CN1125931A - Elevator group control system - Google Patents
Elevator group control system Download PDFInfo
- Publication number
- CN1125931A CN1125931A CN 94192592 CN94192592A CN1125931A CN 1125931 A CN1125931 A CN 1125931A CN 94192592 CN94192592 CN 94192592 CN 94192592 A CN94192592 A CN 94192592A CN 1125931 A CN1125931 A CN 1125931A
- Authority
- CN
- China
- Prior art keywords
- value
- parameter
- searching
- new
- memory storage
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Abstract
A system that controls elevator cars by a group control algorithm with a plurality of parameters is composed of a searching device for the best parameters aggregate which is accessed from the parameter value for the group control algorithm. A plurality of new aggregates are formed in the operation of the system through crossover or variation. The choice aggregates can be stored in memory via an additional record, and incomplete aggregates can be deleted from the memory so that the choice aggregates can be accumulated in the memory. Since the best aggregate is selected from the accumulated aggregates, the system can effectively find the best aggregate.
Description
Background of invention
1. invention field
The present invention relates to elevator group control system, relate in particular to a kind of in order to effectively to search the device of control parameter value best of breed.
2. relevant description of the Prior Art
Elevator group control system be a kind of according to various traffic conditions in the building in order to the system of the many lift cars of effective manipulation.The group control device of system is according to the group control algorithm control operation, as the configuration of lift car.Wherein, group control algorithm is in order to carry out and various functions and the activity of operating relevant elevator cage operation (such as the configuration control of lift car etc.).
Group control algorithm comprises various types of controlled variable.In order to carry out the actv. operation, need be the suitable numerical value of these parameter substitutions according to various traffic conditions in the building and analogue.
As one of basic team control function to should be in the distribution of floor call (from the calling in elevator room) control, when a floor when writing down recently, calculate the estimated valve Em in each lift car (elevator guest room) according to the distribution assessment function of the following stated, so that estimate various condition of sevice according to new floor call and the floor call that write down, such as wait time and reckoning etc.Then, select the lift car of estimated valve Em minimum as institute's assigned elevator car.The elevator door signal lamp or the allied equipment that are arranged on floor were connected before car arrives, and indicated institute's assigned elevator car thus, the passenger that guiding is being waited for (this operation is called the prediction operation).
For example following equation [1] has exemplified in order to determine a kind of function of above-mentioned distribution estimated valve Em, and wherein, i is the floor call numbering; M is the lift car numbering.
Wherein,
Em represents the distribution estimated valve when a new floor call is distributed to lift car m; The wait time of the floor call i that W (i) expression is calculated when a new floor call is distributed to lift car m; M (i) represents when a new floor call is distributed to lift car m the probability (0≤M (i)≤1) of floor call i for full objective state; The prediction error rate (0≤Y (i)≤1) (prediction error refers to the situation when the lift car of another lift car rather than forecast at first arrives this floor) of Y (i) expression floor call i when a new floor call is distributed to lift car m; Pm represents penalize (Penalty) when a new floor call is distributed to lift car m; Bm represents the prize (bonus) when a new floor call is distributed to lift car m; Ca represents full objective state estimation coefficient; Cb represents the prediction error estimation coefficient.
Full objective state estimation coefficient Ca is relative wait time estimated valve W (i)
2Coefficient for full objective state estimation value M (i) addend.If give coefficient Ca with bigger numerical value, system can be during fully laden to operation weighting, rather than to the operation weighting of wait time by this floor.
Prediction error estimation coefficient Cb is relative wait time estimated valve W (i)
2A coefficient for prediction error estimated valve Y (i) weighting.If give coefficient Cb with bigger numerical value, system can distribute car, for preventing the prediction error weighting rather than being the wait time weighting.
Utilize the priority allocation function of penalizing Pm in the equation [1] to comprise for example ride time priority allocation function [2], and economize on electricity priority allocation function [3], as described below.
[2] ride time priority allocation function is a kind of like this function, and the lift car that it attempts to guarantee to accept many floor call is accepted another call distribution from other any floor seldom again.For example, will multiply by calls (Nm) by the value (Pa) that the ride time relative importance value calculates and offer penalty function Pm.
[3] economize on electricity priority allocation function is a kind of like this function, and it guarantees to suspend no lift car and accepts to distribute new calling seldom again.For example, will offer the penalty function Pm that suspends no lift car by the value that economize on electricity relative importance value Pb represents, and the penalty function Pm of other each lift car will be decided to be zero.
Adopt the priority allocation function of equation [1] prize-winning function Bm to comprise for example priority allocation function [4], underloading car priority allocation function [5] and the specific car priority allocation function [6] of contiguous car.
[4] being close to car priority allocation function is a kind of like this function, and it guarantees to be coordinated in contiguous lift car (contiguous lift car) the rapid branch.For example, the value Ba that is expressed as contiguous car relative importance value is set to the prize function Bm of contiguous lift car, and will zero be set to the prize function Bm of other lift car.
[5] underload car priority allocation function is a kind of like this function, and it makes idle lift car or underloaded lift car (underloading lift car) can obtain rapidly distributing.For example, the value Bb of expression underloading car relative importance value is set to the prize function Bm of underloading lift car, and will zero be set to the prize function Bm of other each lift car.
[6] specific car priority allocation function is a kind of like this function, and it makes a particular elevator cab can obtain rapidly distributing.For example, the value Bc that is expressed as specific car relative importance value is set to for example running to the prize function Bm of lift cars such as bottom, roof, observation layer, and will zero be set to the prize function Bm of other each lift car.
As mentioned above, Ca, Cb, Pa, Pb, Ba, Bb and Bc are and distribute the relevant team control parameter of estimation function [1].
Even if after system has utilized distribution estimation function [1] execution distribution function, over a long time the wait relevant with the inexpectancy calling may take place still.Therefore, this group control system also has a kind of additional allocation function [7] and distributes modify feature [8].
[7] be a kind of additional allocation function at the additional allocation function of waiting for over a long time, wherein, elevator just can be carried out various services before the elevator that requires existing distribution is got involved.
[8] be a kind of like this function at the distribution modify feature of waiting for over a long time, the call distribution that it will be waited for over a long time (and forecast) sends the lift car of getting involved to.Wait for over a long time in order to detect one, set and judge reference value DL.
[9] each lift car of group control system all has current voluntarily function during fully laden.If the heavy burden of lift car surpasses reference value DB, it can be without a break by carrying out the floor of distribution.Get involved operation and carried out by the distribution modify feature at a kind of like this calling, promptly lift car passes through during fully laden voluntarily.
[10] be applicable to the distribution modify feature of calling current voluntarily during fully laden, send the distribution and the forecast of floor call to lift car that another is being got involved.Forecast modification to new allocated elevators car is referred to as the forecast modification.As mentioned above, DL and DB also are the team control parameters.
Except the operation relevant, also utilize various types of controlled variable to carry out other operation with floor call.For example, with the condition of various types of controlled variable with elect following operating mode and this selection of cancellation.
[11] selection and cancellation operation in peak time
When the time of origin in peak time mistake, and when the lift car institute ticket call number of times that at first leaves pinao nobile is equal to or greater than reference value DIUPC, select the operation in peak time; On the other hand, then cancel the operation in peak time when the concluding time in peak time is out-of-date.
[12] top (up-peak) operation is gone up in selection and cancellation
During DUPT, when lift car leaves and the main floor passengers number selects to go up the top operation when surpassing the first reference value DUP1; On the other hand, leave and the main floor passengers number surpasses second cancellation when judging reference value DUP2 and goes up the top operation when no lift car.
[13] select and the cancellation (down-peak) operation that goes to the bottom
During the DDPT, when lift car just descending, when ridership wherein surpasses the first reference value DDR1, select to go to the bottom operation; On the other hand, when no lift car just descending, when ridership wherein is equal to or greater than the second reference value DDR2, the cancellation operation of going to the bottom.
Every kind of operation mode comprises following control, also comprises a controlled variable.
[14] operation in peak time
In service in peak time, its numbering is in line at pinao nobile with the lift car of sequence car numbering DIUPN appointment.
Adjust in the operation leaving,, and, still reference value DIUPT is provided with and equals opening the time departure of time because car door is still being opened and postponed to leave even the lift car that at first leaves pinao nobile is sent calling.
Its numbering is appointed as the lift car of waiting for car numbering DIUPW of the opening the door wait of must opening the door, and other lift car then closes a wait.
[15] go up the top operation
In service on last top, the lift car that its numbering is appointed as sequence car numbering DUPN is in line at pinao nobile.
[16] operation of going to the bottom
Going to the bottom in servicely, when calculating, the floor call of the forecast wait time relative orientation pinao nobile direction that calculates obviously has more an amount that is equivalent to relative importance value DDPE.
[17] disperse to wait for operation
Disperse to wait for that operation is a kind of operation of attempting to shorten wait time.When having free elevator car, in advance lift car is evacuated on the floor that occurs next call easily.When selecting this kind operation, do not select the operation in peak time.
When existing quantity to be equal to or greater than the free elevator car of conventional quantity D OHN, and when this kind situation continues at least one standard time DOHT, carry out this dispersions and wait for and moving.
In service in the dispersion wait, lift car can be used as controlled variable at a floor of waiting for or a plurality of floor (wait floor), the lift car number of obeying wait and other factors.
Moreover following additional controlled variable also is used for controlling the quantity of operation car.
[18] power-saving running
Power-saving running is attempted to save electric power by reduce the operation number of elevator automatically according to condition of sevice.When the last five minutes average latency is equal to or less than the first service reference value DESW1, system reduces one with the car number of current operation, when the average latency was equal to or greater than the second service reference value DESW2, system increased by one with the car number of current operation.
As mentioned above, group control algorithm comprises many parameters.These parameters are in order to satisfy various control purposes, such as shortening wait time, improve forecast precision, improve passenger's level of comfort, saving electric power or the like.Yet because the purpose of various parameters has nothing in common with each other, such team control also will be subjected to the influence after the combinations of values in each parameter of substitution on the whole.
In other words, realize the operation of team control efficiently, need find best parameter combinations rapidly according to the various transportation conditions in the building that changes at any time and passenger's various expectations.
Notice that the parameter value combination (series) of the following stated is called " set of parameter values " or abbreviates " set " as.
In traditional optimal set lookup method, the method of calculating that a kind of precision is arranged, wherein verification (for example see 4-51, No. 475 Japan speciallys permit communique or 57-57, No. 168 day disclosure special permission communiques) is all carried out in each possible combination (being each set) of parameter value.
When only adopting a few parameter, there is any problem hardly.
Yet if different classes of parameter has increased, the number of combinations of various parameter values to be verified increases greatly, so select optimal set just to become quite difficult by all possible combination of verification.It below is special description to this kind method.
In the method for calculating of precision, the kind number of supposing parameter is M, and parameter probable value quantity is L.For example, if M=3, L=6, then total simulation number of times is 216 (=L
M).Therefore, when the kind number of parameter or probable value number are not peanut, just need considerable time to determine the best of breed of parameter value, lift car has been carried out demonstration run also is so even if carried out simulation or the group control device by reality, is unpractical therefore.
5-24, No. 067 Japan disclosed a kind of system of special permission communique, suggestion reduces the simulation number of times.That is, when for example adopting two parameters, the optimum value of first parameter is at first searched by this system, finds the optimum value of second parameter then when the value with first parameter is fixed as optimum value.This kind method is called progressive method.
According to a kind of like this progressive method, if M=3, L=6, total simulation number of times is 18, and (=L * M) compares with above-mentioned accurate method of calculation and to significantly reduce.
Yet progressive method is only just effective when various parameters do not have correlativity mutually, when system's control comprise the stronger parameter of a large amount of correlativitys, when being used for the population of parameters of team control, it is inapplicable.
Summary of the invention
The object of the present invention is to provide a kind of control system that can solve the problem of above-mentioned legacy system, it can search optimal set for a population of parameters with strong correlation effectively, also can solve even if the quantity of parameter sets is very big.
Another object of the present invention is to provide a kind of control system that can realize special-purpose lookup method.In order to search an optimal set according to the current techique that is referred to as " genetic algorithm ".
(1) to achieve these goals, one has the control system of searching device and comprises:
In order to storing the memory storage of a plurality of set,
Generating apparatus is gathered as superclass in order to once to select at least one from memory storage, and is once generated the new set that at least one inherits part superclass character,
Estimation unit, each when carrying out group control algorithm with new set, with the result that carries out as the team control performance valve,
Selecting arrangement, by new set is added to memory storage and from memory storage the deletion damaged set improve a plurality of set that are stored in the memory storage, and
The device of extracting is according to the team control performance valve of storing in the memory storage that the improves set optimal set of extracting.
According to said structure, in the set that generates together with heredity and good being integrated into of selection, the probability that generates good set becomes big, has only the subclass (new set) of only having inherited the superclass fine quality just to be stored in the memory device.That is,, in turn upgrade and improve thus being stored in the interior a plurality of set of memory storage by repeating one-period.According to the team control performance valve of each set, the optimal set of from memory storage, finally extracting.To comprise relevant parameters in each numerical value substitution group control algorithm of optimal set, like this, system can carry out the team control such as lift car distributes.
So, can find optimal set effectively according to the present invention or have set to the similar content of optimal set.Soon can reduce transportation load and simulation number of times, and make system find them rapidly thus.
(2) according to the present invention, generating apparatus comprises:
The exchange of values device generates two new set by the value part that exchanges two set selecting from memory storage,
New value displacement apparatus, some parameter value of replacing a set of selecting by the new numerical value that generates with random fashion from memory storage generates a new set, and
Generation method selecting arrangement makes one's options between exchange of values and new value are replaced according to separately probability.
In this structure, by select at random in the exchange of values device " intersection " (crossover) and " variation " in the new value displacement apparatus (mutation) generate new set.
In brief, cross convergence is in solution, and variation has then brought variation to solution.Therefore, intersection can restrain the content of a set combination of storing in the memory storage, but conversely, the variation of set combination does not appear at its commitment, so system can fail to carry out local solution and lose intrinsic solution (best approach).In the case, variation can make system break away from the scope of local solution.On this kind meaning, intersecting and making a variation is complementary relationship.
On the other hand, variation has a kind of invalid trend of good solution that finds by intersection that makes.On this meaning, intersecting and making a variation is competitive relation.Ratio between aberration rate when crossing-over rate when therefore, requirement will be selected to intersect and selection variation is set a suitable value for.By utilizing intersection by any way simultaneously suitably and making a variation the two, system can effectively utilize both advantages, and improves the probability that generates good new set.
In addition, in some cases, system can also utilize to intersect or make a variation and find optimal set.
(3) according to the present invention, generating apparatus comprises:
The superclass selecting arrangement, in order at least one superclass of selection from memory storage, and the parameter selecting arrangement, in order to select various parameters by exchange numerical value or displacement numerical value.
Adopt this kind structure, elected when choosing friends fork, the superclass selecting arrangement is selected two superclass (set to) from memory storage, and when selecting to make a variation, the superclass selecting arrangement is selected a superclass from memory storage.The parameter selecting arrangement is selected the position (crossover location or variation position) of various parameters, replaces parameter value with intersecting or making a variation on this position.
(4) the superclass selecting arrangement selects reference information to carry out the superclass selection according to superclass, to improve the probability that generates good new set.By utilizing this superclass to select reference information, system can make the probability that generates good new set higher.
(5) if the distance between will gathering is selected reference information as superclass, so also can be with the similarity between the set as selection reference.The method allows system relative importance value to be offered the variation or the convergence (convergence) of new set.
(6), then select superclass according to the good degree of each set if the team control performance valve is selected reference information as superclass.This method has improved good superclass with selecteed probability, the result, system enhancement generate the probability of good new set.
(7) if the identity set number is selected reference information as superclass, each original set becomes selection reference.The method has improved the right probability of set of selecting to have mutually different qualities, guarantees the generation that new set changes thus.
(8) preferably revise the condition that superclass is selected by the superclass selecting arrangement according to the progress of searching.For example, multiple superclass alternative condition can be prepared by system, and according to the progress of searching they is exchanged, or changes the reference value of superclass alternative condition according to the progress of searching.Like this, when formation like this, system can improve the probability that generates good new set.
(9) the parameter selecting arrangement selects reference information to select parameter according to parameter, to improve the probability that generates good new set.Therefore, the probability that generates good new set becomes higher.
(10) if with each parameter value difference as parameter selection reference information, then the similarity between each parameter becomes selection reference.This method offers the ability that new set changes or restrains that generates with relative importance value.
(11) if in conjunction with the condition of service of lift car the degree of correlation is selected reference information as parameter, system can improve the probability of selecting to have with the lift car condition of service parameter of the big degree of correlation, has improved the probability that generates better new set thus.
(12) if the content of bonding properties estimated valve is selected reference information with the degree of correlation as parameter, system can improve the probability of selecting to have with the performance estimation value parameter of the big degree of correlation, has improved the probability that generates better new set thus.
(13) preferably according to the progress of searching, revise the parameter alternative condition by the parameter selecting arrangement.For example, system can prepare various parameter alternative conditions and they are exchanged according to the progress of searching, or according to the reference value of various change of circumstance parameter alternative conditions.System can improve the probability that generates good new set.
(14) preferably according to searching progress, revise the selection probability of every kind of generation method by the probability modifier.The utilization of the method permission system improves search efficiency thus by the search procedure and the local search process by intersecting of variation.
(15) when system so constitutes, wish to revise the selection probability by the probability modifier according to for example successful index.According to this structure, system can determine the progress of searching from the degree of convergence of searching, and like this, system can be provided with the selection probability that is fit to process.
(16) in another aspect of this invention, according to the device of searching of the present invention, it comprises:
Memory storage, in order to storing a plurality of set,
The exchange of values device by switching part parameter value select as two set of superclass from memory storage between, generates two new set that herid its superclass character partially,
New value displacement apparatus is selected partial parameters value as a set of superclass by the new numerical value displacement that generates with random device from memory storage, generates a new set that herids its superclass character partially,
Generation method selecting arrangement in order in conjunction with its probability separately, makes one's options between exchange of values method and new value method of replacing,
Estimation unit when group control algorithm is carried out in the new set of each utilization, is searched execution result as the team control performance valve,
Adding device is stored the good new set of satisfying certain additional conditions in the memory storage in order to only to add,
Delete device satisfies the set of damage of certain deletion condition of memory storage in order to deletion, and
The device of extracting is in order to according to improved and be stored in team control performance valve between a plurality of set in the memory storage optimal set of extracting.
According to this kind structure, select through exchange numerical value (intersection) generation set and replace the method that generates set to generate new set with new value (variation) with random fashion.The new set of only satisfying some additional conditions between the new set that is generated just is stored in the memory storage, and the set that damages is then deleted in memory storage.
When repeating this circulation time in an orderly manner, memory storage is only stored good set, so can be from its optimal set of extracting out.Each numerical value with optimal set places in the relevant parameter of group control algorithm then.
According to system of the present invention, can find a kind of set that optimal set or content are in close proximity to the optimal set content effectively thus.That is, system can reduce operation and simulation number of times, realizes therefrom searching rapidly.
(17) in the preferable feature of system, search device and further comprise the additional conditions modifier.
(18) additional conditions are for example determined according to the team control performance valve of each set of storing in the memory storage, and are arranged to more and more tighter.By this kind operation, system can always only be accumulated in good new set in the memory storage, has reduced unnecessary process, and has improved the probability that generates good new set thus.
(19) delete device is for example deleted according to the team control performance valve.By this kind operation, system only stays good set, has optimized the superclass of a plurality of storages thus on the whole.
(20) delete device is for example deleted according to the distance between the set.By this kind operation, system can avoid the dual state that has similar set of a plurality of set in the relative memory storage, and can guarantee the variation of gathering.
(21) device of searching according to the present invention further comprises apparatus for initializing in its preferable feature.If adopt initialization group's initialization as far as possible closely to run into search criterion, then system can reduce the time of searching.
(22) this apparatus for initializing preferably includes first and second kinds of patterns.In first kind of pattern, pre-prepd a plurality of set are used as the initial sets group; In second kind of pattern, in the end improved a plurality of set are used as the initial sets group in the search procedure.If when system begins to search, just choose suitable pattern according to condition, the then system's convergence that can quicken to search.
(23) device of searching according to the present invention comprises further that in its preferable feature searching end determines device.When system has entered a kind of like this state, when promptly set can be expected to improve effectively by system during searching, this device was promptly determined the end of search procedure.If search procedure is still not enough and any search procedure is all invalid, system can cancel search procedure and finish.
(24) wait to estimate that the quantity of gathering is relevant with the improvement on-cycle execution time, and can be as finishing judgment standard.
(25) wait to add the improved degree of quantitaes memory storage of set, and can be as finishing to come judgment standard.
(26) successful index is to treat additional aggregates number and the ratio of waiting to estimate the set number, because the convergence capabilities of its secondary indication search procedure, so can be as the benchmark that finishes to judge.
