JP3094796B2 - Elevator system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an elevator system and, more particularly, to a control device effective for determining a control method for group management control.

[0002]

2. Description of the Related Art In conventional elevator group management control, as disclosed in JP-A-58-52162 and JP-A-58-63668, control parameters are determined by simulation. Further, in Japanese Patent Application Laid-Open No. Hei 3-18566, when the number of parameters must be increased to perform simulation, the number of combinations of parameters increases. There is something that determines the order.

[0003]

The above-mentioned method of the prior art requires a short calculation time in a range where the effect of parameter change can be checked in advance even if the number of parameters to be adjusted is small or even if the number of parameters is increased to some extent. Is also less effective. However, as the number of parameters increases, the range of parameters to be searched becomes enormous, and the calculation time for finding a practical solution also becomes enormous. In addition, when the usage status and the constraint conditions become diverse, the number of combinations of these becomes enormous. In the case of using a method of limiting the search range by using the conventional knowledge processing technology, it is necessary to acquire a large amount of knowledge in advance for this huge amount of combinations. However, the use situation differs for each elevator, and it is difficult to acquire knowledge of all of them in advance. For this reason, the heuristic search method using the knowledge processing technique cannot be used, so that a large amount of simulation is required as described above, and the calculation time is accordingly long, which makes it difficult to follow the aging. .

An object of the present invention is to reduce the time required for adapting to changes in the use situation due to aging of an elevator and the like, and to adjust a control parameter and the like in a practical time and a device therefor. Is to provide.

[0005]

In order to solve the above problems, a parameter search of control parameters is performed by a stochastic multipoint simultaneous search method or a stochastic search means using a genetic algorithm which is a multipoint simultaneous search method. It is something to do.

Further, in order to follow daily aging, the previous search result is recorded, and the search is continued using the result.

Further, in order to confirm whether or not the search result is practical, the evaluation is performed while changing the evaluation conditions within a certain range. For example, when the evaluation is performed by simulation, the simulation is performed by changing the number of passengers on each floor within a certain range.

Further, the parameters up to the upper predetermined value of the search result are evaluated under the condition that the evaluation condition is changed by a certain range,
The parameters are determined based on the result.

[0009]

The learning means learns the use status of the elevator. The simulation means simulates the operation of the elevator using the control parameters set by the search means under the use situation learned by the learning means.

[0010] The search result storage means stores the search result obtained by the search means.

The search means performs the next search based on the previous search result stored in the search result storage means or the search results up to the previous time, and outputs the best value to the control method registration means. The operation control means controls the operation of the elevator using the control method and the control parameters of the control method registration means. As described above, a new search is performed using the search results up to the previous time, so that a practical control method can be determined in a practical time with the influence of aging or the like.

As a search method of the search means, a multi-point simultaneous search method such as a genetic algorithm is used, and the basic control method stored in the basic control storage means is used for the search under certain conditions. It is possible to prevent a situation in which only a local optimum solution is obtained by diverting search results up to the previous time.

Further, the verification means simulates the solution obtained as the best solution by the search means by the simulation means while changing the use situation learned by the learning means within a certain range, verifies the result, and verifies the performance. If it satisfies, it is registered in the control method registration means. Accordingly, the control method used to actually travel control, minute changes in or use conditions, the minor differences in the actual use state, it is possible to prevent the significant performance deteriorates, infinitesimal <br / > A robust control method can be obtained for various changes.