(27) distance between each set can be as a kind of benchmark that finishes to judge, it represents the similarity of a plurality of set in the memory storage on the whole.
(28) device of searching according to the present invention preferably further comprises and searches definite device again, and its variation of various prerequisites when searching beginning is determined to search again.This device permission system searches optimal set automatically under new condition.These prerequisites for example comprise lift car specification, magnitude of traffic flow specification, performance reference value and ratio of controlling reference value or the like.
(29), even the team control performance valve can be stored in the memory storage according to system of the present invention.
(30) search device and can be connected to the expected value setting device, to set the expected value that is associated with search procedure.In the occasion that controlled target is freely set, optimal set can be searched according to the target of having set by system.
(31) in the present invention, new set utilizes the specialized simulation device to estimate, searches device except being connected to group control device, is also connected to simulator.This simulator comprised with group control device in the contained identical group control algorithm of group control algorithm.The execution result of estimation unit simulation is set to the team control performance valve.When adopting this simulator, system can estimate new set and need not to interrupt team control.
(32) in the present invention, group control device for example is installed in and searches in the same building of device (and simulator).Requiring to search the place that device (and simulator) and group control device are installed separately, with communication line with group control device with search device and couple together.If many shared one of group control devices are searched device (and simulator), the cost of system will reduce.
(33) in the present invention, be connected to the group control device of searching device as a utility unit and can be used for carrying out simulation.Therefore, can be without simulator, the cost of system can reduce.
(34) and in the present invention, be arranged on the control device place far away that peels off searching device, group control device and search device and be connected with communication line, if many shared one of group control devices are searched device, the cost of system can reduce.
(35) the present invention can expand and be used for such purpose, promptly by connecting simulator and searching device group control algorithm is estimated.
Relation between the present invention and the GA:
Genetic algorithm has been done elaboration (for example, referring to " genetic algorithm present situation and existing problems " literary composition of delivering on " Measurement and Control " the 32nd the 1st phase of volume of publishing in January, 1993) in all kinds of documents and data.Basic genetic algorithm generally includes a series of loop initialization, superclass is selected, intersects, made a variation and generates.
In " at the genetic algorithm solution of the elevator group controlling car assignment problem of fixing calling " literary composition of in the 34th the automatic guidance associating lecture that held 20 to 22 November in 1991, delivering, a kind of system that genetic algorithm is applied to the lift car team control has been described.
This traditional system utilizes genetic algorithm to carry out optimal allocation to lift car at " calling out displacement ".Therefore, the present invention is identical with traditional system at the system aspects of relevant genetic algorithm, but the purpose that they will be realized is different, and they also have very big difference at aspects such as basic structures.
In brief, the present invention has not only adopted genetic algorithm, and the new technology of searching with optimal parameter value set is provided.These characteristics of the present invention will be embodied in the ad hoc structure of the present invention that comes from the special properties set of parameter values.
Description of drawings
Fig. 1 is a block scheme, the overall structure of first embodiment of its expression system according to the invention;
Fig. 2 is the scheme drawing that apparatus structure is searched in expression, comprising a microcomputer;
Fig. 3 is the scheme drawing of bilge construction within the RAM10C in the presentation graphs 2;
Fig. 4 is the scheme drawing of bilge construction within the ROM10B in the presentation graphs 2;
Fig. 5 is the scheme drawing of expression lift car specification data (ELS) structure;
Fig. 6 is the scheme drawing of expression magnitude of traffic flow specification data (TRS) structure;
Fig. 7 is the scheme drawing of expression team control performance data (PRF) structure;
Fig. 8 is the scheme drawing of expression set of parameter values (EPS) structure;
Fig. 9 is a diagram of circuit, and it represents the content of the control program of first embodiment;
Figure 10 is a diagram of circuit, and it represents the look-up command program of first embodiment;
Figure 11 is a diagram of circuit, and it represents the main program of searching of first embodiment;
Figure 12 is a diagram of circuit, and it represents the initial determining program of searching of first embodiment;
Figure 13 is a diagram of circuit, and it represents the initialize routine of first embodiment;
Figure 14 is a diagram of circuit, and it represents the new set generator program of first embodiment;
Figure 15 is a diagram of circuit, and it represents the estimation routine of first embodiment;
Figure 16 is a diagram of circuit, and it represents the addition program of first embodiment;
Figure 17 is a diagram of circuit, and it represents the delete program of first embodiment;
Figure 18 is a diagram of circuit, and it represents the additional reference value update routine of first embodiment;
Figure 19 is a diagram of circuit, and it represents that first embodiment's searches the end determining program;
Figure 20 is a diagram of circuit, the optimal set that it represents first embodiment program of extracting;
Figure 21 is a block scheme, and it represents second embodiment of the present invention;
Figure 22 is the scheme drawing of RAM among second embodiment of expression;
Figure 23 is a diagram of circuit, the additional reference value update routine of second embodiment of its expression;
Figure 24 is a diagram of circuit, and it represents the initial determining program of searching of second embodiment;
Figure 25 is a diagram of circuit, the initialize routine of second embodiment of its expression;
Figure 26 is a diagram of circuit, and it represents the delete program of the 3rd embodiment;
Figure 27 is a diagram of circuit, and it represents that the 5th embodiment's searches the end determining program;
Figure 28 is a diagram of circuit, and it represents that the 6th embodiment's searches the end determining program;
Figure 29 is a diagram of circuit, the optimum value that it represents the 7th the embodiment program of extracting;
Figure 30 is a block scheme, and it represents the 8th embodiment;
Figure 31 is a diagram of circuit, and it represents the main program of searching of the 8th embodiment;
Figure 32 is a block scheme, and it represents the 9th embodiment;
Figure 33 is a diagram of circuit, and it represents the main program of searching of the 9th embodiment;
Figure 34 is a diagram of circuit, and it represents occurrence rate (emergencerate) update routine of the 9th embodiment;
Figure 35 is a diagram of circuit, and it represents the occurrence rate update routine of the tenth embodiment;
Figure 36 is a block scheme, and it represents the 11 embodiment;
Figure 37 is a diagram of circuit, and it represents the operation main program of the 11 embodiment;
Figure 38 is a diagram of circuit, and it represents the part of the new set generator program of the 11 embodiment;
Figure 39 is a diagram of circuit, and it represents the alternative condition update routine of the 11 embodiment;
Figure 40 is a diagram of circuit, and it represents the alternative condition update routine of the 12 embodiment;
Figure 41 is a diagram of circuit, and it represents the part of the new set generator program of the 13 embodiment;
Figure 42 is a diagram of circuit, and it represents the alternative condition update routine of the 13 embodiment;
Figure 43 is a diagram of circuit, and it represents the alternative condition update routine of the 14 embodiment;
Figure 44 is a diagram of circuit, and it represents the part of the new set generator program of the 15 embodiment;
Figure 45 is the scheme drawing of occurrence rate of each parameter of the 15 embodiment of expression;
Figure 46 is a diagram of circuit, and it represents the part of the new set generator program of the 16 embodiment;
Figure 47 is the scheme drawing of the 17 embodiment of expression;
Figure 48 is the scheme drawing of the 18 embodiment of expression;
Figure 49 is the scheme drawing of expression nineteen embodiment;
Figure 50 is the scheme drawing of expression according to a kind of optimal set lookup method of the present invention.
The detailed description of preferred embodiment [groundwork]
As mentioned above, group control algorithm comprises various types of parameters.In order effectively some lift cars to be carried out team control, need search the parameter value of best of breed according to traffic conditions.Device at this purpose is exactly that optimal set is searched device, and Figure 50 shows the groundwork of searching device according to of the present invention.As mentioned above, the combination of parameter value (sequence) is called as " set of parameter values " or abbreviates " set " as.
As shown in figure 50, by generating new set repeatedly and selecting good set to search optimal set.It is described below specially:
At first, memory device A2 be initialised (A1).For example, pre-prepd a plurality of initial sets are stored in (A2) in the memory device.
Then, generate new set (A4).New set generates by selecting exchange of values (intersection) or new value displacement (variation) at random.Elected when choosing friends the fork method, take out two set (superclass to) from memory device A2 and between two set, exchange a part of parameter value to generate two new set.When selecting variation method, from memory device A2, take out a set (superclass), replace a part of parameter value in this set in order to the new numerical value of random fashion generation, to generate a new set.
Note, select generating mode, superclass and each parameter that exchanges numerical value are therein all carried out basically at random, and each alternative condition can be determined arbitrarily, and each selects element to select probability weight according to it.
Next the new set A 5 that generates is estimated.That is, packed into the group control algorithm of each new set of virtual or actual execution is to obtain execution result.This execution result that obtains is as " team control performance valve ", the performance of the new set of expression, and the new set that will have a good team control performance valve deposits memory device A2 (A8) in.The set that damages is no longer stored and is discarded (A9) or deleted after storage (A10).This good set back-and-forth method (A7) is always only accumulated good set in memory device A2.
Constantly repeating such improving under the on-cycle situation, a plurality of set A 3 that are accumulated in the memory device A2 can make a distinction and improve gradually.Owing in a plurality of set A 3 of being stored, can finally extract best set, so this optimal set group control algorithm of feeding is used for team control as optimal set (A11).
According to described a kind of like this method of searching optimal set, can effectively generate the subclass of having inherited its superclass superperformance.That is, system can improve the probability that is generated good subclass by good superclass, and they can be found rapidly.
Can adopt any method of intersecting and making a variation, select at random but be preferably between the two, like this, the set of a plurality of accumulations can be satisfied convergence suitably and change.
[first embodiment]
Structrual description
Fig. 1 to Figure 20 represents first embodiment according to elevator group control system of the present invention.Fig. 1 represents total system, and it comprises known group control device 1, known simulator 2 and searches device 10.
In order to the writing function of calling out car is when the bid, this calling of record in memory device.The operation controllable function is control the advancing of lift car, stop the decision with sense of motion, so that the calling (floor call of car call and distribution) that the lift car response must must respond.The gate control function is that opening and closing are arranged on the door on elevator cab and each floor.Presentation control function is to notify the passenger institute assigned elevator car of waiting for by lighting building gate signal lamp, and arrives by this building gate light notice passenger elevator car that glimmers.
Group control device 1 transmits a search criterion signal 1a, the condition when optimal set is found in expression to searching device 10.Search criterion signal 1a comprises " the look-up command data " that optimal set is searched in " lift car specification data ", " magnitude of traffic flow specification data " that need simulate the magnitude of traffic flow in the building on computers and the instruction of simulant elevator car group control system on computers.The lift car specification data comprises the data of the type of for example representing its numbering, speed, passenger's limit, the floor that will stop and elevator cab door, and the additional operation of serving or not serving, such as data such as power-saving running and operations in peak time.Magnitude of traffic flow specification data, for example when the magnitude of traffic flow in the building is secondary indication, it comprises the data that are used to make up various eigenwerts, such as per hour Total passenger and floor to the floor rate of traffic flow, and in order to make up the data of various eigenwerts, be multiplied by the ridership of elevator etc. in the unit time such as every relatively floor and each direction, on the other hand, when the magnitude of traffic flow in the building was direct representation, it comprised passenger data at all passengers (such as epoch, floor, destination floor and similar data occur).
In searching device 10, memory device 11 is stored the multiple parameter values set, and the storage team control performance data relevant with corresponding set.Output signal 11a from memory device 11 comprises set of parameter values and team control performance data.
Additional reference value of additional reference value memory device 14 storage is additionally to be deposited with memory device 11 or to abandon in order to determine the new set of so estimating.Additional reference value memory device 14 transmits an additional reference value signal 14a at its mouth.
A kind of addition records that 15 pairs of team control performance datas that comprised by estimated result signal 13b of adder unit are determined generate performance estimated valves, and should be worth and the additional reference value comparison.When this performance estimation value was better than additional reference value, the signal 15a that adder unit 15 generates comprised new set and team control performance valve thereof, and sends it to memory device 11.Cao Zuo result like this, good new set by addition record in memory device 11.
After having satisfied some condition of relevant set record situation, delete cells 16 is made the deletion decision according to team control performance data calculated performance estimated valve to each set.Delete cells 16 is selected the relatively poor set of its performance estimation value, and sends the delete instruction signal 16a of expression set numbering inferior.The result of this operation, the record of named aggregate is promptly deleted in memory device 11.
Whether 17 judgements of end identifying unit are searched and are finished and will search end signal 17a when determining to search end to be sent to maker 12.The result of this operation searches device and stops to generate new set.
Additional reference value modifier 18 utilizes modification signal 18a to revise the additional reference value that is stored in the additional reference value memory device 14.The modification degree depends on the team control performance data of each set in the memory device 11.
Fig. 2 represents the hardware configuration of searching device 10 shown in Figure 1.Among Fig. 2, search device 10 and comprise microprocessor 10A, read-only memory (ROM) (ROM) 10B, random-access memory (ram) 10C and input interface circuit 10D and output interface circuit 10E.In the case, ROM10B has stored the search program of expression microprocessor 10A control step and curing data.RAM10C has stored the arithmetic operation results (service contamination) of microprocessor 10A, search criterion signal 1a that sends into by the outside and the content (input data) of team control performance valve signal 2a, and the content (output data) of analog signal 13a that will outwards send and optimal set signal 20a.
Fig. 3 represents the memory contents of RAM10C shown in Figure 2; Fig. 4 represents the curing data part in the ROM10B memory contents.
Among Fig. 3, ELS is the data of expression car specification; TRS is the data of expression magnitude of traffic flow specification; SCM is the data of expression look-up command.These input data all are included in the search criterion signal 1a shown in Figure 1.
Fig. 5 represents the ad hoc structure of lift car specification data ELS.In example shown in Figure 5, regulation car number is 4; Speed is 120 meters/minute; Passenger capacity is 20 people; Lift car is parked in the 10th buildings, and the 1st building is ground floor, and the 10th buildings are The Highest Tower; The width of door is 1,000 millimeter.Move about priority allocation, [2] running time priority allocation function, [3] economize on electricity priority allocation function, [4] contiguous car priority allocation function, and [5] underloading car priority allocation function all is set to " effectively ", and specific car priority allocation function [6] is changed to engineering noise.About various operations, operation in [11,14] peak time, [12,15] go up top (up-peak) operation, [13,16] going to the bottom, (operation of down-peak) and [17] disperse to wait for that operation all is changed to " effectively ", and [18] power-saving running adds operation as another kind and also is changed to " effectively ".Although expression among Fig. 5, [8] revise operation in response to the distribution that waits as long for calling, and [9] always place " effectively " basically when automatically current function and [10] distributing altering function of at full load.
Fig. 6 represents a kind of magnitude of traffic flow specification data TRS of ad hoc structure.Example shown in Fig. 6 is at business hours scopes (14:00 to 15:00).For example, according to the magnitude of traffic flow result who utilizes group control device 1 actual measurement in advance, ridership is 500 people per hour; Volume of traffic between bottom and any other floor (the 2nd layer to the 10th layer) is 80% (bottom traffic rate) with the ratio of whole volume of traffic; The uplink traffic amount is 50% with the ratio (uplink traffic rate) of whole volume of traffic; The down traffic amount is 50% with the ratio (down traffic rate) of whole volume of traffic.
The PRF that appears at Fig. 3 upper left side refers to expression team control properties data, and the good property of its expression set out of the ordinary is equal to team control performance valve signal 2a shown in Figure 1.
Fig. 7 represents the ad hoc structure of team control performance data PRF.In this example, team control performance data PRF comprises average latency AWT, wait as long for rate RLW, the most common wait time MWT, prediction error rate PRE, forecast volatility RPC, the at full load rate RBP that passes through, average ride time ABT, the most common ride time MBT, power consumption PWC, contiguous car responsiveness RNR (bottom is called out the ratio of handling by near the car of bottom), underloading car responsiveness RLR (will distribute to its ratio by the floor call of underloading car record), and specific car responsiveness RSR (floor call that records specific car will be distributed to its ratio).
Goodbye Fig. 3, the P that is positioned at the upper right side is for being recorded in memory device 11 expression set (perhaps being called good set) quantity data; EPS (1) is that expression is from first to Pmax data of gathering to EPS (Pmax); PRE (1) is to represent the data of corresponding EPS (1) to the team control performance valve of EPS (Pmax) to PRE (Pmax).Set number P, collective data EPS (1) to EPS (Pmax) and team control performance data PRE (1) to PRE (Pmax) corresponding to signal 1a shown in Figure 1.As described below, Pmax is the peaked numeral of the expression set number that can write down.
Fig. 8 represents a kind of ad hoc structure as the set of parameter values of an example.Among Fig. 8, this set comprises 25 types controlled variable.That is, the collective data of each shown in Fig. 3 EPS (1) to EPS (Pmax) all as shown in Figure 8.Team control performance data PRE (1) shown in Figure 3 almost has the structure (its ad hoc structure is also shown in Fig. 7) identical with team control performance data PRE to PRE (Pmax).
Pn shown between top, Fig. 3 right side and middle part is the data of the newly-generated set number of expression; ((Nmax) is expression set number from the data of the 1st to Nmax new set to NPS (1) to NPS.New set number Pn and new set NPS (1) to NPS (Nmax) corresponding to signal 12a shown in Figure 1.As described below, Nmax is the peaked numeral of the expression new set number that can be generated.
The SIM that is arranged in Fig. 3 upper left side then is equivalent to the output data of Fig. 1 simulated conditions signal 13a, and it is by being used to estimate that the collective data of NPSX, lift car specification data ELSX and magnitude of traffic flow specification data TRSX forms.Be used to estimate that the collective data of NPSX is the data of the new set of expression content, its team control performance is passed through simulated estimation, and constitutes EPS shown in Figure 8.When simulation is finished and respectively when the ELS in the pie graph 5 and the TRS among Fig. 6, lift car specification data ELSX and magnitude of traffic flow specification data TRSX are for representing the data of lift car specification and magnitude of traffic flow specification respectively.
It is data corresponding to Fig. 1 estimated result signal 13b that Fig. 3 upper left side is arranged in RES below the SIM, and it comprises estimates times N E, the set NPSY that is used to estimate and team control performance data PRFY.Estimate that times N E is the data that cumulative frequency is estimated in expression.The set NPSY that is used to estimate is the data that are illustrated in by the new set after the simulated estimation team control performance, and it constitutes EPS in Fig. 8.Team control performance data PRFY is the data of the team control performance valve that provides by simulation of expression, and it constitutes PRF in Fig. 7.
BX is whether the data of expression additional reference value want addition record in order to definite estimated new set, and it is corresponding to the additional reference value signal 14a among Fig. 1.
RAP is the data corresponding to addition record signal 15a among Fig. 1, and it comprises addition record times N R, estimates set NPSZ and team control performance data PRFZ.Addition record times N R is the data that the addition record number of times is determined in expression.Estimate that set NPSZ is the data that expression prepares to be recorded in the good new set in the memory device 11, it constitutes EPS in Fig. 8.Team control performance data PRFZ is the expression team control properties data of estimating after set NPSZ carries out the team control simulation, and it constitutes PRF in Fig. 7.
The RP that shows in Fig. 3 left side central portion is the data of expression set number, and this set is deleted its record with respect to the P of the set EPS (1) that has write down to EPS (P) as damaging set, and it is corresponding to the delete instruction signal 16a among Fig. 1.
FLAG is the data (searching permission flag) that continue to search optimal set or finish to search in order to instruction, and it is corresponding to searching end signal 17a among Fig. 1.
CBX is the data in order to up-to-date rewriting additional reference value BX, and it is corresponding to the modification signal 18a among Fig. 1.
STR is the data of restarting to search optimal set in order to instruction, and it is corresponding to the signal of the look-up command again 19a among Fig. 1.
BPD is the output data that is equivalent to optimal set signal 20a among Fig. 1, and it comprises optimal set BPS, lift car specification data ELSY, magnitude of traffic flow specification data TRSY and searches status data SS.Optimal set BPS is the set that has the top performance value in the set of having write down, and it constitutes EPS in Fig. 8.After utilizing optimal set BPS to carry out the team control simulation, lift car specification data ELSY and magnitude of traffic flow specification data TRSY are the data of representing lift car specification and magnitude of traffic flow specification respectively.ELS in their difference pie graphs 5 and the TRS among Fig. 6.Searching status data SS is when having selected the best set merging in the present embodiment it to be arranged to an expression estimation times N E value, the data that state is searched in expression.
GPSO shown in below the BPD of Fig. 3 left side be corresponding to Fig. 1 in the data of initializing signal 21a, it comprise initial sets count PK, a plurality of initial sets IPS (1) to IPS (PK) and a plurality of team control performance data PRI (1) to PRI (PK).It is the data that the set number in when beginning is searched in expression that initial sets is counted PK, and it is arranged to and the identical numerical value of numerical value Pe in order to the end of definite delete procedure at first.A plurality of initial sets IPS (1) are used as to IPS (PK) and search used set group of when beginning, and the EPS in the pie graph 8.Team control performance data PRI (1) PRI (PK) is when utilizing initial sets IPS (1) to carry out team control when simulation, expression team control properties data, the PRF in its pie graph 7 to IPS (PK).
VPD shown in Fig. 3 right side (1) is the performance estimation value that is used to delete judgement to VPD (Pmax); VPE (1) to VPE (Pmax) for when revising additional reference value BX in order to the performance estimation value of setting additional reference value; VPS (1) to VPS (Pmax) be in order to determine the performance estimation value of optimal set when selecting optimal set; The VPN that is positioned at below the GPSO of Fig. 3 left side is as the performance estimation value of adding judgement when determining to estimate whether set NPSY wants addition record.In the present embodiment, the average latency AWT that takes out from the team control performance data places and need not in each performance valve to revise.
NP is the data of the set number of expression set, and its team control performance is newly being gathered NPS (1) to estimating between the NPS (Nmax).
WVPE is data of representing the poorest performance estimation value; BVPE is the data of expression top performance estimated valve; RC searches counting machine, the set number that will use in order to calculating when searching worst-case value WVPE and best values BVPE; BP is the data that expression has the set of records ends number of best values BVPE.
PS1 is the first superclass number of representing in order to the superclass number that generates new set; Equally, PS2 is the second superclass number; PX is the data that parameter (position) number of intersection process or mutation process is carried out in expression; CR is the data that the expression intersection is selected probability (occurrence rate); MR is the data that probability (occurrence rate) is selected in the expression variation.
Among Fig. 4, but Pmax is the peaked data of expression set of records ends number; Nmax is the peaked data that expression can newly-generated set number; In the present embodiment, Pmax is changed to 50, and Nmax is changed to 20.
NEa searches the end decision content, and whether it is in order to judge once that optimal set is searched and restrain together with searching times N E.In the present embodiment, NEa places 1,000 time.
Ps is deletion beginning determined value, and it judges once in order to count P together with set of records ends whether the set that has damaged will delete; Pe is that deletion finishes decision content, and it is in order to judge once whether the delete procedure to damaging set finishes.In this first embodiment, Ps places 50, and Pe places 30.
AVPE is the data of expression compensation value, and this compensation value is added to the worst value WVPE of performance estimation value when additional reference value CBX is provided with.In other words, promptly revise additional reference value with the compensation value AVPE that is added to the worst value WVPE.As compensation value, be arranged to usually zero second or zero second above value; In the present embodiment, AVPE was changed to 1 second.GPS1 to GPS4 is corresponding to conventional operation (business hours), operation in peak time, goes up the initialization set group that the top is moved and gone to the bottom and move.Each initialization set group GPS (1) constitutes initialization to GPS (4) and gathers group GPS0 in Fig. 3.
The operation explanation
Referring to Fig. 9 to 20, the operation of first embodiment will be described below.Fig. 9 represents the main portion of a control program in the group control device 1.This control program comprises group control algorithm, and group control device 1 is operated according to control program.Group control algorithm itself is known.
Among Fig. 9, step 221 is carried out the floor call logging program.The floor call that takes place during especially, with passenger's action button is recorded in the memory device.When any lift car is handled this calling, i.e. this call record of cancellation.
Step 222 is carried out allocator, especially, utilizes the distribution assessment function of equation [1] to calculate the distribution estimated valve of each elevator cab relatively.Its estimated valve is that minimum lift car is assigned with and tackles this calling.Except the basic Distribution Calculation of assessment function, this step also comprises the operation based on following various functions, the time priority degree distribution function [2] of promptly travelling, economize on electricity distribution function [3], contiguous car relative importance value distribution function [4], underloading car priority allocation function [5] and specific car priority allocation function [6].
In step 223, carry out the distributing altering program.Especially, system detects the deterioration of the floor call service that distributes as mentioned above, and carries out to distribute and reset it.This step comprises according to the processing [8] of distributing altering operation to waiting as long for calling, and the distribution retouching operation [10] of working as the current floor call of car at full load.
Step 224 is carried out the operation procedure in peak time.Especially, according to selection and the cancellation condition [11] to the operation in peak time, select or the cancellation operational mode, as selecting the operation in peak time, system is according to operation control operation in peak time [14].
Step 225 carry out to go up the top (operation procedure of up-peak).Especially, according to selection and the cancellation condition [12] to the operation of last top, select or cancel this operational mode, if select to go up the top operation, system is then according to last top operation control operation [15].
(the operation procedure of down-peak) is carried out and to be gone to the bottom to step 226.Especially, according to selection and the cancellation condition [13] to the operation of going to the bottom, select or cancel this operational mode, if select to go to the bottom operation, system then moves [16] according to the operation control of going to the bottom.
Step 227 is carried out and is disperseed to wait for operation procedure.Especially when moving, off-peak time selects to disperse to wait for operation the top operation or the operation of going to the bottom in the selection.When selecting to disperse to wait for operation, system is according to disperseing to wait for operation control operation [17].
Step 228 is carried out the power-saving running program.Especially, in order to consider the economize on electricity under the operation service situation,, drop into the number of the lift car of service by increase and decrease and control operation [18] according to power-saving running.
Notice that whole team control program (comprising control program shown in Figure 9 and look-up command program shown in Figure 10) is (for example per 100 milliseconds once) who periodically carries out.
Figure 10 represents the look-up command program that group control device 1 is equipped with.This program is a kind of by searching the command lookup program that device 10 is carried out.
Among Figure 10, execution in step 232 when obtaining optimal set with respect to any magnitude of traffic flow is taken out optimal set signal 20a from searching device 10, being stored in the memory device of group control device 1 corresponding to the optimal set BPS of its magnitude of traffic flow TRS.Simultaneously, also store to be included in and search status data SS in the optimal set signal 20a.
Step 232 and 233 is judged previous searching by searching status data SS.Equal NEa (=1 if search status data SS, 000), because search procedure is finished, step 234 determines to find the magnitude of traffic flow of optimal set, and up-to-date generation and transmission search criterion signal 1a, it comprises specification data TRS, the lift car specification data ELS of the magnitude of traffic flow and the look-up command data SCM that is changed to " 1 ".
Conversely, if the status data SS that searches that step 233 is determined is more than 1 or 1, this expression search procedure begins, then in step 235, group control device 1 is rewritten as " 0 " with the value of the look-up command data SCM in the search criterion signal 1a, and at the new search criterion signal 1a of its mouth output.In order to search the optimal set of every kind of magnitude of traffic flow, group control device 1 is sequentially selected four types the magnitude of traffic flow, and they are corresponding to conventional operation (business hours), operation in peak time, the upward top operation and the operation of going to the bottom.
In step 234, although magnitude of traffic flow specification data TRS survives according to the measured result of group control device 1, but group control device 1 also can be connected to an existing traffic measurement mechanism, the transportation condition data that latter accumulation has been collected (for example advancing/go out ridership, calls of car or the like) are also delivered to group control device 1 with data, according to these data, group control device 1 can generate magnitude of traffic flow specification data TRS.
Even when searching device 10 adopt a kind of pattern that optimal set also fixedly is provided in search procedure, how much group control device 1 is failure-free if can having according to searching of storing in step 231 optimal set that status data SS judgement obtained so far.For example, represent that one is searched the incipient stage if search status data SS, then group control device 1 is owing to used by actual being set to, thus not only can adopt by searching the set that device 10 provides, and can adopt set with gratifying result of use.This method can be prevented the team control performance degradation of locking system.Represent to find half or terminal stage if search status data SS, judge that then from the optimal set of searching device 10 be unusual failure-free, like this, system performance can be utilized this team control operation set and improve before search procedure finishes fully.
Figure 11 represents to be stored in the search program of searching in the device 10 (main program).This program is stored in the ROM10B.
Among Figure 11, step 25 is carried out one and is searched determining program again, and it has function of searching unit 19 again shown in Figure 1.Step 26 is carried out and is searched the beginning determining program, and device 10 judges whether it searches the time for resetting optimal set.Referring now to Figure 12,, it represents a kind of determination methods of searching again.
Among Figure 12,, search the search criterion signal 1a that device 10 receives from group control device 1 in step 261, and at RAM 10C stored lift car specification data ELS, magnitude of traffic flow specification data TRS and look-up command data SCM.Then,, search device 10 and detect the variations of look-up command data SCM, as detect variation,, search device 10 and will search beginning flag STR and be changed to " 1 " in step 265 from " 0 " to " 1 " at next procedure 262.Otherwise, judge that when searching device 10 look-up command data SCM are not still when " 0 " changes to " 1 ", device 10 judges in detection step 263 whether lift car specification data ELS is different from the lift car specification data ELSX that has been found so far, and judges in step 264 whether magnitude of traffic flow specification data TRS is different from the magnitude of traffic flow specification data TRSX that has been found so far.If ELS is different from ELSX, or TRS is different from TRSX, then will search start mark STR and be changed to " 1 " in step 265, otherwise, then will search start mark STR and be changed to " 0 " in step 266.
Judge that to 265 the reason restart to search is need search optimal set once more when searching the situation variation in step 263, like this, optimal set that very might current record no longer is best.For example, when the magnitude of traffic flow in the building changes because of occupant's change, or some group control algorithm change will be searched when improving systemic-function again.
Referring to Figure 11,, search device 10 and judge whether needs initialization once more according to the result of step 26 in step 27.On the one hand, STR equals " 0 " and means to search and carrying out.On the other hand, STR is not equal to " 0 " and means to search Halfway Stopping and start anew and search, and perhaps, begins to search again after searching end current.Therefore, when STR equaled " 0 ", operating process entered generator program 29, but when STR is not equal to " 0 ", carried out initialize routine in step 28, and then, after to various data initializations, process enters generator program 29.
Referring to Figure 13, the operation of initialize routine 28 will be described below.
Among Figure 13,, from previously stored a plurality of primary data groups, select the primary data group that is suitable for specifying the magnitude of traffic flow in step 281.Each primary data group comprises that initial sets counts the initial sets of PK, PK item and the team control performance data of PK item.
For example, specifying under the situation of conventional time range, from a plurality of primary data GSP1 to GSP4, selecting the primary data group that is fit to conventional operation, with the selected primary data group record DS Data Set that is initialization GPS0 shown in Figure 3 according to magnitude of traffic flow specification data TRS.The DS Data Set of initialization GPS0 comprise initial sets count PK, a plurality of initial sets IPS (1) to IPS (PK) and team control performance data PRI (1) to PRI (PK).
The value that initial sets to be set is counted PK finishes judgment value Pe identical (=30) with deletion.
In step 282, initial sets is counted PK as set number P input; Initial sets IPS (1) imports to IPS (PK) conduct set of records ends EPS (1) to EPS (P); Team control performance data PRI (1) is input into team control performance data PRE (1) to PRE (P) to PRI (PK).That is, as shown in figure 50, the initialization A1 of execute store A2.
In step 283, as initialization, will estimate that times N E is changed to zero, addition record times N R is changed to zero, and the set of estimation is counted NP and is changed to zero, searches permission flag and is changed to " 1 ", and crossover probability CR is changed to 1.0, and variation probability MR is changed to 0.01.So close this end of program.
Referring to Figure 11, step 29 is carried out the generator program corresponding to maker among Fig. 1 12.At first, step 30 judges to search whether continue.Be " 0 " if search permission flag FLAG, process is returned in step 26 and is searched the beginning determining program, and on the other hand, if searching permission flag FLAG is " 1 ", process enters new set generator program in step 31.
Referring to Figure 14, new set generator program will be described below.
Among Figure 14,, judge whether also NE new set is keeping at first in step 311.If estimation set number NP is less than maxim Nmax, owing to there is the also new set of NE reservation, process withdraws from estimation set newly from the step 311 of this program at once.Otherwise if estimate that set number NP is maxim Nmax or bigger, perhaps, if the estimation of all new set is all finished, process enters step 312 and does initialization, Pn is counted in the set of having survived place zero.
At next procedure 313, team control performance data PRE (1) from relative P set of records ends EPS (1) to EPS (P) takes out each average latency AWT (1) to AWT (P) to PRE (P), and the performance estimation value VPS (1) that is placed on the maxim judgement is to VPS (P).Determine that to the inverse of VPS (P) each set is chosen as the probability of superclass (occurrence rate) according to performance estimation value VPS (1).
Then, by the process of repeating step 314 to 324 as described below, generate new set until maxim Nmax.
At first, in step 314 institute is generated set number Pn and add 1.In step 315, determine a numerical value zero and [CR+MR] (crossing-over rate and aberration rate and) between random number, choose at random generating mode.If random number is less than CR (=1.0), chosen manner is " intersection ", if random number is CR (=1.0) or more than 1.0, then chosen manner is " variation ", according to some with reference to the ratio that can revise between crossing-over rate (select intersect probability) CR and aberration rate (select make a variation the probability) MR.
If select " intersection " in step 316, process enters step 317.The reciprocal value that provides respective performances estimated valve VPS for each set is as a weighted value.This weighted value represents to select the probability of this set.Then, in scope, generate two random numbers from " 0 " (as lower limit) to " performance estimation value VPS (1) is to VPS (P) sum reciprocal " (as the upper limit).Then, select two set in conjunction with the random number that is generated.Now, suppose that two superclass (intersect set to) are for the set EPS (PS1) of number PS1 with count the set EPS (PS2) of PS2.
At next procedure 318, produce the random number of numerical value between 0 and 25, and select by a specific parameter PX of this random number.This parameter is shared between two superclass, and has determined to carry out the parameter position of exchange of values.
In step 319, exchange PX numerical value among the superclass EPS (PS1) and PX numerical value among the superclass EPS (PS2) mutually.This operation generates two new set.These two new set be set to the new set NPS (Pn) of number [Pn] and count [Pn+1) new set NPS (Pn+1).
At last, the new set that will generate in step 320 is counted Pn and is added 1.
On the other hand, if select " variation " in step 316, process enters step 321.The reciprocal value that provides respective performances estimated valve VPS for each set is as a weighted value.This weighted value represents to select the probability of this set.Then, in scope, generate a random number from " 0 " (as lower limit) to " performance estimation value VRS (1) is to all sums reciprocal of VPS (P) " (as the upper limit).Then, select a set in conjunction with the value of generation random number.
As step 318, in step 322, produce a random number, to select wherein to produce the parameter PX of variation.
In step 323, another random number of existence between [minimum] and [maximum] (can get specific parameter) by number PX.The numerical value that replaces the middle number of good set EPS (PS1) PX of number PS1 with this random number.In this operation, the set that is generated is exactly the new set NPS (Pn) of number Pn.
At next procedure 324, judge whether the new set with requisite number generates.Because two new set are once generating in " intersection ", if Pn+2＞Nmax can finish the generation of new set.Before this finishes, then generate new set, until number Nmax by repeating step 314 to 324.When finishing the generation of new set, the set number of the new set that will at first estimate in step 325 or the set of estimation are counted NP and are changed to 1.
" intersection " is the lookup method of a kind of convergence solution (convergence of solutions).Otherwise " variation " is a kind of lookup method that changes solution (possessing varied solutions) that has.That is, if process is only undertaken by intersection, search direction limits, and has increased the probability of missing best solution thus.Yet except intersecting, if adopt variation more suitably, system can avoid local solution.In the case, these two kinds of methods are complementary relationship.Yet, even adopt the danger of variation to be that best solution finally finds, also may be destroyed.In the case, these two kinds of methods are again competitive relation.
Therefore, will form the risk of competitive relation for fear of both, and utilize their complementary relationship simultaneously, in first embodiment, CR compares with crossing-over rate, and aberration rate MR is arranged to minimum value.
In the present embodiment, generate 12 (=Nmax) after the new set, team control performance is estimated in each new set relatively.Yet also can adopt other method.
For example, by maxim Nmax is set at " 1 ", can be the team control performance of each " intersection " or " variation " estimation, and can repeats with respect to new set.
In this connection, when maxim Nmax becomes big, can shorten time of run because of new set can generate at once.Yet in the case, system need guarantee that high-capacity RAM10C is arranged.Best, contact time of run and memory span are determined maxim Nmax.
Referring to Figure 11, the estimation routine of step 33 will be described below.This estimation routine is corresponding to the simulator among Fig. 1 13, and it offers simulator and carry out group control algorithm by will newly gathering, and obtains execution result thus.
Referring to Figure 15, below will describe this estimation routine in detail.
Among Figure 15, step 331 generates the simulated conditions data, and it comprises set NPSX, lift car specification data ELSX and the magnitude of traffic flow specification data TRSX that is used to estimate.Promptly, system will newly gather NPS (NP) and be provided with to the set NPSX that is used to estimate, and the lift car specification data ELS and the magnitude of traffic flow specification data TRS that will be included in respectively in the search criterion signal 1a are provided with to lift car specification data ELSX and magnitude of traffic flow specification data TRSX.At next step 332, search device 10 in order to outputing to simulator 2 as the simulated conditions data SIM of simulated conditions signal 13a, make simulator 2 carry out virtual team control operation.In step 333, process is waited for and is finished simulation.
In step 335, will estimate that times N E adds 1, and generate estimated result data RES, it comprise the set NPSY that estimates times N E, be used to estimate (=NPSX) and team control performance data PRFY (=PRF).Then, in step 336, the numerical value that will be used for the set number of new estimation of gathering adds 1.
Referring to Figure 11, the addition program of step 34 is with respect to the adder unit among Fig. 1 15, and it judges whether new set (unit number NP) is recorded.This addition program is described with Figure 16.Among Figure 16,, from team control performance data PRFY, take out average latency AWT, and it is set to performance estimation value VPN and is used for addition record and judges in step 341.In step 342, performance estimation value VPN and additional reference value BX are compared, whether to be recorded in the memory device to judge it.If VPN is equal to or greater than BX, system does not allow this record, and terminator 34 immediately.
Otherwise, if VPN less than BX, process enters step 343, makes addition record times N R add 1, produces the data RAP be used for addition record thus, it comprise addition record times N R, estimate set NPSZ (=NPSY) and team control performance data PRFZ (=PRFY).In step 344, new set simultaneously, is made the numerical value of the set P that has write down add 1 by the set of addition record for number [P+1].
Referring to Figure 11, the delete program of step 35 is corresponding to the delete cells among Fig. 1 16, and its deletes the set of performance estimation value difference.
Referring to Figure 17, delete program is described now.Whether among Figure 17, compare set of records ends in step 351 and count P and deletion beginning judgment value Ps, be the time of deletion record to judge it.If P, judges then that it is not the orthochronous that deletion damages set less than Ps, process quits a program 35 immediately.If P is equal to or greater than Ps, judge that then it is the orthochronous that deletion damages set, so damage set, become deletion until set number P and finish judgment value Pe by repeating step 352 to 359 deletions.
In step 352, take out average latency AWT (1) respectively to AWT (P) from team control performance data PRE (1) to PRE (P), and respectively it is set to performance estimation value VPD (1) to VPD (P).Carry out initialization in step 353 then, the damage set that detection will be deleted.That is, be used in the counting machine RC that searches and be changed to 1, make the worst value WVPE of performance estimation value be changed to zero, and make deletion set number RP be changed to zero.
Determine to have the set (set number RP) of the worst performance estimation value by the process of repeating step 354 to 357.That is, when determining that in step 354 its performance estimation value VPD (RC) is inferior to any set of the worst existing value WVPE, promptly be set at the worst value WVPE with performance estimation value VPD (RC) is up-to-date in step 355.The Counter Value RC that will be used to search is set at deletion set number RP.In step 356 counting machine RC is added 1, and judge whether finished searching of all set relatively in step 357.
In step 358, deletion has the record of the set of the worst value WVPE (its deletion set number is RP), and deletion team control performance data record PRE (RP).The set numerical value P that has write down also subtracts 1.Like this, the set of reservation is newly numbered from 1 lifting successively, so set is stored again, process finishes in step 358 then.
In step 359, judge that set after the deletion counts that P equals or finish decision content Pe less than deletion.As denying the process of system's repetition above-mentioned steps 352 to 358.When P becomes when being equal to or less than Pe, this delete program finishes.
In the present embodiment, although step 351 will be deleted beginning judgment value Ps and deletion end judgment value Pe is set to 50 and 30 respectively, they also can be set to other number.
Deletion beginning judgment value PS is arranged at a scope, and wherein this numerical value is no more than the maxim Pmax that can be stored in the set in the RAM10C.If deletion beginning judgment value is changed to Ps=Pe+1, a set by addition record whenever, all having one, to replace set deleted.When RAM10C did not have enough memory capacitys, this method was very easily.
Some set that the deletion end judgment value Pe of step 359 means reservation become superclass.If it is little that deletion finishes judgment value Pe, owing to be difficult to keep the variation of the new set that is generated, the probability that generates good new set will reduce.Otherwise, be big if deletion finishes judgment value Pe, the new set that is generated will be guaranteed to change, the result, the probability that generates good new set will increase.Yet as the actv. search procedure, not wishing provides bigger number for deletion finishes judgment value Pe, because will increase generating run like this.