When the search means employs a method of obtaining the best value by repetitive evaluation such as a genetic algorithm, the verification means changes the use state learned by the learning means at predetermined intervals, within a certain range, it may be obtained a robust control method in advance to infinitesimal changes.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is an overall configuration diagram of the present invention. The hall call information of the hall call buttons 101 to 10n is sent to the hall call collection unit 31 in the group management control device 3 through the transmission line 2. The car information such as the car call of the car call buttons 41 to 44 provided in the car 91 to 94 is sent to the car control devices 61 to 64 through the car transmission lines 51 to 54, and together with other car information. It is sent to the car information collection unit 32 and the assignment control unit 33 in the group management control device 3 through the unit transmission line 7. The operation data collected by the hall call collection unit 31 and the car information collection unit 32 is sent to the learning unit 34,
It is recorded in the learning result table 35. The search unit 36 searches for a control method by referring to the learning result table 35 by a stochastic or multipoint simultaneous search method. The search unit 36
The control method is sent to the simulation unit 37, and the simulation unit 37 checks the conditions of the learning result table 35 and the search unit 3
The simulation which simulates the operation of the elevator is performed by using the control method designated by 6, and the result is sent to the search unit. The search unit 36 searches for a new control method candidate with reference to the simulation result of the simulation unit 37 and repeatedly sends the candidate to the simulation unit 37. When the optimum control method is determined in this way, the result is sent to the search result table 38 and the evaluation expression table 39. The assignment control unit 33 determines an elevator or the like to be assigned to the hall call based on the control method of the evaluation expression table 39, the contents of the hall call collection unit 31, the car information collection unit 32, and the car information. Send to ~ 64. The car control devices 61 to 64 control the hoisting machines 81 to 84 according to the assignment signal, and allow the cars 91 to 94 to service hall calls.

FIG. 2 is a schematic flowchart showing the operation of the present invention. First, the learning unit 34 collects operation data and stores it in the learning result table 35 (201). The search unit 36 searches for a control method based on the contents of the learning result table 35 and the contents of the search result table 38, and determines one or more control method candidates (202). The determined control method is sent to the simulation unit 37 (20
3), a simulation is performed for each control method under the usage status stored in the learning result table (204), and the result is sent to the search unit 36 (205). The search unit 36
Looking at the simulation result, a search end condition is determined whether the result has converged, the predetermined condition is satisfied, or the search has been completed a predetermined number of times (206). If the search condition is not satisfied, the process proceeds to (202). When the return condition is satisfied, one or more control methods with good results are registered in the search result table 38 (207), and the best control method is written in the evaluation expression table 39 (208). The search result table 38 records the latest selected control method for each time period, for each traffic volume, or for each traffic flow. When a hall call occurs, the assignment control unit 33 performs operation control such as determination of an assigned elevator using the control method written in the evaluation expression table 39 (209), and performs this for a certain period of time, for example, one day or a certain number of times. After repeating the traffic volume / hall call until it is determined (210), the process returns to (201) and the search for a new control method is repeated. In the search processing of (202), the search is performed with reference to the contents of the search result table 38 in the second and subsequent searches.

According to this embodiment, the search time can be reduced by referring to the search results up to the previous time.

FIG. 3 is a flowchart when a genetic algorithm is used, and FIG. 4 is an explanatory diagram thereof.
First, a predetermined number of control method candidates are created based on random numbers or the learning results of the learning result table 35, and these are set as initial values (301). The control method is represented as a set of control parameters as shown in FIG. 4A, and here, a case where six sets of control methods are prepared as shown in FIG. Next, the simulation unit 37 is started, and a simulation result when the operation control is performed using this control method is obtained (302). Each control method is ranked based on the simulation result (303). This is Figure 4
This corresponds to the evaluation of (c). Here, it is checked whether the end condition is satisfied (304), and if so, the control method is written in the search result table 38 and the evaluation expression table 39 (305). If the termination condition is not satisfied, the upper control method is left (306). This is the selection in FIG. The parameters of the remaining control methods are exchanged, and the combinations are exchanged to create a control method to be executed next (30).
7). This is the crossover process of FIGS. Next, the parameters of the set of control methods are partially changed using random numbers (308). This is the mutation process shown in FIG. Thereafter, the process returns to (302) and the subsequent processing is repeated.

According to the present embodiment, the optimal control method can be obtained in a practical calculation time even when the number of parameters constituting the control method is large due to the effect of the multipoint simultaneous search by the genetic algorithm. Becomes

In this embodiment, when creating the initial values,
Although the control method candidates are created based on the random numbers or the learning results of the learning result table 35, it is possible to obtain the optimum value in a shorter calculation time by using the contents of the search result table 38 for the second and subsequent times. Become.

FIG. 5 is an explanatory diagram for verifying the above control method. In the hill-climbing method used in the conventional example as shown in FIG.
Since the linearity is substantially satisfied within a certain range, the control method is not so deteriorated even if the usage condition slightly changes, so that the robustness of the control method is guaranteed to some extent. However, the robustness of the control parameters is not guaranteed in the stochastic simultaneous multipoint search, so it is necessary to separately verify the control method determined as the optimum value for the robustness. For example, in the genetic algorithm, points 1a, 1b, 1c, and 1d as shown in FIG.
Are newly generated from the control methods 2a, 2b, 2c, 2
When a d control method as 2d is likely to be selected as the optimum value. Here, the parameter is used as a control parameter, but the effect of such a control method often changes sensitively to a change in the use situation.
If the learning result is a traffic flow as shown in FIG. 5C, the total traffic volume may be slightly changed while the traffic volume ratio of each floor remains unchanged as shown in FIG. By changing the ratio of the traffic volume on each floor as in the above, a predetermined number of traffic flows whose usage status has been changed within a certain range is created, and simulation is performed based on these traffic flows, and the control results are good. Is adopted as a control method. Here, the method of changing the use situation is to increase or decrease the traffic flow in a predetermined range as shown in FIGS. 5D and 5E, to increase or decrease the number of people on each floor in the predetermined range, and to perform them simultaneously. And the one with a good control result includes each usage situation,
The simulation result of the traffic flow has a good average value, a difference between the best value and the worst value and a small standard deviation, or a result of comprehensively judging these.

[0022] FIG. 6 is an embodiment in which to validate a method of controlling the above follow. The search unit 36 creates a verification traffic flow in which the usage status is changed within a certain range based on the learning content of the learning result table 35, and registers the verification traffic flow in the verification data table 310. After the search is completed, the search unit 36 simulates a control method of a higher search result based on the traffic flow in the verification data table 310, and registers a good one of the results in the search result table 38 and the evaluation expression table 39. I do.

FIG. 7 is a flowchart for verifying the control method of the above follow. First, learning result table 3
5 is the same as that shown in FIG. 3 until a verification traffic flow is created (701) and the end condition is checked (304). Before writing in the search result table 38 and the evaluation expression table 39 (305), the verification traffic flow is read (70).
2), the simulation unit 37 simulates a higher control method of the search result based on the verification traffic flow (703). Next, the simulation result is statistically processed to obtain an average, a standard deviation, and the like (704), and a good one is selected from the results (705). Figure 3
Is the same as

According to the present embodiment, it is possible to confirm the robustness of the control method adopted in advance, so that it is possible to select a control method that causes little performance degradation even if there is a slight change in the use situation.

FIG. 8 is a flowchart in the case where the use state is slightly changed every predetermined number of times. In order to select a robust control method as described in FIGS. 5, 6, and 7, instead of verifying the higher search results after the search as shown in FIGS. The traffic flow used for the repetitive calculation is changed slightly in advance so that only a robust traffic flow is obtained as a search result. The process up to (308) is the same as that in FIG. After the control method to be simulated is created, the traffic flow is minutely changed by random numbers based on the contents of the learning result table 35 to create a traffic flow to be used for the simulation (308-1). The traffic flow creation method used in this simulation is shown in FIG.
Is the same as After that, the process returns to (302) and is the same as FIG. According to the present embodiment, it is possible to determine a control method that is robust to minute changes in the use situation in advance.

FIG. 9 is an explanatory diagram when the basic control method is mixed at a fixed rate. In FIG. 2, when the past search result is used as the next search result, the convergence of the solution is promoted, and the calculation time required to find the optimal or practical best solution is reduced. However, even when the search method of the multipoint simultaneous search is used, the solution gradually converges and settles down to the local optimum solution, so that a true optimum solution cannot be searched. In particular, as shown in FIG. 9A, the point x is a true optimal solution, and the point y is a local optimal solution. , FIG. 9
As shown in (c), the position of the optimal solution may change. When the true optimal solution and the local optimal solution are reversed, the method of referring to the past search results uses the points i,
There is a possibility that a search is performed only in a range such as j and k, and a true optimal solution cannot be obtained. Therefore, as shown in FIG. 9D, the basic control method such as the points p, q, and r is mixed in the search at a fixed interval and at a fixed rate so that a true optimum solution can be searched.

FIG. 10 shows an embodiment in which the basic control method is mixed. This embodiment is the same as FIG. 1 except that there is a basic control method table 311 referred to by the search unit when mixing the basic control methods.