Therefore, wish according to the kind of the combination of carrying out two set that intersect, controlled variable and quantity or the like, or sometimes by test with make mistakes to determine that deletion finishes judgment value Pe.Because Pe=30, present embodiment has been guaranteed combination (=30 * 29 ÷ 2).
Referring to Figure 11, the additional reference value update routine of step 36 is with respect to the additional reference value modifier 18 among Fig. 1, and it revises additional reference value BX according to the recording status of set in the memory device 11.
Figure 18 has described the additional reference value update routine.Among Figure 18,, take out each average latency AWT (1) to AWT (P) from team control performance data PRE (1) to PRE (P) in step 361, and place performance estimation value VPE (1) to VPE (P) in order to set reference value.In step 362, system's operation is to determine the worst value WVPE at performance estimation value VPE (1), in order to set reference value to VPE (P).This operation is identical with the process of step 353 to 357.In step 363, obtain modification value CBX by calculating [the worst value WVPE one right value AVPE], in step 364, modification is worth CBX places additional reference value BX that it is made an amendment.
In the present embodiment, compensation value AVPE is placed regularly one second that searches between beginning and the end.That is, additional reference value BX (referring to the average latency) being set made it less than one second phase weekly.Yet system also can use other numerical value.
If compensation value AVPE is arranged to a higher value, it is strict that the condition of addition record becomes gradually, so it will not be a very large value, so that obtain good set as much as possible in limited estimated time.Otherwise,, just can increase the probability that addition record has many set of similar features and a little performance difference if compensation value AVPE is arranged to zero second.Therefore, must determine this numerical value suitably according to search criterion.
Referring to Figure 11, searching of step 37 finishes determining program corresponding to searching end judging unit 17, and its judges whether search optimal set finishes.Below be described with Figure 19.
Among Figure 19, step 371 is according to estimating times N E and searching end judgment value NEa and judge whether search procedure finishes.If NE＜NEa judges that then search procedure thoroughly do not finish, and will search permission flag FLAG in step 372 and be changed to " 1 ", continue search procedure.If NE 〉=NEa judges that then search procedure all carries out, and will search permission flag FLAG and place " 0 ", finish search procedure.
In the present embodiment, be set to 1,000 although search end judgment value NEa, judgment value NEa also can be worth for other.
Usually be difficult to determine to estimate that how many times is just enough actually.This is because various search criterions are depended in the convergence of searching substantially, such as kind and quantity, the content of initial sets, the method that generates new set and the condition of addition record of controlled variable.
In order to obtain to represent the many good set of good team control performance, search end judgment value NEa and should place a big as far as possible value.Yet, become too big if search the accumulated value NE of number of times, will spend the considerable time finishes search procedure, causes the relatively poor search procedure of effect thus.Therefore, in order to obtain many good set effectively, must determine to search end judgment value NEa according to search criterion.
Referring to Figure 11, the optimal set of step 38 takes out the extractor 20 of program corresponding to Fig. 1, in order to take out an optimal set from each set.Be described hereinafter with reference to Figure 20.Among Figure 20, step 381 is taken out each average latency AWT (1) to AWT (P) from team control performance data PRE (1) to PRE (P), and places performance estimation value VPS (1) to be used for optimum value to VPS (P) to judge.Then, at step 382 system initialization to detect optimal set.That is, the counting machine RC that searches is placed 1; The best values BVPE of performance estimation value is placed 9,999; To gather number BP and place zero.
Next, determine to have the optimal set (set number BP) of top performance estimated valve by the process of repeating step 383 to 386.That is,, the performance estimation value VPS (RC) and the best values BVPE of previous acquisition compared in step 383.Be better than best values BVPE if detect performance estimation value VPS (RC), performance estimation value VPS (RC) placed best values BVPE, the value of the counting machine RC that searches is placed set number BP in step 384.In step 385 the counting machine RC that searches is added 1.In step 386, judge whether searching of all set finishes.Generate optimal set data BPD in step 387, it comprises optimal set BPS, lift car specification data ELSY, magnitude of traffic flow specification data TRSY and searches status data SS.Promptly, to have best values BVPE set in be placed in optimal set BPS, the content identical with magnitude of traffic flow specification data TRSX with the lift car specification data ELSX of simulated conditions data SIM is changed to lift car specification data ELSY and magnitude of traffic flow specification data TRSY respectively.Estimation numerical value NE till will arriving this moment is changed to and searches status data SS.
At last, the optimal set signal 20 that will comprise optimal set data BPD in step 388 is delivered to group control device 1.
Referring to Figure 11, as mentioned above, search optimal set if finished, process is returned step 26, and repeated execution of steps 26,27,30 to 38, searches until detecting to have finished, and will search permission flag FLAG and reset to " 0 " in searching the end determining program.In search procedure,, then carry out and search again if the content of lift car specification data and magnitude of traffic flow specification data changes.That is, search permission flag FLAG and become " 1 ", carry out from each step of step 31 beginning.
The advantage of first embodiment
As mentioned above,, good set can be generated effectively, optimal set can be searched effectively thus according to first embodiment.Because the simulator 2 of team control is and group control device 1 independent device mutually, can disturb original team control to operate so search optimal set.
Because new set generates with two kinds of methods, so first embodiment can accept " intersection " and " variation " characteristic effectively.In other words, the new set of generation can have suitable variation and convergence simultaneously, so combine by extensive and local search, can find optimal set earlier.
In first embodiment, superclass according to the performance estimation value with select probability to each superclass weighting after just selection, so system can improve the probability of selecting good superclass, in other words, system can improve the probability that generates good new set, and the outstanding character of superclass has been inherited in these new set.
Adopt first embodiment, because additional reference value is revised according to the worst value in a plurality of performance estimation values, system can avoid needing the useless process of deletion immediately behind addition record, and it is tighter to become gradually thus.End is judged, the superclass condition is revised selects to count up to according to addition record with parameter if search, and then system can carry out suitable processing according to searching the progress situation.
In first embodiment, also can see,, and be maintained, realize rational set record so will consider the capacity of memory device because the set of being write down can be less than certain number that is drawn by delete procedure.As a result, can from set as much as possible, select the more good set.
Adopt first embodiment,,, when generating new set, always provide good set thus as the superclass so system can the retention property estimated valve be good set because the set of performance estimation value for difference constantly deleted by system.
In first embodiment,, and provide optimal set at its mouth even extractor is also searched optimal set in search procedure.Therefore, even in search procedure and between used life, group control device 1 also can obtain optimal set and to be found finishing such as need not.
Because group control device 1 can obtain to search status data (estimating times N E) from optimal set in search procedure, the value so the optimal set that it can be provided use in search procedure judges rightly.
Adopt first embodiment, search continuously and can proceed to the time of searching and reach till certain number, prevent from before enough searching, to finish search procedure.
And in first embodiment, system also has locating function again, and like this, when any lift car specification data and the variation of magnitude of traffic flow specification data, even after searching end, good set also can restart to search automatically in system.Therefore, even for a certain reason, the instruction that begins to search of sending from group control device has been delayed, and system still can begin rapidly to search.In early days the stage, system can obtain the optimal set corresponding to up-to-date team control condition thus.When lift car specification data or the variation of magnitude of traffic flow specification data, even during searching, system still can utilize again locating function automatically to start anew to search again under new team control condition.
In first embodiment, prepare initial sets group in advance corresponding to each magnitude of traffic flow specification.When beginning to search, only initial sets group can select to the magnitude of traffic flow in system when himself initialization.Therefore, system can provide the set of optimizing degree as superclass from the beginning, realizes thus searching rapidly.When the set provide as optimal set was provided in search procedure, even at the commitment of search procedure, good team control performance still can shared by system on certain program.
In first embodiment, the team control performance data PRF that simulator 2 is obtained is stored in the memory device 11.This team control performance data PRF comprises a plurality of data shown in Figure 7.The performance estimation value VPE (1) that be included in performance estimation value VPN that some data in the team control performance data PRF are used to addition record, is used to set reference value to VPE (P), the performance estimation value VPD (1) that is used to delete judgement is to VPD (P) and be used for performance estimation value VPS (1) that optimum value judges and replace to VPS (P).Therefore, obtain the simulation that whenever do not need of each performance estimation value at needs.Comprise shared data when (such as the average latency) in the performance estimation value, have only shared data to be stored naturally as team control performance data PRF.
[generation method and system of selection outline]
Be divided into two kinds by n for the generation method that groups of individuals (superclass) generates the individuality (subclass) of (n+1) Dai Xin.First method is that newly-generated individuality (subclass) is removed to generate next new individuality (subclass) as superclass, and the latter belongs to and superclass the same generation i.e. (n+1) generation.On the other hand, second method is exactly not adopt aforesaid way, especially, and wherein:
N is for groups of individuals in Gn (Mn) expression, and its compilation length is Mm;
The groups of individuals that Gn* (j) expression is new, its compilation length is j, comprises new individual and generated on groups of individuals Gn (Mn) basis at least;
Gn (i) expression has the individuality of n algebraically i; And
Gn* (j) expression has the new individuality of new groups of individuals Gn* (j) number j, two kinds of methods is arranged in order to the newly-generated new individual gn* (j+l) with number (j+l) that will increase, and generates new groups of individuals Gn* (j+l).[generation method A]: in generation method A, have only contemporary groups of individuals Gn (Mn) to generate new groups of individuals gn* (j+l) as subclass from intersection or variation as superclass.[generation method B]: in generation method B, all or part of contemporary groups of individuals Gn (Mn) and new groups of individuals gn* (j) are used as superclass, to generate new individual gn* (j+l) as coming from the subclass that intersects or make a variation, wherein:
Gn(Mn)＝{gn(l)，gn(2)，…，gn(Mn)}；
Gn*(j)＝{gn*(l)，gn*(2)，…，gn*(j)}；
Gn* (j+l)={ gn* (l), gn* (2), gn* (j), gn* (j+l) } note another method (hereinafter referred to as " generation method Ba ") being arranged as modification, wherein to generation method B, only adopt the individuality that is suitable for superclass, it makes the new groups of individuals Gn* (j) of a part become superclass.
On the other hand, whether be left of future generation individual according to contemporary individuality, can be to the system of selection classification of groups of individuals of future generation.That is, wherein Gn* (Mn*) represents new groups of individuals, and its compilation length is Mn*; Gn+l (Mn+l) represents groups of individuals of future generation, and its compilation length is Mn+l, and two kinds of method A and B are as described below.[system of selection A]: system of selection A is a kind of like this method, wherein, Mn+l the individuality of under certain threshold condition that departs from new groups of individuals Gn* (Mn*), selecting (new individual gn+l (i) (i=l ... Mn+l), as groups of individuals Gn+l of future generation (Mn+l).Adopt this method, it is of future generation individual that contemporary groups of individuals Gn (Mn) never is left.[system of selection B]: system of selection B is another kind of method, wherein, new individual gn+l (the i) (i=l of the Mn+l individuality of under certain threshold condition that departs from all or part of contemporary groups of individuals Gn (Mn) and new groups of individuals Gn* (Mn*), selecting, Mn+l), as groups of individuals Gn+l of future generation (Mn+l), as long as:
Gn*(Mn*)＝{gn*(l)，gn*(2)….，gn*(Mn*)}；
Gn＋l(Mn＋l)＝{gn＋l(l)，gn＋1(2)，…，gn＋l(Mn＋l)}。Note another kind of method (hereinafter referred to as [back-and-forth method Ba]) being arranged, wherein, in the middle of current groups of individuals Gn (Mn), can not can not keep as of future generation individual as the individuality of the superclass of new groups of individuals Gn* (Mn*) as modification to [back-and-forth method B].Moreover, also have another kind of method (hereinafter referred to as [back-and-forth method Bb]) as a kind of modification to back-and-forth method B, wherein, be not suitable as the new individuality of the superclass of new groups of individuals Gn* (Mn*), can not keep as of future generation individual.[generation method and system of selection in conjunction with example] (A) carries out the optimal set lookup method in conjunction with method of formation B and back-and-forth method B, and this method is as follows:
In the middle of this combination, the function of [method of formation B] is distributed to maker 12 and addition unit 15, the function of [back-and-forth method B] is distributed to addition unit 15 and delete cells 16.
First embodiment method of formation Ba as modification to [method of formation B], back-and-forth method Bb as modification to [back-and-forth method B].From the addition unit 15 of first embodiment new set only addition record be suitable as the set of superclass, and they are used for the viewpoint of next new set as one of superclass, can think that unit 15 and maker 12 are shouldered the responsibility of the function of [method of formation Ba].Simultaneously, from the addition unit 15 of first embodiment only addition record be suitable as the set of superclass, and with they viewpoints as possible superclass of future generation, addition unit 15 and delete cells 16 are shouldered the function of [back-and-forth method Bb].(B) when in conjunction with [method of formation A] and [back-and-forth method A], this method is as follows.Memory device 11 is divided into two, promptly is divided into and gathers group zone and addition record set group zone the present age.Then, maker 12 generates a new set group the good set group from the present age.Addition unit 15 utilizes certain benchmark of new set group to select and a plurality of set of addition record.On the other hand, as long as the addition record number reaches a predetermined value (for example Pe), delete cells 16 is just deleted all set in present age groups, and the set group of addition record upgrades this generation through displacement.Then, repeat these steps.
In this kind combination, the function of method of formation A is distributed to maker 12 among first embodiment, the function of [back-and-forth method A] is distributed to addition unit 15 and delete cells 16.(C) when in conjunction with [method of formation A] and [back-and-forth method B (or Bb)], its method is as follows:
Therefore, in the middle of this kind combination, the function of [method of formation A] is distributed to maker 12 among first embodiment, the function of [back-and-forth method B (or Bb)] is distributed to addition unit 15 and delete cells 16.(D) when in conjunction with [method of formation B (especially be Ba] and [back-and-forth method A], its method is as follows:
Adopt this kind combination, the function of inciting somebody to action [method of formation B (especially being Ba)] is distributed to maker 12 and the addition unit 15 of first embodiment, and the function of [back-and-forth method A] is distributed to addition unit 15 and delete cells 16.[performance estimation value]
As a reference, each used performance estimation value is summarized as follows according to the device and the purposes that realize in the present embodiment: the performance estimation value that [19] optimal set is judged: VPS (l) is to VPS (P)
Device: extractor 20.
Purposes: select optimal set.[20] the performance estimation value of addition record judgement; VPN
Device: addition unit 15.
Purposes: judge the new set of addition record.[21] the performance estimation value of deletion judgement: VPD (1) is to VPD (P)
Device: delete cells 16.
Purposes: judge deletion at set of records ends.[22] the performance estimation value of additional reference value is set: VPE (1) is to VPE (P).
Device: reference value modifier 18.
Purposes: for referencial use when revising additional reference value BX.[23] the performance estimation value that superclass is selected probability is set: VPS (1) is to VPS (P).
Device: maker 12.
Purposes: select superclass.
Wherein, for each set, various aspects are depended in its good set.For example, it depends on what degree is the controlled target of being instructed by group control device satisfy to.At this moment, can utilize directly as the performance estimation value and order the relevant team control performance valve (see figure 7) of controlled target.
About intersecting, the good property as the superclass of each set depends on the variation in the memory device.This is because the intersection of being made up of the set that has different performance mutually can improve the probability of the better set of generation.
" allocation index " can be used as its variation of performance estimation value representation.Be in each set at center relatively, this allocation index for example can be defined as other set number, makes the distance between the set be equal to or less than certain value.Wherein, the distance between the set forms the defined hyperspace definition of element with a plurality of set.Similarity between allocation index is represented to gather, index value are low more, and be excellent more as the performance of superclass.
Allocation index can be defined as the total distance with other set.In the case, index value is big more, and is excellent more as the performance of superclass.Allocation index can be defined as the distance of those and this set other set number greater than a predetermined value.In the case, index value is big more, and is excellent more as the performance of superclass.
Next, describe the method for searching the performance estimation value in detail.In first embodiment, any performance estimation value includes average latency AWT.Yet, might change the content of each performance estimation value according to its purposes.For example, utilize mutual different performance estimation function can calculate each performance estimation value E.
In a word, the performance estimation function relevant with the team control performance represents that with following formula wherein F (X) is the function of X.
E＝F(X1，X2，…，Xi，…，Xn，T1，T2，…，Ti，…，Tn)
Wherein,
N: the estimation item number of team control performance,
Xi: estimate i (i=1,2 ...., performance valve n),
Ti: an estimation i (i=1,2 ..., performance reference value n).
Performance reference value Ti represents that one [expected value] reaching the most at last as the team control performance maybe must satisfied [limit].[limit] comprises [higher limit] and [lower limit].No matter the performance reference value that provides is as [expected value] or as [higher limit] or [lower limit], all do different determining according to the team control purpose that is provided with.
Note, when performance reference value Ti means " expected value " relevant with performance estimation function [24], | Xi-Ti| becomes more little, and it is good more that performance will become.When performance reference value Ti referred to [higher limit], (Ti-Xi) become big more, it is good more that performance will become.When performance reference value Ti referred to [lower limit], (Xi-Ti) become big more, it is good more that performance will become.
In a word, the performance estimation value shown in the function [19] to [23] depends on performance estimation function, estimation item wherein and performance reference value.
Below be some special example, in order to the operational method of explanation performance estimated valve.[25] first routine performance estimation functions (in first embodiment): controlled target: [reducing the average latency as much as possible].Performance reference value: T1=0 second.(T1: [expected value] of average latency).Performance estimation function: E=|AWT-T1|=AWT (AWT: the average latency).Addition record is judged: and E＜BX (BX: additional reference value, for example, BX=15 second .).[26] second routine performance estimation functions (under the situation of following second embodiment): controlled target: [making the average latency approach predetermined value as far as possible] performance reference value: T1=20 second.(T1:[average latency [expected value]).Performance estimation function: E=|AWT-T1| (AWT: the average latency).Addition record is judged: and E＜BX (BX: additional reference value, for example, BX=3 second).[27] the 3rd routine performance estimation functions: controlled target: [wait rate RLW and predicated error rate RPE approach each expected value as far as possible when making average latency AWT, length].Performance reference value: T1=15 second, T2=2%, T3=3% (wherein, T1: [expected value] of average latency, T2: when long [expected value] of wait rate, T3: [expected value] performance estimation function of predicated error rate: E=Ap * | AWT-T1|+Ag * | RLW-T2|+Ar * | PRE-T3| (wherein, Ap, Ag and Ar are coefficient of weight).The judgement of addition record: E＜BY (BY: total estimation reference value, for example, BY=10).[28] the 4th routine performance estimation functions: controlled target: [wait rate RLW and predicated error rate RPE are as far as possible little when making average latency AWT, length usually].The performance reference value: T1=0 second, T2=0%, T3=0% (T1 wherein: [expected value] of average latency, T2: when long [expected value] of wait rate, T3: [expected value] of predicated error rate.Performance estimation function: E=AP * AWT+Ag * RLW+Ar * RPE (wherein Ap, Ag and Ar are coefficient of weight).Addition record is judged: and E＜BY (BY: total estimation reference value, for example, BY=1000).[29] the 5th routine performance estimation functions: controlled target: [estimate item number make average latency AWT, wait rate RLW and predicated error rate RPE are retained in its permission separately as much as possible when long increases in scope] performance reference value: T1a=15 second, T1b=3 second, T2=2%, T3=3% (T1a wherein: [expected value] of average latency, T1b: average latency permissible aberration scope, when T2:[is long [higher limit] of wait rate, T3: [higher limit] of predicated error rate.Performance estimation function: E=f (| AWT-Tla|-T1b)+f (RLW-T2)+f (RPE-T3) (wherein, f (X) represents a function, i.e. f (X)=1, X 〉=0 and f (X)=0, X＜0).Addition record is judged: and E＜BY (BY: total estimation reference value, for example, BY=1).[30] the 6th routine performance estimation functions: controlled target: [average latency AWT is remained in its predetermined tolerance band from its best values (minimum value), and reduce wait rate RLW when long as much as possible].Performance reference value: Tla=BVPE (second), Tlb=2 second, T2=2% (wherein, T1a: [expected value] of average latency, T1b: the permissible aberration scope of average latency, T2: when long [expected value] of wait rate).Performance estimation function: E=(100-RLW) * f (T1b-| AWT-BVPE|) (function of f (X) expression wherein, i.e. f (X)=1, X 〉=0 and f (X)=0, X＜0). optimum value is judged: Max{E}.
The first and second routine performance estimation functions [25], [26] are applicable to estimates that item only is the also very simple a kind of situation of an average latency and its controlled target, therefore, can produce estimation function with comparalive ease.Utilize estimation item shown in Figure 7 replace the average latency be very easy to (certainly, though estimate item exceed shown in Figure 7 also can).
On the other hand, for a plurality of estimations, because that controlled target becomes is variable, it is complicated that the performance estimation function will become.When adopting the third and fourth performance estimation function [27] and [28], in conjunction with the deviation between estimation item and the expected value to performance estimation function weighted calculation.Wherein, the method for each estimation of contact relative importance value estimating target realization degree generally is known usually.Especially, when the controlled target of estimating item was opposite mutually, this was a method very easily.
Shown in the 5th performance estimation function [29], wherein, the team control performance valve drops on (for example judges tolerance band, be equal to or less than higher limit, be equal to or higher than lower limit, perhaps be retained in certain value with departing from of expected value) the quantity of estimation item, estimate to search according to each, and can estimate the team control performance according to this quantity.
Moreover each estimated valve is respectively according to two or more different performance estimation function calculation, and finishes according to the combination of a plurality of estimated valves that addition record is judged, deletion is judged, optimal set is judged or the like.
Shown in the 6th performance estimation function [30], two conditions are wherein arranged, be to maintain the average latency apart from minimum value (E1=|AWT-BVPE|, in predetermined tolerance band of E1 〉=T1b), wait rate (E2=RLW therein when long, Min{E2}) be minimum, at first can through performance estimation function E1 select candidate's optimum value, can finally select optimum value by through performance estimation function E2 then.Shown in the 6th performance estimation function [30], can comprehensive two groups of performance estimation functions and two conditions, rewrite a performance estimation function and a Rule of judgment, select optimal set to utilize performance estimation function and Rule of judgment.
In first embodiment,, also can divide this selection probability equally although the selection probability of superclass depends on the performance estimation value.Under situation as first embodiment, trend towards generating new set with similarity, therefore, the variation that is stored in the set in the memory device 11 can disappear.When this changes disappearance,, also may exist the convergence problem of local solution even in the starting stage of searching.Therefore, maybe must reduce under the situation of calculated amount, the selection probability of the superclass that provides is divided equally in the problem that must avoid initial convergence.[second embodiment]
Referring to Figure 21 to 25, second embodiment will be described below.In describing second embodiment, its part different with first embodiment is described mainly.
Figure 21 is a figure corresponding to Fig. 1, represents second embodiment.Among Figure 21, performance reference value setting device 3 comprises a Personal Computer, and it provides reference value signal 3a for group control device 1.Reference value signal 3a comprises [the control reference value] that device is searched in team control performance [performance reference value] and control.In the present embodiment, the performance reference value is average latency [expected value]; The control reference value provides [designated value] to additional reference value BX.Reference value signal 3a can directly present to searching device 10 (estimator 13, reference value updating block 18, again search unit 19 and initialization unit 21).
Figure 22 represents the memory contents of RAM10C corresponding to Fig. 3.Among Figure 22, TGT is for forming one of data item of holding within the search criterion signal 1a, it comprises the data TAW (wait time expected value) of expression average latency AWT it [expected value], and the data TCB (additional reference designated value) of the designated value of expression additional reference value BX.For example, wait time expected value TAW placed 5 seconds, and additional reference designated value TCB placed 3 seconds.TAWX is the data of being transcribed by wait time expected value TAW, in order to the calculated performance estimated valve.
When input reference signal 3a, group control device 1 takes out performance reference value (wait time expected value TAW) and the control reference value (additional reference designated value TCB) that is included in the reference value signal 3a.Operation as step 234 among Figure 10, group control device 1 is fed to search criterion signal 1a and searches device 10, and it comprises lift car specification data ELS, magnitude of traffic flow specification data TRS, look-up command data SCM, wait time expected value TAW and additional reference designated value TCB.If search procedure begins (seeing Figure 11), then in searching device 10, finish generation, estimation, addition, deletion and additional reference value successively and revise, this process and first embodiment are similar.When beginning determining program 26 when searching device 10 input search criterion signal 1a (seeing Figure 25) according to searching, RAM10C storage lift car specification data ELS, magnitude of traffic flow specification data TRS, look-up command data SCM, wait time expected value TAW and additional reference designated value TCB, as shown in figure 22.
In second embodiment, addition program 34 and delete program 35 have special feature.In second embodiment, according to the performance estimation function shown in the second performance estimation function [26] (E=|AWT-T1| calculated performance estimated valve, and relatively performance estimated valve E and additional reference value BX mutually.Less than additional reference value BX, then determine addition record as performance estimation value E.Promptly with the designated value difference of actual value and expected value relatively, when this difference during less than designated value, the set that addition record is new.Therefore, in the step 341 (seeing Figure 26) of addition program 34, execution calculating [VPN＜---| AWT-TAWX|], addition record is judged set performance estimation value VPN thus.
In the step 352 (seeing Figure 17) of delete program 35, and execution calculating [VPD (P)＜---| AWT (P)-TAWX|], deletion is judged set performance estimation value VPD (1) thus to VPD (P).
Figure 23 represents the content of the additional reference value update routine 36 of second embodiment.
Among Figure 23, the additional reference designated value TCB substitution modification value CBX that reads in step 40.At next step 402, rewrite additional reference value BX with modification value CBX.In step 403, read wait time expected value TAW, rewrite the current performance reference value of using (wait time expected value TAWX) with this value.
When the modification finished additional reference value, as carrying out judgement that search procedure is finished among first embodiment and to extract (the seeing Figure 11) of optimal set, suppose in the program of extracting 38 of optimum program, make the method for calculating of performance estimation value be different from first embodiment.Promptly, in the extract step 381 (seeing Figure 20) of program 38 of optimal set, in such a way optimal set is judged calculated performance estimated valve VPS (1) to VPS (P), promptly [VPS (1)＜---| AWT (1)-TAWX|, ..., VPS (P)＜---| AWT (P)-TAWX|].
Extract after the program 38 having handled optimal set, process enters the step of searching determining program 25 again.Referring to Figure 24, below will be described in and search again in the determining program 25, search the control step of beginning determining program 26.Notice that Figure 24 is equivalent to Figure 12 of first embodiment.
Among Figure 24, in step 261, provide search criterion signal 1A by group control device 1, then, store lift car specification data ELS, magnitude of traffic flow specification data TRS, look-up command data SCM, wait time expected value TAW and additional benchmark designated value TCB by RAM10C.Then, begin determining program 26 similar (seeing Figure 12) with searching of first embodiment, to 264, SCM changes into [1] from [0] with the look-up command data in step 262, and detects lift car specification data ELS or magnitude of traffic flow specification data TRS.In step 265, when detecting at least one and change, search opening flag STR and place [1], and instruct the beginning of search procedure after by following first kind of mode initialization.
On the other hand, if any of look-up command data SCM, lift car specification data ELS and magnitude of traffic flow specification data TRS do not represented to change, then judge in step 267 whether wait time expected value TAW changes.That is, the wait time expected value TAWX and the wait time expected value TAW of Set For Current compared mutually, to judge any variation.If TAW is not equal to TAWX, then be labeled as [changing], will search opening flag STR in step 265 then and place [1].Otherwise,, then be judged as [not changing] if TAW equals TAWX, the operation of execution in step 269.
In step 269, judge whether additional reference designated value TCB changes.That is, the additional reference value BX of additional reference designated value TCB and Set For Current is compared mutually, to determine to exist any variation.If BX equals TCB, then be judged as [not changing], will search opening flag STR in step 266 then and place " 0 ", thus, if searching optimal set, then the maintenance state if search procedure is finished, then keeps its completion status.Otherwise,, then be judged to be [changing] if BX is not equal to TCB, in step 268, search device 10 and will search opening flag STR and place [2], then by after the second way initialization, the beginning of instruction search procedure.Adopt the initialization of first kind of mode to comprise to the initialization of memory device and total initialization (to estimating the initialization of times N E, addition record times N R or the like).The initialization of the employing second way is meant the total initialization procedure except memory device.
When at step 269 change additional reference value BX, carry out being, even the value that additional reference value BX changes into more restricted (promptly littler) also formerly continues search procedure in the scope of search procedure without doubt by the initialized reason of the second way.From the viewpoint of convergence, when utilizing the generation that is used as initial sets and selecting set to restart search procedure, can obtain better search efficiency.Especially, when the performance estimation function only comprised mono-estimation item and only changes additional reference value BX, the initialization of the second way was just more suitable.
Supposition is now searched opening flag STR and is placed [0], and like this, whether judged result just becomes allows optimal set rest on current state (continuation state or done state), still from the beginning restarts.In the case, the step 27 of process from Figure 11 enters step 30, and judges according to the numerical value of searching permission flag FLAG whether search procedure is carried out or finished.If search procedure is carried out, FLAG equals [1], and process enters step 31, searching end determining program 37, will search permission flag FLAG and be re-set as [0], repeating step 31 to 38 thus, 26 and 27 operation finishes until judging to search in step 27.
Finished if search, searched and finish determining program 37 and will search permission flag FLAG and place [1], like this,, process has been waited for searched again by repeating step 26,27,30,26 successively.
If during searching or change lift car specification data ELS, magnitude of traffic flow specification data TRS, performance reference value TAW or control reference value TCB afterwards, then will search opening flag STR and place [1] or [2] by searching beginning determining program 26, process enters initialize routine 28 from step 27.
Figure 25 represents initialize routine 28.
Figure 25 is corresponding to Figure 13 of first embodiment.Referring to Figure 25, it judges according to the value of searching opening flag STR whether the initialization of memory device is essential in step 284.If specify the initialization of adopting first kind of mode, place [1] if promptly will search opening flag STR, then be similar to the initialize routine 28 (seeing Figure 13) of first embodiment, read initial sets group and team control performance data corresponding to magnitude of traffic flow specification data TRS in step 281.Then, in step 282, utilize initial sets group and team control performance data to carry out initialization.
In step 285, start the update routine of additional reference value.Its content of the update routine of the additional reference value of step 285 is identical with the content of program shown in Figure 23 36.In this step 285 new additional reference value (BX) and new performance reference value (wait time expected value TAWX) are set.Then, in the initialize routine 28 (see Figure 13) of step 283, also carry out total initialization as first embodiment.The initialization of first kind of mode of employing that Here it is.
On the other hand, when step 284 is judged when searching opening flag STR for " 2 ", promptly when specifying the initialization of passing through the second way, then only total initialization of execution in step 283.That is, needn't execution in step 281 and 282 operation, will before searching beginning, be recorded in set group in the memory device as the initial sets group.Especially, when finishing to search at last, in memory device 11, kept good set EPS (1) to EPS (P), and used its team control performance data PRE (1) to PRE (P).The initialization of the employing second way that Here it is.
As mentioned above, in second embodiment, utilize performance reference value setting device 3, the expected value TAWX of average latency and the designated value BX of additional reference value can be provided from the outside.Therefore, can or search team control and find the optimal set that meets required strategy.
In second embodiment, during searching or search the variation that detects reference value TGT (wait time expected value TAW and additional reference value BX) after the end, can search again thus.Therefore, when the control policy of people's team control for a change, or the people just automatically performs again and searches, so can obtain optimal set rapidly for a change during reference value TGT.
In second embodiment, can the situation when beginning to search select the initialization of the first kind of mode or the second way, so just can carry out suitable initialization.For example, when revising additional reference value BX a little, can utilize the rapid execution of initialization of the second way to search again.When additional reference value BX change is big, just need naturally to have adopted the initialization of first kind of mode.Otherwise, in second embodiment, although when detecting the variation of performance estimation function (estimating item, performance reference value, structure), employing be the initialization of first kind of mode, when change than hour, also can adopt the initialization of the second way.
In brief, adopt first kind still to be that the initialized mode of the second way selects to have produced such problem, be exactly which set group, promptly can under new search criterion, realize effectively searching with the initialized set group of GPS0 at one that obtained at that time.Yet a kind of like this computing time of judging that needs are considerable is so be considered as unrealistic.Therefore, need be according to above-mentioned change item and change amount selection mode.
Above-mentioned performance estimation function is an example, thus also can adopt other function, such as performance estimation function [24] or first to the 6th routine performance estimation function ([25]-[30]) or the like.
Above-mentioned performance estimation value is not only the performance estimation value of addition record judgement, and is performance estimation value [19], the performance estimation value [21] of deletion judgement and performance estimation value [22] or the similar value that reference value is judged that optimal set is judged.[the 3rd embodiment (another embodiment of delete cells)]
Below another embodiment of delete cells 16 will be described according to Figure 26, and illustrate to be primarily aimed at its part different with first embodiment.
Figure 26 represents the delete program 35 of the 3rd embodiment.This diagram of circuit is corresponding to Figure 16 of first embodiment.
At first,, calculate a formula NRH=NR-NRX, count NRH, new record to count NRH be the addition record number, be used for the new set of existence after the delete procedure in the end to calculate new record in step 411.Record when in the end judging is counted NRX and is one counts the represented value of NR by addition record when having carried out last delete procedure.
Next, in step 412, relatively new record is counted NRH and deletion beginning judgment value NRa, to have judged whether the time of deletion.Deletion beginning judgment value NRa for example is set at 10 times.
If NRH＜NRa then is judged as the time that is not the deletion set, then, process directly quits a program 35.On the other hand, if NRH 〉=NRa, being judged as is deletion time, then, at that time addition record numerical value of N R is inserted final record when judging count NRX to upgrade.
Then, by repeating step 414 to 422 deletion set, reach deletion until units P and finish judgment value Pe.In the step 283 (seeing Figure 13) of initialize routine 28, the initial value of NRX is initialized as [0] (not shown).
In step 414, the distance D ST between the set of computations (i, j) (wherein, i, j=1,2 ..., p; I ≠ j).(i, j), (i, relation j) is as follows in relevant two set for distance D ST.
DST (i, j)=|| EPU (i)-EPU (j) || [31] (wherein, i, j=1,2 ..., P; I ≠ j).
If the parameter value EPS (i) of each set, (i=1,2 ..., P) done normalization method with respect to each parameter.Be that each parameter value is expressed as the possible peaked ratio that is drawn by each parameter relatively, it is worth between 0 to 100.
For example, if be assumed to 50,000 as one of parameter by the possible maxim that fully laden estimation coefficient Ca is got, the normalized parameter value of relevant fully laden estimation coefficient Ca is calculated as 20 ({=(10 in set EPS (1) shown in Figure 8,000 ÷ 50,000) * 100}.Equally, (maxim: 1,600) during normalization method, it is calculated as (400 ÷ 1,600) * 100=25 as the parameter value Cb of prediction error coefficient.
Owing to always have 25 controlled variable, therefore, DST (i, j) will for 0≤DST (i, j)≤500.When the step 387 (seeing Figure 20) of the program 38 of extracting at optimal set generates optimal set data BPD, each normalized parameter value is converted to original value again.(for example, under the situation of fully laden estimation coefficient Ca, being converted to 20 * 50,000 ÷ 100=10 again, 000).
In step 415, (i, set j) is to Pd1, Pd2 select to form shortest distance DST.Then, judge set to Pd1 in step 416, whether the feature of Pd2 is identical.That is (Pd1 Pd2) with judgment value DSTa, judges homogeneity according to comparative result, to compare distance D ST.This judgment value DSTa for example is made as 25.
If DST (Pd1, Pd2)≤DSTa, process enters step 417, respectively from two set Pd1, the team control performance data PRE (Pd1) of Pd2, PRE (Pd2) take out average latency AWT (Pd1), and AWT (Pd2) is set to the performance estimation value VPD1, the VPD2 that delete then.
In step 418, compare performance estimated valve VPD1 and VPD2, with decision set to be deleted.If VPD1＜VPD2 determines that deletion set number is the set of Pd2, process enters step 419.Then, the record of deletion set EPS (Pd2) and team control performance data PRE (Pd2).Then, the value of several P of record cell is subtracted 1.The set number is redistributed into remaining set, and process finishes in step 419.
If at step 418 VPD1 〉=VPD2, judge that then one of set Pd1 is deleted, process enters step 420.Then, the record of deletion set EPS (Pd1) and team control performance data PRE (Pd1).Then, the value of several P of record cell is subtracted 1.The set that stays renumbers, and process finishes in step 420.
In step 416, if DST is (Pd1, Pd2)＞DSTa, the feature of then judging them is inequality, process enters step 421 then, in this step, finds out with respect to all unit number 1 to P and to be used to delete the performance estimation value VPD (1) of judgement to VPD (P), therefrom regulation has the set of worst-case value, then its set number is made as Pd1.This step 421 is similar to the step 352 of delete program 35 (seeing Figure 17) among first embodiment to 357, and the Therefore, omited is to its description.Deletion set number RP among Figure 17 is equivalent to gather number Pd1.
In step 422, the set after the judgement deletion is counted P and whether is become deletion end judgment value Pe or littler than Pe.As not, process repeating step 414 to 422, and become when being equal to or less than Pe the operation of end delete program 35 as P.
As mentioned above, in the 3rd embodiment, a pair set that has same feature according to the distance D ST regulation between the set, and delete a set in this pair set, like this, characteristic different a plurality of set mutually can be retained in the memory device 11, guarantees the variation of memory device 11 thus.When selecting a pair of set, the pair set with best similarity (shortest distance) has preoption, so the mutual different a plurality of set of characteristic can keep.
In the 3rd embodiment, when set of centering is gathered in deletion, keep the set of better performance estimation value, and the set of deletion poorer performance estimated valve, like this, the team control performance that is stored in a plurality of set in the memory device 11 all can maintain on the high level.Should be noted that in the 3rd embodiment,, will leave out the set of the poorest performance estimation value successively when not existing when leaving out an identical pair set.Therefore, unnecessary set will continue deletion, and finishing judgment value until deletion is Pe.[the 4th embodiment (searching another embodiment that finishes to judge)]
Referring to Figure 19, below another embodiment that finishes judging unit is searched in description.In the 4th embodiment, the main description part different with first embodiment.
In the step 371 of Figure 19, use addition record times N R to replace and estimate times N E, search end judgment value NEa with searching end judgment value NRb replacement.Particularly, addition record times N R for example is made as 200.
That is,, continue search procedure,, finish to search if NR 〉=NRb then will search permission flag FLAG in step 373 and be changed to [0] if NR＜NRb then will search permission flag FLAG in step 372 and be changed to [1].
Usually, become a higher value, and when searching when restraining to a certain extent, the addition record number of times is tending towards reducing with the ratio of searching number of times when searching times N E.Therefore, to search in order finishing effectively, must to consider the result who searches operation, promptly the set number of addition record is judged the end of searching.Therefore, in the 4th embodiment, search procedure proceeds to addition record times N R (expression addition record number) always and has reached to search and finish judgment value NRb.
As mentioned above, according to the 4th embodiment, system can avoid stopping its search procedure before effectively finishing search procedure.[the 5th embodiment (for searching another embodiment that finishes to judge)]
Referring to Figure 27, below another embodiment that finishes judging unit is searched in description.In the 5th embodiment, will the part different with first embodiment be described mainly.
Figure 27 represents that searching of the 5th embodiment finish determining program 37, and it is corresponding to Figure 19 of first embodiment.
In step 431, calculate the estimation times N EH that lives through according to formula NEH=NE+NEX.The estimation times N EH that lives through in the end finishes to judge new estimative figure afterwards.Estimation times N EX when in the end finishing to judge is illustrated in the estimation times N E that determines in the last circulation when finishing.After step 432 is judged at the eleventh hour, estimate whether number of times is equal to or greater than certain number of times.If the estimation times N EH that lives through is worth NEb less than certain, then will search permission flag FLAG and be changed to " 1 " in step 433, continue to search.The estimation times N EH that especially, for example will live through is changed to 20.
If step 432 is judged the estimation times N EH that lives through and is equal to or greater than certain value NEb that then process enters step 434, and calculates a successful index RSC.At first, calculate new record times N RH according to formula NRH=NR-NRX.This new record times N RH is illustrated in and finishes the number of times that the new set of addition record is afterwards judged in last end.According to formula RSC=NRH ÷ NEH, calculate successful index then.In the end the record times N RX of Pan Duaning represents to have finished the value of the addition record times N R that is calculated when finishing to judge in circulating the last time.
At next step 435, be updated in last estimation times N EX and record times N RX when judging according at that time estimation times N E and addition record times N R.In step 436,, judge whether to finish to search according to estimating times N E, search terminal stage judgment value NEC, successful index RSC and searching end judgment value RSCa.Wherein, search terminal stage judgment value NEC and for example place 600.Search end judgment value RSCa and for example place 0.05.
If NE＜NEC or RSC 〉=RSCa then judge and search still not yet in effect finishing, search permission flag FLAG and place [1], continue to search in step 433.If NE 〉=NEC and RSC＜RSCa then judge to search and effectively finish, and will search permission flag FLAG in step 437 and place [0], and stop to search.
It should be noted that, step 436 comprises the relevant reason of estimating the condition of times N E, be to prevent that search procedure from not having enough estimation number of times just to finish because of having lowered successful index RSC, in the starting stage of searching, this situation may be due to scale GP-SO, crossing-over rate CR and the aberration rate MR of initial sets.If do not produce this class problem, the condition of relevant estimation times N E will be no longer essential by the Rule of judgment of searching end, thus, even use the condition relevant with successful index RSC just enough.