FIG. 11 is a flowchart when the basic control method is mixed. It is the same as FIG. 3 until a part is replaced by a random number (308). After replacing a part of the control parameters, it is checked whether or not the number of repetitions has reached a predetermined value (308-2). If the number has not reached, the process returns to (302). If the number has been reached, the basic control method is changed to the basic control method. Read from the table 311 and mix (30
8-3). The following is the same as FIG.

According to this embodiment, a true optimal solution can be obtained while shortening the calculation time by referring to the past search results.

FIG. 12 shows an embodiment in which fuzzy evaluation is performed. When searching daily and adopting the best value as a control method, it is more feasible to keep various forms of solution by fuzzy evaluation than strict evaluation in order to secure diversity of solution. In some cases, the tracking performance can be improved. Therefore, the search unit 36 performs fuzzy evaluation using the fuzzy rules of the fuzzy rule table 312 when evaluating the control method in FIG.

According to the present embodiment, the variety of solutions can be easily ensured, so that the ability to follow aging can be improved.

FIG. 13 shows an example of the learning result table. In this example, the traffic flow is classified for each time zone, and the number of people getting on and off in both directions of UP and DOWN is recorded for each floor, but may be stored according to the time zone.

FIG. 14 shows an example of the search result table. Here, an example in which a control method that separately includes an evaluation formula used for a general floor and an evaluation formula used for a specific floor is selected. As described above, the serviceability can be changed for each floor by using a different evaluation formula for each specific floor, specific zone, and specific elevator, but the number of parameters of the control method is greatly increased. An optimization method for efficiently searching for such a large number of control parameters is required. By configuring the search table as in the present embodiment, it is possible to efficiently search for an optimal control method when different services are provided for each floor, each car, and each zone.

FIG. 15 shows an example of the evaluation formula table.
In this example, the control method and the value of the parameter to be used are stored for each traffic flow, but may be stored for each time zone.

FIG. 16 is an explanatory diagram when different evaluation formulas are used for each car and each floor. As shown in FIG. 16A, the serviceability is changed on each floor and each zone.
As shown in FIG. 16B, in order to provide a car having different characteristics for each floor, control is performed by using a different evaluation formula for each floor and each car performing a special service as shown in FIG. 16C. Need to be done. FIG. 16A shows a case where a priority zone is provided and the waiting time of the floors belonging to the priority zone is particularly improved. In this figure, the priority zone is from the 42nd floor to the 50th floor, and the waiting time of the other floors. Is controlled to be 30 seconds or less, whereas it is controlled to be 20 seconds or less.

FIG. 16B shows an example in which a car having a different property is provided for each floor. In this example, the 10th floor is the waiting time priority floor, and the 6th floor is the car congestion degree priority floor. When a hall call occurs on the 10th floor waiting time priority floor, an elevator A that is close to the 10th floor and has a short waiting time is allocated even if it is somewhat crowded. If a hall call occurs on the 6th floor, which is the priority floor of the congested floor in the car, an empty elevator C is allocated even if the waiting time is somewhat long.
In order to realize such control, a plurality of evaluation formulas as shown in FIG. 16C are required, and the number of control parameters used in one traffic flow and time zone is determined by the number of parameters used in the past. The number is several times the number, and the number of combinations is tens of thousands. Therefore, a stochastic search means as used in the present invention is required.

FIG. 17 shows an embodiment in which the search means is switched according to the set control method. The control device is set in the control method setting table 313 for each traffic flow or for each time zone by the setting device 21. Here, the control method includes normal control, energy saving control, multi-target control for controlling a plurality of control targets such as waiting time, boarding time, degree of congestion in a car, floor-based control, car-based control, and the like. The number of parameters of the method is different from one, several, ten or more. The search method switching unit 314 selects a search method of the search unit 36 based on the control methods registered in the control method table 313. FIG. 18 shows an example of the control method table. When the normal control or the energy saving control with one parameter is set, the search unit 36 performs the hill-climbing method 361.
Search using. When multi-target control with several parameters is set, a search is performed using a search 362 by inference. The heuristic search method uses the knowledge of the knowledge base 362-1 to perform parameter search in the inference unit 362-2 by designating a parameter range and a simulation order. If floor-specific or car-specific control in which the number of parameters is ten or more is specified, or if there is no search knowledge in the knowledge base 362-1, a stochastic search method such as a genetic algorithm is used. Perform a search.