In the 5th embodiment, search end as mentioned above, can high precision judge search whether fully restrain thus according to the successful index RSC judgement that obtains by estimation number of times and addition record number of times.Therefore, need not to repeat without meaning to search, but also find to efficiency optimal set.
In addition, in the 5th embodiment, detect the starting stage of searching by estimating times N E, during this starting stage, even becoming each, successful index RSC finishes judgment value RSCa less than searching, search procedure does not finish yet, like this, do not have enough estimation number of times, search procedure will can not finish.[the 6th embodiment (searching another embodiment that finishes to judge)]
Referring to Figure 28, below with description search the knot judging unit 17 another embodiment.In the 6th embodiment, the main description part different with first embodiment.
Figure 28 represents that searching of the 6th embodiment finish determining program 37, and it is corresponding to Figure 19 of first embodiment.
Distance D ST between step 451 set of computations (i, j).(i j) calculates by above-mentioned formula [31] distance D ST.Notice that the step 414 of delete program 35 much at one among this calculating and the 3rd embodiment.
In step 452, according to the distance D ST of aforementioned calculation (i j) calculates similar set and counts NDST, promptly its DST (i, j)≤the set number of DSTa.DSTa is a judgment value of judging that set is whether identical, and in the 6th embodiment, it is the same with the 3rd embodiment to be set to 25.At next step 453, calculate to search and finish judgment value NDSTa.In a word, when searching in convergence, many set have the optimal set trend on every side that is converged in.Identity set is counted NDST and is detected a kind of like this trend as index, and the convergence of searching depends on that identity set counts NDST and the whole set group ratio of closing number.If judgment threshold this ratio relatively places 80%, then calculate to search and finish judgment value NDSTa according to NDSTa={PX (P-1) ÷ 2} * 0.8, because when the set number that has write down is P, whole number of combinations becomes { Px (P-1) ÷ 2}.
At this moment, if the set number of addition record reaches maxim Pmax, although will delete the set that need not under a certain reference point, identical set is counted NDST and also will be changed according to how deleting the set that need not.Equally, identical set is counted NDST and can be changed according to the method that generates set or addition record.Therefore, judgment threshold is not limited in 80%, must carry out suitable modification according to other factors.
In step 454, count NDST and search according to identity set and finish judgment value NDSTa and judge to search whether finish.If NDST＜NDSTa then judges not search yet and fully finishes; To search permission flag FLAG in step 455 and be changed to [1], and continue to search; And the operation that stops searching end determining program 37.If NDST 〉=NDSTa then judges to search and fully carries out; To search permission flag FLAG in step 456 and be changed to [0], and stop to search; And the operation that stops searching end determining program 37.
As mentioned above, in the 6th embodiment, judge according to the distance D ST between the set and to search end, like this, the convergence of can high Precision Detection searching.Therefore, need not to repeat without meaning to search, can carry out search procedure effectively thus.Search the end Rule of judgment and be not subjected to above-mentioned those conditional convergences, can adopt other condition.[the 7th embodiment (another embodiment that optimal set is extracted)]
Referring to Figure 29, another embodiment of extractor 20 will be described below.In the 7th embodiment, the main description part different with first embodiment.
Figure 29 represents the optimal set of the 7th the embodiment program 38 of extracting, and it is corresponding to Figure 20 of first embodiment.
In step 471, take out average latency AWT (1) respectively from team control performance data PRE (1) to PRE (P) to AWT (P), be set to the first performance estimated valve VPS1 (1) then to VPS1 (P).In step 472, taking-up waits as long for time RLW (1) to RLW (P) from team control performance data PRE (1) to PRE (P) respectively, is set to the second performance estimated valve VPS2 (1) then to VPS2 (P).In step 473, find out the minimum value of the first performance estimated valve VPS1 (1) to VPS1 (P), they are set to best values BVPE.
In step 474, select optimal set BP according to the performance estimation value.(VPS1 (i)-BVPE) is equal to or less than a plurality of set i of BZ promptly to find its numerical value from memory device 11.Then, will wherein have minimum second the set and be chosen as optimal set BP with reference to estimated valve VPS2 (i).Promptly finish the selection of two steps.BZ is a reference value, and expression departs from the tolerance band of best values BVPE, is set to two seconds in the present embodiment.
Mode with similar first embodiment generates optimal set data BPD then, and at next step 388, BPD delivers to group control device 1 with the optimal set data.
As mentioned above, in the 7th embodiment, adopted for two steps selected to extract optimal set.Estimating with two under the situation of judging some relative importance value, because two steps were selected to be applied to extract, so can be according to this relative importance value optimal set of extracting.Can certainly adopt three steps or multistep to select.The 7th embodiment is in the first-class content that is same as above-mentioned the 6th performance estimation function [30] of meaning.[the 8th embodiment (the another kind of method of calculating of team control performance valve)]
As 57-57, No. 168 day disclosure special permission communiques (KOKAI) are described, with the simulator 2 of actual group control device 1 first embodiment of replacement, can obtain the team control performance of new set.Below be described referring to Figure 30 and 31.
Figure 30 represents the system of the 8th embodiment, and it is corresponding to Fig. 1 of first embodiment.Figure 31 represents the operation of group control device 1, and it is corresponding to Fig. 9 of first embodiment.The 8th embodiment below will be described, the wherein main description part different with first embodiment.
Referring to Figure 30, when basis is searched the search criterion signal 1a command lookup of device 10 by group control device 1 input, search device 10 and promptly carry out estimation routine (seeing Figure 15) in the step 33 of operating sequence, generation is similar to the simulated conditions data of first embodiment and exports simulated conditions signal 13a.
Yet, as shown in figure 30, in the present embodiment signal 13a is fed to group control device 1.Group control device 1 enters the trail run mode after receiving this signal 13a.Below will utilize diagram of circuit shown in Figure 31 to describe this operation in detail.
Among Figure 31, step 491 judges whether it is trail run, and judges whether to begin trail run in step 492.When trail run sign FLG is " 0 " and when signal 13a does not comprise the appointment that begin trail run, move according to step 221 to the conventional team control of 229 execution.
On the other hand, when from the content of signal 13a, detecting beginning trail run mode in step 492, in step 493 trail run sign FLG is changed to [1], and by parameter value the interim escape of current use is set in step 494.Write the set (new set) that is included in the estimation among the signal 13a and replace these escapes.
Then, in service in conventional team control, carry out team control in step 221 to 229 and move.Between trial run period, by means of trail run sign FLG be step 491 after [1] to 495, carry out the team controls operation with step 221 to 229.
After (for example one hour) has been carried out a period of time in the team control operation, group control device 1 detects the end of trail run in step 495, make trail run sign FLG be reset to [0] in step 496, send the parameter sets of escape back in step 497, simultaneously, calculate the team control performance data PRF (such as average latency, wait as long for time) relevant with trail run.In step 498, team control performance data PRF is fed to searches device 10, as team control performance valve signal 2a.Then, group control device 1 returns normal condition, and finishes team control by step 221 to 229 and move.
As mentioned above, when obtaining team control performance valve signal 2a by trail run, shown in Figure 30 searches device 10 according to team control performance valve signal 2a obtained performance estimated valve VPN, and compares performance estimated valve VPN and estimate reference value BX, to judge whether wanting addition record newly to gather.
As mentioned above, in the 8th embodiment, estimate new set in actual device, like this, be tending towards becoming longer owing to obtain the time of optimal set, although not really suitable, simulator 2 is optional, and system cost is reduced.[the 9th embodiment (generating another embodiment of new set)]
Although in first embodiment, fixed crossing-over rate CR and aberration rate MR, the 9th embodiment is characterised in that and can revises crossing-over rate CR and aberration rate MR according to searching situation.
Referring to Figure 32 to 34, the 9th embodiment will be described below.Mainly will describe and first or second part that embodiment is different.
Figure 32 represents the integral structure of the 9th embodiment, and it is corresponding to Fig. 1 of first embodiment.Among Figure 32, occurrence rate modifier 4 is revised crossing-over rate CR and aberration rate MR (selection rates of various generating modes) according to searching situation.
Figure 33 represents the operating sequence of the 9th embodiment, and it is corresponding to Figure 11 of first embodiment.Notice that except adding the step 50 corresponding to the occurrence rate update routine of the function of occurrence rate shown in Figure 32 (emergence rate) modifier 4, the operating sequence 100 with shown in Figure 11 is identical basically for it.
Figure 34 represents the occurrence rate update routine.In step 501, calculate the estimation times N EN that lives through according to formula NEH=NE-NEX, it is illustrated in last end and judges the new afterwards estimation number of times of carrying out.The estimation times N EX that the identical value of estimation times N E when the last end of carrying out in the former circulation is judged is set in the end judge.Step 502 judges to have estimated whether to have carried out certain number of times or more times number since the last circulation.If the estimation times N EH that lives through then finishes the occurrence rate update routine less than a certain value NEb (for example 20 times).
On the other hand, if judge that in step 502 the estimation times N EH of experience is equal to or greater than certain value NEb, program enters step 503, calculates successful index RSC in this step.
At first, according to formula NRH=NR-NRX calculate from last finish to judge begin the number of times that addition record represents newly to gather the record NRH of number of times.Calculate successful index RSC according to formula RSC=NRH ÷ NEN.The same number NR of the addition record when former circulation finishes to judge at last is set in the end judge the times N RX of record.At next step 504, the record times N RX that is updated in the estimation times N EX of last judgement and in the end judges according at that time estimation times N E and addition record times N R.
To 510, revise crossing-over rate CR and aberration rate MR in step 505 according to estimation times N E, the first judgment value NEd1, the second judgment value NEd2, successful index RSC, success ratio judgment value RSCb, success ratio judgment value RSCc.For example, NEd1=500; NEd2=800; RSCb=0.10; RSCc=0.05.
If NE≤NEd1 and RSC≤RSCb, if or be in the state lower searching first half success index than expectation value, the crossing-over rate CR and the aberration rate MR that then judge Set For Current are improper, process enters step 505,507 and 509, reduce crossing-over rate CR (for example reducing 0.001) slightly from currency thus, and increase aberration rate MR (for example increasing by 0.001) slightly from currency.In addition,, adopt the mode opposite, be about to crossing-over rate CR and be provided with, be provided with aberration rate MR to such an extent that be slightly less than currency (for example 0.001) less times greater than currency (for example 0.001) with above-mentioned modification in step 509.
In fact also can remove a kind of so successful index of reduction with another kind of method.That is, if crossing-over rate can not use too greatly, then make crossing-over rate CR less, and make aberration rate MR bigger.Otherwise, can not use if crossing-over rate is too little, then make crossing-over rate CR bigger, and make aberration rate MR less.
In a word,, then change the ratio of the selection probability of every kind of generation method, to remove a kind of like this reduction state if search the state that is in reduction.
If NE 〉=NEd2 and RSC≤RSCC, if or searching after-stage, the success index is in the state lower than expectation value, judge that then good set search trends towards convergence, process enters step 505,506,508 and 510 successively, make crossing-over rate CR that increase (for example increasing by 0.001) is arranged on the basis of currency slightly thus, and make aberration rate MR on the basis of currency, slightly reduce (for example reducing 0.001).
If above-mentioned condition does not satisfy, judge that then the crossing-over rate CR of Set For Current and aberration rate MR are all suitable, and stop the occurrence rate update routine, no longer these ratios are made amendment.
As mentioned above, in the 9th embodiment,, and can judge according to this and revise crossing-over rate CR and aberration rate MR according to the progress of estimating that times N E and successful index RSC judgement is searched.Therefore, can than have fixedly crossing-over rate CR and fixedly the system of aberration rate MR earlier find good set, and can shorten the time of searching.As a result, system can improve search efficiency.
Especially, in the 9th embodiment, if at the starting stage of searching (or first half), the success index becomes and is lower than expectation value, then process is arranged to the value lower than currency with crossing-over rate CR, and aberration rate MR is arranged to the value more higher than currency, like this, carry out in search procedure widely when variation and to add temporary, system can improve the probability of the set of the more excellent team control performance of generation.System also can remove the lowered state of searching.
In addition, in the 9th embodiment, if searching after-stage (or latter half), the success index becomes and is lower than expectation value, then process is arranged to the value more higher than currency with crossing-over rate CR, and aberration rate MR is arranged to the value lower than its currency, like this, when local search procedure is added temporary, system can make search procedure restrain as early as possible.System also can remove the lowered state of searching.[the tenth embodiment (another embodiment that occurrence rate is revised)]
Referring to Figure 35, another embodiment of occurrence rate modifier 4 is described below.Figure 35 represents an occurrence rate update routine, and it is that occurrence rate update routine (seeing Figure 34) by the 9th embodiment is through the amended program of part.
Among Figure 35, step 502 judges whether to have reached certain and estimates times N Eb (for example 40 times).Step 506 judges whether to be at last and searches the stage (or NE 〉=NEd2) not.At last, in step 510, be provided with crossing-over rate CR than currency slightly larger (for example big slightly 0.001), though be provided with aberration rate MR than currency slightly smaller (for example little 0.001).Therefore, in the final stage of searching, along with the propelling of search procedure, it is bigger that crossing-over rate CR becomes gradually, otherwise it is less that aberration rate MR then becomes gradually.This part, other operation is identical with the 9th embodiment except above-mentioned.
As mentioned above, in the tenth embodiment, because propelling along with search procedure, system changes generation into from generation weighting variation and mainly is fixed against intersection, so at the starting stage of searching (first half), system can keep the team control changes of properties, and in the end the stage (latter half) can make search procedure restrain as early as possible.As a result, can find set effectively with more excellent team control performance.[the 11 embodiment (intersection another embodiment to selecting)] referring to Figure 36 to 39, below will describe the 11 embodiment, and itself and first or second part that embodiment is different mainly are described.
Figure 36 represents the system of the 11 embodiment.In this embodiment, utilize distance between the set as selecting the condition of set to (intersect to) to intersecting purpose.In addition, in the present embodiment, system constructing is become can be according to the distance modification condition (hereinafter referred to as " distance condition ") between the set.A superclass alternative condition modifier 5 shown in Figure 36 is in order to revise distance condition according to searching situation.
Figure 37 represents the operating sequence of the 11 embodiment, and it is corresponding to Figure 11 of first embodiment.Among Figure 37, step 31 generates new set.Although the operation that should generate is as described in Figure 14 of first embodiment, present embodiment is selecting intersection still to be different from first embodiment aspect the process of PS1, PS2 (step 317).To add step 52 corresponding to the superclass alternative condition update routine of superclass alternative condition modifier 5 (seeing Figure 36), can revise above-mentioned distance condition thus.Other structure of present embodiment uses the mode identical with first embodiment to constitute.
Figure 38 is illustrated in the operation of the step 317 that comprises in the new set generator program 31 (seeing Figure 14).In the step 317a of Figure 38, be [0] with the value initialization of counting machine RC.In the present embodiment, counting machine RC utilizes the right selection time counting number of distance condition pair set between set.According to distance between formula [31] set of computations.
At next step 317b, system to selecting probability weight, selects to form two a pair of set PS1, PS2 according to the size of performance estimation value at random.Next step 317C judges the set that utilizes distance condition selects whether to have finished certain number of times or more times number (for example 10 times).
If by the certain number of times of repeating step 317d to 317h or more times number, still can not select to satisfy two set PS1 of distance condition, PS2, two set PS1 that then step 317b selected, the set that PS2 is defined as intersecting is right.
On the other hand, the number of times that the set of using when distance condition is selected is during less than certain number of times (RC＜10), and process enters step 317c, enters 317d then, and the value of counting machine RC is added 1.At step 317e, calculate the distance D ST between two set PS1, the PS2 that step 317e selects.Promptly such with formula [31], calculate
DST＝||EST(PS1)－EPS(PS2)||。As mentioned above, each parameter value is normalized to a numerical value between 0 and 100.At step 317e,, then whether satisfy as the distance condition that intersects to one of alternative condition at step 317f to 317h judging distance DST if calculated distance D ST.
If condition selection marker SELS is changed to [1], and if first alternative condition is appointed as distance condition, process enters step 317f, and whether 317g is equal to or greater than first at this step judging distance DST and selects reference value DSTb1.If DST＜DSTb1 then abandons satisfying two set PS1, PS2 of this distance condition, process is returned step 317b once more, and the repetition identical process starts anew.
On the other hand, if condition selection marker SELS is changed to [2], and if second alternative condition is appointed as distance condition, then as the above, whether be equal to or less than second at step 317h judging distance DST and select reference value DSTb2.If DST＞DSTb2 judges that then alternative condition satisfies, process enters step 317b and starts from scratch and restart.
The operation of step 317b to 317h repeatedly reaches certain number of times (10 times) until distance condition, or until finding two set satisfying distance condition.If generate two set of satisfying DST 〉=DSTb1 or DST≤DSTb2, then make these two set become rule and intersect PS1, PS2, process enters step 318.Those steps after the step 318 are identical with the step among first embodiment.
As implied above, system can utilize distance condition to select intersection right.This distance condition can be revised by alternative condition update routine 52 according to the situation of searching.
Figure 39 represents the detailed content of step 52 (seeing Figure 37) alternative condition update routine.
To 524, the step 501 in the 9th embodiment in the occurrence rate update routine 50 (seeing Figure 34) is moved successful index RSC to 504 operation in step 521.Usually, should be defined as the addition record number of times divided by estimating number of times by the success index.
To 536, with estimating times N E, the first judgment value NEd1, the second judgment value NEd2, successful index RSC and success ratio judgment value RSCd, RSCe, RSCf revise and select reference value DSTb1, DSTb2 in step 525.For example, NEd1=500; NEd2=800; RSCd=0.10; RSCe=0.05; RSCf=0.05.For example, at the step 283 of initialize routine 28 (seeing Figure 13), SELS=1; DSTb1=250; DSTb2=250.
Step 525 is judged NE≤NEd1 not, promptly searches still to be in (first half), in step 527, SELS is changed to [1], and first alternative condition is appointed as intersection to alternative condition.Then, in step 529, more successful index RSC and success ratio judgment value RSCd.If successful index RSC less than success ratio judgment value RSCd (RSC≤RSCd), then judge since Set For Current first select the value of reference value DSTb1 too limited, so the set that meets this condition is very little, stop successful index RSC higher thus.Therefore, in step 531, select reference value DSTb1 to be provided with slightly smallerly (for example little 5%) with first.
On the other hand, if NE 〉=NEd2 perhaps, that is searches and is in [final stage], then process enters step 525,526,528, SELS is changed to [2]; And second alternative condition is appointed as intersection to alternative condition.Then in step 529, more successful index RSC and success ratio judgment value RSCf.(RSC≤RSCf), then judge because the value of the second selection reference value DSTb2 of Set For Current is too limited so qualified set very little, stops successful index RSC higher thus if successful index RSC is less than success ratio judgment value RSCf.Therefore, in step 532, select the value of reference value DSTb2 to be provided with slightly largerly (for example big 5%) with second.
If the intersection that successful index RSC, then judges current selection greater than success ratio judgment value RSCf will improve without doubt to the selection reference value of alternative condition, finish the operation of alternative condition update routine 52 on currency.
In step 525,526, as judgement NEd1＜NE＜NEd2, when being [midway], then at more successful index RSC of step 533 and success ratio judgment value RSCe.(RSC≤RSCe), think that then current alternative condition is improper is so must change alternative condition if successful index RSC is less than success ratio judgment value RSCe.Therefore, after step 534 judges whether current alternative condition is first or second alternative condition, promptly alternative condition is changed to second alternative condition (SELS=2), or change to first alternative condition (SELS=1) in step 536 in step 535.
If successful index RSC greater than success ratio judgment value RSCe, then judges the alternative condition no problem of current selection, finish the operation of alternative condition update routine 52.
As mentioned above, in the 11 embodiment, owing to will represent between two set between the set of similarity distance as intersecting right selection reference, so can gather with generating newly with the search procedure of part widely.
Promptly, if select that pair set of reference value DSTb1 to obtain the selection relative importance value apart from being equal to or greater than first between its set, if and make intersection and have two sets match of different characteristic as far as possible mutually, this process then can improve the probability that generates set, although may suffer setbacks or lose performance and convergence is relatively poor with better team control performance.On the contrary, if select that pair set of reference value DSTb2 to obtain the selection relative importance value apart from being equal to or less than second between its set, if and make intersection and have two sets match of same characteristic features as far as possible mutually, although reduced the probability that generates set, can improve the probability that generates new set with good team control performance with good team control performance.
In the 11 embodiment, system is especially according to starting stage or the final stage of estimating that the number of times judgement is searched, in the starting stage of searching, for being equal to or greater than first, distance between its set select that pair set of reference value DSTb1 that the selection relative importance value is provided, to search procedure weighting widely.In the final stage of searching, for being equal to or less than second, distance between its set select that pair set of reference value DSTb2 that the selection relative importance value is provided, and to searching the convergence weighting, system can improve its search efficiency for this reason.
Especially, in the 11 embodiment, utilize to intersect alternative condition is searched, wherein, select that pair set of reference value DSTb1 that the selection relative importance value is provided for distance between its set is equal to or greater than first, selecting reference value DSTb1 to place than its currency with first is little value, so the starting stage of searching, if successful index RSC becomes less than expectation value, then crossing condition will become milder.As a result,, search the low state that is in, at this moment the first suitable selection reference value DSTb1 will be set automatically, to remove low state because the first selection reference value DSTb1 that is distributed is improper.
Especially, in the 11 embodiment, utilize to intersect alternative condition is searched, wherein, select that pair set of reference value DSTb2 that relative importance value is provided for distance between its set is equal to or less than second, selecting reference value DSTb2 to be arranged to than its currency with second is big value, so searching final stage, if successful index RSC becomes less than expectation value, then crossing condition will become milder.As a result,, search the low state that is in, at this moment second a suitable selection reference value DSTb2 will be set automatically, to remove low state because the second selection reference value DSTb2 that is distributed is improper.
Especially, in the 11 embodiment, be to intersect first alternative condition is searched midway during searching, if successful index RSC becomes less than expectation value, then this condition changes to second and intersects to alternative condition, does not make its suitable present case so work as this alternative condition, and makes and search when entering low state, system can be converted to suitable alternative condition automatically with it.
Moreover, especially in the 11 embodiment, intersect when searching under the alternative condition second when searching final stage, if successful index RSC becomes less than expectation value, then this condition changes to first intersection to alternative condition, so when this alternative condition does not make its suitable present case, make and search the low state that enters, system can be converted into suitable alternative condition automatically.[the 12 embodiment (intersection another embodiment)] to selecting
Referring to Figure 40, another embodiment of alternative condition modifier 5 is described.Figure 40 represents an alternative condition update routine, and it is that a kind of part of the alternative condition update routine 52 (seeing Figure 39) of the 11 embodiment has been made the form of revising.
In Figure 40, step 522 judges whether the estimation number of times is a fixed number NEb (for example 50 times).Step 526 judges whether it is second period (after-stage), if it is not (NE＜NEd2), then in step 527 first condition is appointed as intersection to alternative condition in second period; In step 531, select reference value DSTb1 to be provided with smallerly slightly (for example, little by 2%) with first.Otherwise, if judge that in step 526 it is (NE 〉=NE2d), then in step 528 second condition is appointed as intersection to alternative condition in second period; In step 532, select reference value DSTb2 to be provided with slightly smallerly (for example little 2%) with second.
As mentioned above, in the 12 embodiment,, can change intersection to alternative condition according to the progress of searching utilizing first to select intersection that reference value DSTb1 uses to alternative condition period.That is, select the value of reference value DSTb1 to be provided with first of this starting stage in period greater than the value during this after-stage in period, like this, in the starting stage in this period, to the weighting of team control changes of properties, and at the after-stage in this period, then to searching the convergence weighting.
Equally, in the 12 embodiment,, intersect to alternative condition according to searching the progress conversion utilizing second to select set that reference value DSTb2 uses to alternative condition period.That is, select reference value DSTb2 to be provided with second of this starting stage in period less than the value of after-stage in this in period, like this, in the starting stage in this period, to the weighting of team control changes of properties, and at the after-stage in this period, to searching the convergence weighting.Therefore, system can improve search efficiency by the conversion alternative condition.[the 13 embodiment (another embodiment that cross parameter is selected)]
In first embodiment, wherein the parameter value cross parameter (parameter position) of being replaced is selected at random to two superclass.On the contrary, the maker among the 13 embodiment, its characteristics are poor (parameter error) with parameter value as the parameter alternative condition, and revise wherein parameter alternative condition according to the situation of searching.
Referring to Figure 41 and 42 maker is described.In this 13 embodiment, the main description and the 11 part that embodiment is different.
Figure 41 represents the content of the middle step 318 of new set generator program 31 (seeing Figure 14) of the 13 embodiment.Among Figure 41,, be " 0 " with the value initialization of counting machine RC at step 318a.In the present embodiment, counting machine RC determines the parameter error condition as one of cross parameter alternative condition in order to calculation times.Step 318b produces a random number in 0 to 25 scope, to determine relevant therewith parameter PX.This is identical with the described operation of first embodiment.
Step 318c compares number of times and a certain number of times of determining the parameter error condition.Even when when the certain number of times of repeating step 318d to 318h or more times number (for example 10 times), still failing to find to satisfy the cross parameter PX of parameter error condition, then judge and considered that the selection of parameter error fully finishes, process finishes in step 318, with the parameter PX that determines to select at step 318b as cross parameter.
On the other hand, at step 318c, if the number of times of determining the parameter error condition less than certain number of times (RC＜10), process enters step 318c, enters step 318d then, counting machine RC adds 1.At next step 318e, calculate selected two the set PS1, PS2 PX numerical value poor [| EPS (PS1)＜PX＞-EPS (PS2)＜PX＞|], to find distance D STP.Each Parameters Transformation is become numerical value between 0 and 100 and normalization method.In the extract step 387 (seeing Figure 20) of program 38 of optimal set, when generating optimal set data BPD, in group control device 1, convert each parameter to useful numerical value again.
As mentioned above, the difference DSTP in that step 318e calculates PX the parameter value of two set PS1, PS2 has judged whether to satisfy by selecting the deviation condition of appointment at step 318f to 318h then.
When first alternative condition was appointed as parameter error condition (SELS=1), process entered step 318f, enters step 318g then, judged whether deviation D STP is equal to or greater than first and selects reference value DSTc1 this moment.If DSTP＜DSTc1 then abandons not satisfying the cross parameter PX of parameter error condition, process is returned step 318b, starts from scratch to repeat identical operation.
Equally, when second alternative condition is appointed as parameter error condition (SELS=2), judge at step 318h whether deviation D STP is equal to or less than second and selects reference value DSTc2.(DSTP＞DSTc2), process enters step 318b, starts anew again if alternative condition does not satisfy.
The operation of repeating step 318b to 318h, until the certain number of times of definite parameter error condition (10 times) or more times number, or until finding the cross parameter that satisfies the parameter error condition.If detect the cross parameter PX that meets DSTP 〉=DSTc1 or DSTP≤DSTc2, judge that then parameter PX is regular cross parameter PX, process enters next step 319.Because identical among the later step of step 319 and first embodiment is so this part is no longer described.
Figure 42 represents to revise a kind of method of above-mentioned alternative condition according to searching progress.
Figure 42 is illustrated in an alternative condition update routine of operating sequence 100 the 52nd step (seeing Figure 37).
Except step 531,532, the alternative condition update routine of the 13 embodiment is identical with the alternative condition update routine (Figure 39) of the 11 embodiment.Therefore, mainly will describe wherein, suppose step 283 (seeing Figure 13), specific data SELS=1 in initialize routine 28 to selecting the operation to 532 of step 529 that reference value DSTc1, DSTc2 proofread and correct; Select reference value DSTc1=50; Select reference value DSTc2=50.
Searching the starting stage (NE≤NEd1), if successful index RSC is less than expectation value RSCd (RSC≤RSCd), selected first alternative condition (SELS=1) simultaneously, judge that then the cross parameter value satisfy condition is too little, and stop successful index RSC to select reference value DSTc1 to limit to very much to become bigger with improper because of first of Set For Current.Then, process enters step 529, enters 531 again, at this moment, is provided with smallerly slightly the first selection reference value DSTc1 (for example little 5%), finishes the operation of alternative condition update routine 52.
On the other hand, at the final stage of searching (NE 〉=NEd2), if successful index RSC is less than expectation value RSCf (RSC≤RSCf), selected second alternative condition (SELS=2) simultaneously, judge that then the cross parameter meet this condition is too little, and stop successful index RSC to select reference value DSTc2 to limit to very much and improperly become bigger because of second of current setting.Then, process enters step 530 and enters 532 then, at this moment, selects reference value DSTc2 to be provided with slightly largerly (for example big 5%) with second.
If successful index RSC, judges then that the selection reference value of the alternative condition of current selection increases without doubt greater than expectation value RSCd on currency, and the operation of end alternative condition update routine 52.
As mentioned above, in the 13 embodiment, owing to obtained to represent the parameter error of the similarity of two selected set, and, considered team control changes of properties and the convergence of searching so can make to generate owing to be provided with the parameter alternative condition according to this parameter error thereafter.
Promptly, if being equal to or greater than first, its parameter error select the controlled variable of reference value DSTc1 in selection, to give relative importance value, if and intersection can be matched with its characteristic different parameter as far as possible mutually, although then select to suffer trouble or loss performance and convergence relatively poor, still can improve the probability that generates new set with good team control performance.If being equal to or less than second, its parameter error select the controlled variable of reference value DSTc2 to obtain the selection relative importance value, if and intersection can be matched with its characteristic identical parameter as far as possible mutually, although then may reduce the generating probability of new set, but still can improve the generating probability of new set with good relatively team control performance with good team control performance.
In the 13 embodiment, according to starting stage and the terminal stage of estimating that the number of times judgement is searched; In the starting stage of searching, carry out search procedure to the weighting of team control performance variation, wherein, its parameter error is equal to or greater than first and selects the controlled variable of reference field value DSTc1 to obtain the selection relative importance value; Otherwise, in the final stage of searching, carry out searching the search procedure of convergence weighting, wherein, its parameter error is equal to or less than second and selects the controlled variable of reference value DSTc2 to obtain the selection relative importance value.Therefore, can be according to carrying out the generation of considering the team control performance variation and searching convergence the period of searching.
In the 13 embodiment, when the cross parameter alternative condition of using the first selection reference value DSTc1 use is searched, selecting first reference value DSTb1 to be arranged to than its currency is little value, like this, if successful index RSC becomes littler than expectation value in the starting stage of searching, then alternative condition will become milder.As a result, owing to the improper low state that is in of searching that makes of the first selection reference value DSTb1 that is distributed, its numerical value is modified to suitable value automatically.In the 13 embodiment, when utilizing the cross parameter alternative condition of using the second selection reference value DSTc2 to search, selecting reference value DSTb2 to be arranged to than its currency with second is little value, therefore, the final stage of searching, if successful index RSC becomes less than expectation value, then alternative condition will become milder.As a result, second selected improper the making of reference value DSTb2 to search when being in low state owing to what distribute, its numerical value is adapted to desired value automatically.
In the 13 embodiment, when the first cross parameter alternative condition of searching the first selection reference value DSTc1 of utilization application is midway searched, if successful index RSC becomes less than expectation value, then it is converted to the second cross parameter alternative condition of using the second selection reference value DSTc2.As a result, make and search when being in low state being not suitable for present case because of the first cross parameter alternative condition, this condition can be transformed into suitable alternative condition automatically.
In the 13 embodiment, when the second cross parameter alternative condition of searching the second selection reference value DSTc2 of utilization application is midway searched, be smaller than expectation value if successful index RSC becomes, then convert the first cross parameter alternative condition of using the first selection reference value DSTc1 to.As a result, make and search when being in low state being not suitable for present case because of the second cross parameter alternative condition, this condition can convert suitable alternative condition automatically to.
Like this, can remove searching of low state, and select reference value and improve search efficiency by change thus according to searching situation conversion alternative condition according to the system of the 13 embodiment.[the 14 embodiment (another embodiment that alternative condition is revised)]
Referring to Figure 43, another embodiment of alternative condition modifier 5 will be described below.Figure 43 represents the operating procedure of alternative condition update routine 52, and it is that part has been made the program of revising (seeing Figure 42) among the 13 embodiment.
Among Figure 43, step 522 judges whether the estimation number of times is a fixed number NEb (for example 50 times).Step 526 judges whether to be second period (after-stage), if not (NE＜NE2d), then in step 527 first condition is appointed as the cross parameter alternative condition in second period; In step 531, select reference value DSTb1 to be provided with slightly smallerly (for example little 2%) with first.
Otherwise, if be judged as for second period (NE 〉=NE2d), then second condition is appointed as the cross parameter alternative condition in step 526 in step 528; In step 532, select reference value DSTb2 to be provided with slightly smallerly (for example little 2%) with second.
Like this, according to searching the change of circumstance condition relevant with parameter error.Except the above, other operation is identical with the 13 embodiment's.
As mentioned above, in the 14 embodiment, utilize to use the first cross parameter alternative condition of selecting reference value DSTb1 in period, selecting reference value DSTb1 to be provided with to such an extent that work energetically the value of after-stage in this first of this starting stage in period, for this reason in period, the alternative condition that is provided with makes the starting stage in this period tighter than after-stage, thereby, in the starting stage in this period, to the weighting of team control performance variation, and in the final stage in this period, to searching the convergence weighting.
Equally, in the 14 embodiment, utilizing the cross parameter alternative condition of using the second selection reference value DSTb2 in period, with this in period after-stage second select reference value DSTb2 to be provided with less than the value of this starting stage in period, for this reason, the alternative condition of setting makes the after-stage in this period become tighter than the starting stage.As a result, starting stage in this period to the weighting of team control performance variation, in the final stage in this period then to searching the convergence weighting.
Although in the 9th to the 14 embodiment, be according to estimating that times N E judges whether to be the forebody searched or latter half of, or be not the starting stage or the final stage of searching, but also can be by replace estimating that with addition record times N R times N E judges whether to be the forebody searched or latter half of, or starting stage or final stage not for searching.[the 15 embodiment (another embodiment that parameter is selected)]
Next, will utilize Figure 44 and 45 to describe another embodiment of maker.In the 15 embodiment, judge the selection probability (occurrence rate) of each parameter according to the degree of correlation of the degree of correlation of traffic stream characteristics and team control performance estimation item.The basic structure of the 15 embodiment is identical with second embodiment's, therefore, below will mainly describe wherein and second part that embodiment is different.
Figure 44 represents newly to gather the content of step 318 in the generator program 31 (seeing Figure 14).
Among Figure 44, step 318j to 318q judges that according to the magnitude of traffic flow specification parameter occurrence rate RPA (1) of 25 parameters is to RPA (25).Especially at step 318j to 318l, according to content such as ridership, bottom traffic ratio, uplink traffic ratio, down traffic ratio and be included in similar ratio among the magnitude of traffic flow specification data TRS, distinguish the form of the magnitude of traffic flow.That is, judge peak time section, on the top, go to the bottom and in the middle of the conventional time period, any actually situation is current traffic conditions.
If be judged as the conventional time period, then the occurrence rate RPA1 (1) of each parameter of will be before preparing for the conventional time period at step 318m is set to parameter occurrence rate RPA (1) to RPA (25) to RPA1 (25).Equally, when being judged as peak time, RPA2 (1) is set to parameter occurrence rate RPA (1) to RPA (25) to RPA2 (25) at step 318n; When being judged as the top during time, RPA3 (1) is set to parameter occurrence rate RPA (1) to RPA (25) to RPA3 (25) at step 318P; When being judged as when going to the bottom the time, RPA4 (1) is set to parameter occurrence rate RPA (1) to RPA (25) to RPA4 (25) at step 318q.The row 10B of Figure 45 is expressed as parameter occurrence rate RPA1 (1) that the various magnitudes of traffic flow prepare to RPA1 (25), and RPA2 (1) is to RPA2 (25), RPA 3 (1) to RPA3 (25), and RPA4 (1) is to RPA4 (25).
Among Figure 45, the feature of the magnitude of traffic flow is at the conventional time period, about parameter occurrence rate RPA (1) to RPA (25), for parameter (the parameter numbering=1 to 9 of substantial connection being arranged with the magnitude of traffic flow, 22 to 25), occurrence rate is arranged to " 10 ", and for the magnitude of traffic flow parameter that almost it doesn't matter (parameter numbering=18 to 21), occurrence rate is arranged to " 0 ".For with the magnitude of traffic flow the appropriate controlled variable that concerns (parameter numbering=10 to 17) being arranged, occurrence rate is arranged to " 5 ".
When the feature of the magnitude of traffic flow at peak time during section, related parameter occurrence rate RPA2 (1) is arranged to RPA2 (25), for parameter (the parameter numbering=1 to 9 of substantial connection being arranged with the magnitude of traffic flow, 18 to 21), occurrence rate is arranged to " 10 ", and for the magnitude of traffic flow parameter that almost it doesn't matter (parameter numbering=10 to 17), occurrence rate places " 0 ".For with the magnitude of traffic flow controlled variable (parameter numbering=22 to 25) of moderate relation being arranged, occurrence rate places " 5 ".
When the feature of the magnitude of traffic flow at last top during the time period, about parameter occurrence rate RPA3 (1) to RPA3 (25), for the parameter (parameter numbering=1 to 13) of substantial connection being arranged with the magnitude of traffic flow, occurrence rate places " 10 ", and for the magnitude of traffic flow parameter that almost it doesn't matter (parameter is numbered 14 to 21), occurrence rate places " 0 ".For with the magnitude of traffic flow controlled variable (parameter is numbered 22 to 25) of moderate relation being arranged, occurrence rate places " 5 ".
When the feature of the magnitude of traffic flow when going to the bottom the time period, about parameter occurrence rate RPA4 (1) to RPA4 (25), for parameter (the parameter numbering=1 to 9 of substantial connection being arranged with the magnitude of traffic flow, 14 to 17), occurrence rate places " 10 ", and for the almost unallied parameter of the magnitude of traffic flow (parameter numbering=18 to 21), occurrence rate places " 0 ".For with the magnitude of traffic flow controlled variable (parameter numbering=22 to 25) of moderate relation being arranged, occurrence rate places " 5 ".
As mentioned above, step 3l8j to 3l8q determines that according to the magnitude of traffic flow specification parameter occurrence rate RPA (1) of 25 controlled variable is to RPA (25).Each magnitude of traffic flow RPA1 (1) is to RPA1 (25), and RPA2 (1) is to RPA2 (25), and the value of RPA3 (1) to RPA3 (25) and RPA4 (1) to the occurrence rate of RPA4 (25) is not limited to value shown in Figure 45.If express the degree relevant relatively, then any value might be set with each magnitude of traffic flow feature.The occurrence rate that between each controlled variable, might have little difference.
Next referring to Figure 44, step 3l8r according to compensation value RPAA (1) to RPAA (25) correction parameter occurrence rate RPA (1) to RPA (25), compensation value RPAA (1) is proportional to the degree of correlation (for example average latency) of RPAA (25) and team control performance estimation item.
Compensation value RPAA (1) is provided with by above-mentioned performance reference value setting device 3 to RPAA (25).As described in second embodiment, performance reference value setting device 3 provides [expected value] of average latency as its output, estimate reference value BX [designated value], identical with second embodiment, in this 15 embodiment, except above-mentioned, also provide as that estimate and [degree of correlation] average latency as compensation value.
Therefore, will be included in reference value data TGT in the search criterion signal 1a by group control device 1 and offer and search device 10, the latter comprises wait time expected value TAW, additional reference designated value TCB and compensation value RPAA (1) to RPAA (25).RPAA shown in Figure 45 (1) is to the compensation value of the expected value of RPAA (25) expression average latency.
About compensation value RPAA (1) to RPAA (25), for (parameter numbers=8 with the controlled variable that substantial connection was arranged as the average latency of estimating item, 22,23), compensation value is changed to " 10 ", and for the controlled variable (parameter numbering=10 to 21) that almost it doesn't matter with it, compensation value reset.For the controlled variable that has the appropriateness relation with it (parameter numbering=1 to 7,9,24,25), compensation value is put " 5 ".
On the other hand, if the estimation item is not the average latency but economizes on electricity, about compensation value RPAA (1) to RPAA (25), for (parameter numbers=4 to 7 with the controlled variable that substantial connection was arranged as the average latency of estimating item, 22 to 25), compensation value is put " 10 ", and for the controlled variable (parameter numbering=9 to 21) that almost it doesn't matter with it, compensation value reset.For the controlled variable that the appropriateness relation is arranged with it (parameter numbering=1 to 3,8), compensation value is put " 5 ".
Even estimate that item is not as mentioned above, can determine that compensation value RPAA (1) is to RPAA (25) according to relevant degree equally.If express relatively and a degree of correlation of estimating, then can any value be set to RPAA (25) for compensation value RPAA (1).Might there be little difference in compensation value between each controlled variable.
Next,, produce the random number of its value between [0] and [parameter occurrence rate RPA (1) is to RPA (25) summation], with the parameter numbering PX that determines to intersect or make a variation at step 3l8S.Then, process enters next step 319, because the later step of step 319 and first embodiment's is identical, so this part description has just been omitted.
As mentioned above, in the 15 embodiment, with the special traffic traffic characteristic parameter value of substantial connection is arranged, because the degree of correlation between parameter and the magnitude of traffic flow feature is set to the parameter alternative condition and obtains relative importance value in variation, improved the probability that generation is had the new set of good team control performance thus.
In the 15 embodiment, be the degree of correlation generation ratiometric occurrence rate of each parameter setting with magnitude of traffic flow feature, and according to this occurrence rate selection parameter, so select easily substantial connection and the easy parameter that influences the team control performance are arranged, uprised so generate the probability of new set with good team control performance with the special traffic traffic characteristic.