According to the present embodiment, when parameters can be optimized sufficiently by the conventional search method, the parameters can be determined by the conventional method, the calculation time can be reduced, and the performance equal to or higher than the conventional one can be secured. Can be.

[0039]

According to the present invention, there is provided a parameter adjusting method and apparatus capable of adjusting the control parameters and the like in a practical time by shortening the time required for adapting to a change in the use situation due to aging and the like. Can be provided.

[Brief description of the drawings]

FIG. 1 is an overall configuration diagram of the present invention.

FIG. 2 is a schematic flowchart showing the operation of the present invention.

FIG. 3 is a flowchart when a genetic algorithm is used.

FIG. 4 is an explanatory diagram when a genetic algorithm is used.

FIG. 5 is an explanatory diagram in the case of verifying a higher-level control method.

FIG. 6 is an embodiment for verifying a higher-level control method.

FIG. 7 is a flowchart in the case of verifying a higher-level control method.

FIG. 8 is a flowchart in a case where the usage status is minutely changed every predetermined number of times.

FIG. 9 is an explanatory diagram when a basic control method is mixed at a fixed ratio.

FIG. 10 shows an embodiment in which a basic control method is mixed.

FIG. 11 is a flowchart when a basic control method is mixed.

FIG. 12 is an example in the case of performing fuzzy evaluation.

FIG. 13 is an example of a learning result table.

FIG. 14 is an example of a search result table.

FIG. 15 is an example of an evaluation expression table.

FIG. 16 is an explanatory diagram in the case of using different evaluation expressions for each floor and each car.

FIG. 17 shows an example in which a search method is switched.

FIG. 18 is an example of a control method table.

[Explanation of symbols]

3 ... group management control device, 31 ... hall call collection unit, 32 ...
Car information collecting unit, 33: assignment control unit, 34: learning unit,
35: learning result table, 36: searching unit, 37: simulation unit, 38: search result table, 39: evaluation expression table, 310: verification data table, 311: basic control method table, 312: fuzzy rule table.

──────────────────────────────────────────────────続 き Continued on the front page (72) Inventor Yoshiaki Ichikawa 7-2-1, Omika-cho, Hitachi City, Ibaraki Pref. Energy Research Laboratory, Hitachi, Ltd. (72) Inventor Kenji Yoneda 1070 Imo, Katsuta-shi, Ibaraki Hitachi, Ltd. Mito Factory (56) References JP-A-4-20467 (JP, A) JP-A-4-235869 (JP, A) (58) Fields investigated (Int. Cl. ⁷ , DB name) B66B 1 / 18

Claims (4)

(57) [Claims] 1. A plurality of elevators serving on a plurality of floors, a use state collecting means for collecting the use state of elevators, a learning means for learning the use state of the elevators,
A search means for searching for an optimum control method based on the use situation learned by the learning means, wherein the control methods up to a predetermined upper-ranking order among the control methods searched by the search means An elevator system comprising: simulation means for simulating elevator operation by changing the learning content of the learning means; and means for determining an optimal control method by the search means based on the simulation result. 2. A system comprising: a plurality of elevators working on a plurality of floors; a use state collecting means for collecting the use state of the elevator; a learning means for learning the use state of the elevator; Simulation means for simulating elevator operation using a learning result of the learning means to try the control method searched by the search means in an elevator system comprising: a search means for searching for an optimum control method based on a use situation. An elevator system, wherein the simulation means uses the control method selected by the search means, and changes a use state used for the simulation within a predetermined range every time the simulation is repeated a predetermined number of times. 3. A basic control method registering means for registering a basic control method in advance, wherein said simulation means adds a basic control method registered in said basic control method registration means to a control method for trial under predetermined conditions. 3. The elevator system according to claim 1, wherein a simulation is performed. 4. An elevator system comprising: a plurality of elevators operating on a plurality of floors; and an elevator operation control device operated based on a request from a user.
Learning means for learning the usage situation of the user, simulation means for predicting the performance of the plurality of control method candidates with respect to the usage situation learned, and optimization using the performance prediction results of the simulation means Searching means for searching for a suitable control method, wherein the simulation means performs performance with respect to a plurality of usage situations in which a target usage situation is changed within a predetermined range for a predetermined number or all of the plurality of control methods. The elevator system, wherein the prediction is performed, and the search means selects an optimal control method using the performance prediction results in the plurality of use situations.