In the 15 embodiment since when carrying out team control not by its occurrence rate being placed zero selects and the irrelevant controlled variable of magnitude of traffic flow feature, so can prevent from intersection fully or make a variation to be applied to and parameter that magnitude of traffic flow feature has nothing to do.
In the 15 embodiment, be set to the parameter alternative condition owing to parameter and as the degree of correlation between the estimation item of estimation purpose, so have the parameter value of substantial connection in variation, to obtain relative importance value with estimating item, improved the probability that generation is had the new set of good team control performance thus.
In the 15 embodiment, for each parameter is provided with occurrence rate, it is proportional with the degree of correlation as the estimation item of estimation purpose, and according to this occurrence rate selection parameter, so trend towards selecting to influence easily the parameter of team control performance easily, so improved the probability that generates new set with good team control performance.
In the 15 embodiment, owing to does not select and estimate an irrelevant controlled variable, so can prevent fully and will intersect or variation is applied to the parameter that has nothing to do with estimation item as estimating target by its occurrence rate being made as zero.
Moreover, in the 15 embodiment, be combined to form the parameter condition with parameter and as the degree of correlation between the degree of correlation between the estimation item of estimating target and controlled variable and the magnitude of traffic flow feature, so improved the probability that generates new set with good team control performance.
Like this,, reduced generation, estimation, the addition record of useless new set and judged and similar process, made and search and effectively to carry out according to the 15 embodiment.
Note,, then estimate the importance of item according to each if the estimation function of team control performance comprises a plurality of estimations items, to each occurrence rate weighting, or the additive correction value.[the 16 embodiment (intersection another embodiment)] to selecting
Next, will another embodiment of maker be described.In this embodiment, select the intersection set according to the quantity of identity set.About the 16 embodiment, next its part different with first embodiment will be described mainly.
In new set generator program 31 (seeing Figure 11), step 317 obviously is different from first embodiment in order to the operation of selecting according to this embodiment to intersect to PS1 and PS2 (seeing Figure 14), below describes its operation with Figure 46.
Among Figure 46, at first step 317j according to equation [31] calculate each set (if i, j=1,2 ..., P; Distance D ST between the i ≠ j) (i, j).The calculating of calculating the step 414 (seeing Figure 26) in the delete program 35 with the 3rd embodiment is identical.Each parameter value is normalized to the value between 0 and 100.
Step 317K will gather number i and be initialized as 1, the operation of repeating step 317l to 317n, until detect at step 317P all set (i=1,2 ..., P) the occurrence rate RSA (1) of She Dinging is to RSA (P).At step 317l, be j ≠ i, and j=1,2 ..., all numbers of P find DST (i, j)≤set of DSTa (identity set number) counts MDST (i).Whether DSTa is a judgment value, identical mutually in order to judge two set, and in the 16 embodiment, is placed on 25 as the 3rd embodiment.
At step 31 7m, count MDST (i) according to identity set, with formula RSA (i)=1 ÷ { the occurrence rate RSA (i) of MDST (i)+1} set of computations i.That is,, be provided with occurrence rate greatlyyer along with identical set number diminishes.In order to be next step calculating occurrence rate, step 317n will gather number and add 1.
Like this, when having determined occurrence rate RSA (1) for all set to RSA (P), produce two random numbers at step 317q at last with value between [0] and [occurrence rate (RSA (1) is to RSA (P) summation], then, to RUA (P), select two superclass PS1, PS2 according to each random number and occurrence rate RUA (1).
This is the end of step 317 operation, two set PS1 and PS2 be confirmed as standard intersect right.Then, process enters next step 318.Step after the step 318 is identical with first embodiment's, omits so this part is described.
As mentioned above, in the 16 embodiment,, make the probability of set to intersecting with mutual different characteristic so improved owing to select intersection right according to identical set number.Therefore, improved the probability that generates new set with good team control performance.
Note, in the above-described embodiments,, also can adopt other method although only selected a cross parameter (this parameter selection method so-called " a bit intersecting ").Can adopt the method for selecting two or more cross parameters (" multiple spot intersection ") simultaneously.Can also adopt a kind of being called the method for " evenly intersecting ", wherein, prepare in advance to have the program row (mask) with parameter numbering equal length,, determine which superclass can convert subclass to range gene (parameter value) according to each value by the mask appointment.This is identical with [variation].
Although this set has comprised 25 parameters, its quantity and content only are as an example, and the present invention can be applied to any class parameter sets used in the group control algorithm.[the 17 embodiment (another embodiment of system architecture)]
In above-mentioned all embodiment, although with group control device 1 with search in the elevator(lift) machine room that device 10 is installed in the building, and obtain optimal set by on-line operation, also can adopt other method.
For example, such as shown in figure 47 for embodiment 1 to 6 and 8 to 15, also can will search the central monitoring position that device 10 and simulator 2 are installed in elevator maintenance company, and utilize communicator 4A and 4B connection and locating device 10 and group control device 1 by telephone wire.In the case, communicator 4A can carry out data communication with other building that communicator 4B is housed.Adopt this kind structure, usually a cover is searched device and a simulator uses for a plurality of group control devices.Search device 10, simulator 2 and communicator 4A and can be installed in manager's room or safety guard-safeguard center.
According to the 17 embodiment, by a shared costliness search device 10 and a simulator 2, can reduce the cost of system.
Especially in the 8th embodiment, as shown in figure 48, can also be installed in the manager's room in building or the monitoring center of elevator maintenance company with searching device 10, and utilize communicator 4A and 4B connection and locating device 10 and group control device 1 by telephone wire.
When group control device 1 during from the delivery of its factory, when being found best set to merge to write down then by simulator shown in Figure 49 2, maybe when the initialized set of record crowd GPS1 to GPS4, it also is useful searching device 10.Searching device 10 and simulator 2 can be realized by a microcomputer.Moreover, group control device 1, search device 10 and simulator 2 also can be a microcomputer.
Claims (37)
1. system according to many electric schools of group control algorithm team control car, described group control algorithm comprises a plurality of parameters, and described system comprises searches device, in order to search optimal set in the set that offers described group control algorithm as parameter value combination, it is characterized in that the described device of searching comprises:
In order to store the memory storage of a plurality of set;
Generating apparatus in order to selecting one or more set as one or more superclass, and generates the new set of one or more succession part superclass character from memory storage;
Estimation unit in order to when carrying out described group control algorithm with each new set, is searched execution result as the team control performance valve;
Selecting arrangement, in order to by described new set is added to described memory storage and from described memory storage the deletion damaged set improve a plurality of set that are stored in the described memory storage; And
The device of extracting is in improving and being stored in a plurality of set in the described memory storage, according to the described team control performance valve optimal set of extracting.
2. the system as claimed in claim 1 is characterized in that described generating apparatus comprises:
The exchange of values device generates two new set by the value part that exchanges two set selecting from described memory storage;
New value displacement apparatus, the partial parameters value of replacing a set of selecting by the new numerical value that produces with random fashion from described memory storage generates a new set; And
Generation method selecting arrangement is in order to make one's options between exchange of values and new value displacement according to certain probability.
3. the system as claimed in claim 1 is characterized in that described generating apparatus comprises:
The superclass selecting arrangement is in order to select one or more set from described memory storage;
The parameter selecting arrangement in order to select parameter together with described one or two set, carries out exchange of values or new value displacement by this parameter;
The exchange of values device, between two set selecting at described superclass selecting arrangement, exchange generates two new set by the part numerical value of the parameter that described parameter selecting arrangement is selected;
New value displacement apparatus, the parameter value of a set of being selected by described parameter selecting arrangement by the new numerical value displacement that produces with random fashion, selected by described superclass selecting arrangement generates a new set; And
Generation method selecting arrangement is in order to make one's options between exchange of values and new value displacement according to certain probability.
4. system as claimed in claim 3 is characterized in that, described superclass selecting arrangement selects reference information to carry out the superclass selection according to superclass, so that the generating probability of good new set to be provided.
5. system as claimed in claim 4, it is characterized in that, described superclass selects reference information to be the distance between each set, distance between described each set of described superclass selecting arrangement calculating, and from described memory storage, select at random between a pair of wherein said each set apart from the set of satisfying certain condition.
6. system as claimed in claim 4, it is characterized in that, it is described team control performance valve that described superclass is selected reference information, and described superclass selecting arrangement is selected one or two set thus at random according to the selection probability weight of described team control performance valve to each set from described memory storage.
7. system as claimed in claim 4, it is characterized in that, it is the identity set number that described superclass is selected reference information, described superclass selecting arrangement calculates described identity set number for each set, according to the selection probability weight of described identity set number, and from described memory storage, select one or two set thus at random to each set.
8. system as claimed in claim 3 is characterized in that further comprising modifier, in order to revise the superclass alternative condition according to the progress of searching.
9. system as claimed in claim 3 is characterized in that, described parameter selecting arrangement selects reference information to select described parameter according to parameter, to improve the probability that generates good new set.
10. system as claimed in claim 9, it is characterized in that, described parameter is selected reference information poor for two parameter values that will exchange between two set, described parameter selecting arrangement calculates described poor, and selects its described difference to satisfy the parameter of certain condition at random.
11. system as claimed in claim 9, it is characterized in that, it is that lift car utilizes the degree of correlation between situation and each parameter that described parameter is selected reference information, and described parameter selecting arrangement is according to the selection probability weight of the described degree of correlation to each parameter, and selects described parameter thus at random.
12. system as claimed in claim 9, it is characterized in that, it is the degree of correlation between described performance estimation value content and each parameter that described parameter is selected reference information, and described parameter selecting arrangement is according to the selection probability weight of the described degree of correlation to each parameter, and selects described parameter thus at random.
13. system as claimed in claim 3 is characterized in that further comprising modifier, in order to revise the parameter alternative condition according to the progress of searching.
14. system as claimed in claim 2 is characterized in that further comprising the probability modifier, in order to revise the selection probability of each generation method according to the progress of searching.
15. system as claimed in claim 14 is characterized in that, described probability modifier basis is added to the set number of described memory storage and successful index of ratio calculating of estimated set number, and according to the described selection probability of described successful index modification.
16. system according to many lift cars of group control algorithm team control, described group control algorithm comprises a plurality of parameters, and described system comprises searches device, in order to search optimal set in the set that offers described group control algorithm as parameter value combination, it is characterized in that the described device of searching comprises:
In order to store the memory storage of a plurality of set;
The exchange of values device by exchange the partial parameters value between two set selecting as superclass from described memory storage, generates two new set that herid its superclass character partially;
New value displacement apparatus, the partial parameters value of replacing a set of selecting as superclass by the new numerical value that produces with random fashion from described storage arrangement generates a new set that herids its superclass character partially;
Generation method selecting arrangement is in order to select exchange of values method and new value method of replacing in conjunction with the probability of every kind of method;
Estimation unit is when carrying out group control algorithm with one or more new set, in order to search execution result as the team control performance valve;
Adding device is stored to described memory storage in order to only the good new set of satisfying certain addition condition is added;
Delete device satisfies the damaged set of certain deletion condition in order to deletion from described memory storage; And
The device of extracting, in order in the middle of a plurality of set that improve and be stored in described memory storage according to the team control performance valve optimal set of extracting.
17. system as claimed in claim 16 is characterized in that further comprising modifier, in order to revise additional conditions or a plurality of additional conditions.
18. system as claimed in claim 17 is characterized in that, described additional conditions depend on the described team control reference value of each set, and are defined as more and more tighter.
19. system as claimed in claim 16 is characterized in that, it is damaged set that described delete device is deleted its performance estimation value.
20. system as claimed in claim 19 is characterized in that, described delete device is according to the distance between the set, and deletion is gathered identical set with another.
21. system as claimed in claim 16 is characterized in that further comprising apparatus for initializing, searches in order to initialization.
22. system as claimed in claim 21 is characterized in that, described apparatus for initializing comprises first initialize mode and second initialize mode; In described first initialize mode, a plurality of set of before having prepared are used as initialization; In second initialize mode, will in the end search the improved a plurality of set of circulation time as initialization, select described first initialize mode and described second initialize mode according to searching the beginning condition thus.
23. system as claimed in claim 16 is characterized in that further comprising the end judgment means, in order to judge the end of searching according to searching situation.
24. system as claimed in claim 23 is characterized in that, described end judgment means is according to the end of estimating that the judgement of set number is searched.
25. system as claimed in claim 23 is characterized in that, described end judgment means is judged the end of searching according to addition set number.
26. system as claimed in claim 23 is characterized in that, described end judgment means is according to judging the end of searching as the addition set number and the successful index of the ratio of estimating the set number.
27. system as claimed in claim 23 is characterized in that, described end judgment means is with respect to the distance between a plurality of set set of computations of storing in the described memory storage, and judges the end of searching according to the distance between the set.
28. system as claimed in claim 16 is characterized in that, further comprises searching judgment means again, in order to search again according to judging according to the variation of finding to be provided when searching beginning.
29. system as claimed in claim 16 is characterized in that, the described team control performance valve of described memory device stores is to distribute according to each set.
30. system as claimed in claim 16 is characterized in that, the described device of searching is connected with an expected value setting device, in order to set an expected value in conjunction with search procedure.
31. system as claimed in claim 16, it is characterized in that, the described device of searching is connected to group control device, this group control device comprises described group control algorithm and controls the operation of described many lift cars, and the described device of searching is connected to a simulator, this simulator comprises the group control algorithm identical with described group control device, and the estimated result of the described execution of described estimation unit is set to the team control performance valve.
32. system as claimed in claim 31 is characterized in that, described device and the described simulator searched is away from described group control device setting, and the described device of searching is connected by communication line with described group control device.
33. system as claimed in claim 16, it is characterized in that, the described device of searching is connected to group control device, this group control device comprises described group control algorithm and controls the operation of described many lift cars, when described estimation unit is simulated described group control algorithm whenever described group control device, be about to performed result and be set to the team control performance valve.
34. system as claimed in claim 33 is characterized in that, the described device of searching is away from described group control device setting, and the described device of searching is connected by communication line with described group control device.
35. the system in order to many lift cars of team control is characterized in that comprising:
Simulator, it comprises the group control algorithm in order to many described lift cars of team control; And
Search device, it is connected to described simulator, and in order to search the set with optimum parameter value according to described group control algorithm, the described device of searching comprises:
In order to store the memory storage of a plurality of set;
The exchange of values device generates two new set that herid its superclass character partially by mutual switching part parameter value select as two set of superclass from described memory storage between;
New value displacement apparatus is selected partial parameters value as a set of superclass by the new numerical value displacement that produces with random fashion from described memory storage, and generating portion is inherited a new set of its superclass character;
Generation method selecting arrangement in order to the own probability in conjunction with every kind of method, is selected exchange of values method and new value method of replacing;
Estimation unit in order to when carrying out group control algorithm with one or more new set, is searched performed result, as the team control performance valve;
Adder arrives described memory storage in order to the good new set extra storage that only will satisfy certain addition condition;
Delete device satisfies the damaged set of certain deletion condition in order to deletion from described memory storage; And
The device of extracting is improving and is being stored in the optimal set of extracting among a plurality of set in the described memory storage according to the team control performance valve.
36. system according to many elevator cabs of group control algorithm team control, described group control algorithm comprises a plurality of parameters, and described system comprises searches device, in order to search optimal set in the set that offers described group control algorithm as parameter value combination, it is characterized in that the described device of searching comprises:
In order to store the memory storage of a plurality of set;
The chiasma type generating apparatus, by switching part numerical value between selecting as two set of superclass from described memory storage, generating portion is inherited two new set of its superclass character;
Estimation unit in order to when carrying out group control algorithm with one or more new set, is searched performed result as the team control performance valve;
Selecting arrangement by described new set is added to described memory storage, and improves a plurality of set that are stored in the described memory storage by the damaged set of deletion from described memory storage; And
The device of extracting is improving and is being stored in the optimal set of extracting in a plurality of set in the described memory storage according to the team control performance valve.
37. system according to many lift cars of group control algorithm team control, described group control algorithm comprises a plurality of parameters, and described system comprises searches device, in order to search optimal set in the set that offers described group control algorithm as parameter value combination, it is characterized in that the described device of searching comprises:
In order to store the memory storage of a plurality of set;
The anomaly generating apparatus, by selecting from described memory storage as the partial parameters value in the set of superclass with the new numerical value displacement that produces at random, generating portion is inherited a new set of its superclass character;
Estimation unit in order to when carrying out group control algorithm with one or more new set, is searched performed result as the team control performance valve;
Selecting arrangement is by being added to described new set described memory storage and by the damaged set of deletion from described memory storage, improving a plurality of set that are stored in the described memory storage; And
The device of extracting is improving and is being stored in the optimal set of extracting in a plurality of set in the described memory storage according to the team control performance valve.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN94192592A CN1044219C (en) | 1994-05-17 | 1994-05-17 | Elevator group control system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN94192592A CN1044219C (en) | 1994-05-17 | 1994-05-17 | Elevator group control system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1125931A true CN1125931A (en) | 1996-07-03 |
| CN1044219C CN1044219C (en) | 1999-07-21 |
Family
ID=5039552
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN94192592A Expired - Fee Related CN1044219C (en) | 1994-05-17 | 1994-05-17 | Elevator group control system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1044219C (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101837910A (en) * | 2009-03-19 | 2010-09-22 | 株式会社东芝 | Elevator cluster management system and method thereof |
| CN102689824A (en) * | 2011-03-25 | 2012-09-26 | 三菱电机株式会社 | Parameter in elevator and recommended device of equipment |
| CN109715541A (en) * | 2016-09-19 | 2019-05-03 | 通力股份公司 | It is the method for service mode by elevator setting |
Family Cites Families (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5255345A (en) * | 1988-02-17 | 1993-10-19 | The Rowland Institute For Science, Inc. | Genetic algorithm |
| JPH01226678A (en) * | 1988-03-04 | 1989-09-11 | Hitachi Ltd | Elevator controller |
| US4935877A (en) * | 1988-05-20 | 1990-06-19 | Koza John R | Non-linear genetic algorithms for solving problems |
-
1994
- 1994-05-17 CN CN94192592A patent/CN1044219C/en not_active Expired - Fee Related
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101837910A (en) * | 2009-03-19 | 2010-09-22 | 株式会社东芝 | Elevator cluster management system and method thereof |
| CN102689824A (en) * | 2011-03-25 | 2012-09-26 | 三菱电机株式会社 | Parameter in elevator and recommended device of equipment |
| CN102689824B (en) * | 2011-03-25 | 2015-01-07 | 三菱电机株式会社 | Parameter in elevator and recommended device of equipment |
| CN109715541A (en) * | 2016-09-19 | 2019-05-03 | 通力股份公司 | It is the method for service mode by elevator setting |
| CN109715541B (en) * | 2016-09-19 | 2022-01-28 | 通力股份公司 | Method for setting elevator in service mode |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1044219C (en) | 1999-07-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN1117022C (en) | Elevator group control device and elevator group control method | |
| CN1047145C (en) | Traffic means controlling apparatus background of the invention | |
| CN1297933C (en) | Full-lext search device capable of excecuting merge treatment and logon/deletion treatment | |
| CN1221909C (en) | Method for assigning job in parallel processing method and parallel processing method | |
| CN1287282C (en) | Method and system for scheduling real-time periodic tasks | |
| CN1018069B (en) | Group-control method and apparatus for elevator system with plural cayes | |
| CN1120425C (en) | Memory controller and memory control system | |
| CN1939830A (en) | Elevator group management system and control method therefor | |
| CN1299177C (en) | Data management device, computer system and storage medium of storage program | |
| CN1055899C (en) | Traffic means controlling apparatus | |
| CN1760804A (en) | Information processor, information processing method, and program | |
| CN1207716A (en) | Estimation of lobby traffic and traffic rate using fuzzy logic to control elevator dispatching for single source traffic | |
| CN1181551A (en) | Cluster control system | |
| CN1427335A (en) | Circuit set controlling system | |
| CN1878460A (en) | Installation work management method, installation machine and preparation support method, installation assembly line | |
| CN1677286A (en) | Control system for movable body | |
| CN1745347A (en) | Integrated optimization and control using modular model predictive controller | |
| CN1857980A (en) | System and display for an elevator group supervisory and method for supervising a plurality of elevators | |
| CN1819383A (en) | Group-based BCU methods for on-line dynamical security assessments and energy margin calculations of practical power systems | |
| CN1577316A (en) | Security management system in parallel processing system by os for single processors | |
| CN1641567A (en) | Method and apparatus for performing fast speculative pop operation from a stack memory cache | |
| CN101038672A (en) | Image tracking method and system thereof | |
| CN1534504A (en) | Paralled processing system using single processor operating system and parallel processing program | |
| CN1581636A (en) | Network system, server, apparatus management method and program | |
| CN1044219C (en) | Elevator group control system |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 19990721 |