JP3090832B2 - Elevator group control device - 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 group management and control system in which a plurality of elevators are used for a plurality of floors, and more particularly to an elevator group management and control system characterized by hall call assignment control.

[0002]

2. Description of the Related Art In recent years, when a plurality of elevators are juxtaposed, an elevator which responds to a hall call on each floor is provided by a microcomputer in order to improve the operation efficiency of the elevators and the service to elevator users. It is being done to use a small computer such as this to make the assignment rational and quick.

[0003] That is, when a hall call occurs, the most suitable elevator to service the hall call is selected and assigned, and other elevators are prevented from responding to the hall call.

[0004] In the group management control device of such a system, recently, grasp of inter-story traffic of each building such as measurement of car call registration data when responding to each hall call in real time, data measurement during getting on / off, etc. The demand unique to the building is grasped on the basis of the measurement data and used for hall call assignment control.

In such a situation, the hall call assignment control of the elevator generates an optimum value of each evaluation index on the group management performance with respect to the occurrence of the hall call, and determines the optimum car.

[0006]

Generally, the evaluation criterion of "optimal" in terms of group management performance is specific to various buildings such as a hotel, a tenant building, a building occupied by one company, an elevator user, and a building management company. Depends on the situation
Group management control performance must be matched to the evaluation criteria for each building, and it was not possible to automatically set evaluation criteria specific to each building. A new elevator group system which can automatically set an evaluation criterion specific to each building by applying and correcting the connection state of the neural network according to the actual use state after the operation of the building has already been disclosed in Japanese Unexamined Patent Publication No. Hei.
No. 76 is known.

However, the process of extracting the result data after system operation as teacher data and the process of re-learning the neural network using the teacher data require a long time, and the calculation time varies due to fluctuations in demand. The processing is not completed within a predetermined time and overlaps with other calculations, delaying the allocation control processing, or performing allocation control using old learning data, so that optimum allocation cannot be performed in the current use state of the building.

In particular, in learning, the time until convergence varies, and depending on how the initial conditions are given, the processing does not converge to a predetermined error, and the processing becomes an infinite loop.

[0009] In contrast, the clock frequency changes and for example 64-bit CPU (Central Processing Unit), a micro
A method of improving the performance of the CPU and solving the above problem by adopting a processor or the like can be considered. However, this leads to a significant increase in cost, and a large result cannot be expected for the cost increase.

[0010] The present invention has been made in view of such circumstances, and even if traffic demand and processing data fluctuate, it does not affect other tasks, and the effective result data during system operation is not affected. Processing such as extraction / synthesis and learning / updating of a partial model using the data as teacher data can be performed, and data processing can be performed reliably, leading to improvement in system reliability. In addition, calculation start time and time limit Is set in advance to a time zone with small demand,
It is an object of the present invention to provide an elevator group management control device that leads to load distribution of U.

[0011]

According to a first aspect of the present invention, a plurality of elevators are put into service on a plurality of floors, and an optimum unit is obtained by performing a predetermined evaluation operation on a common hall call generated. Is applied to an elevator group management control device that applies the learning function of the neural network to the assignment control and corrects the connection state of the neural network according to the actual use state after the building has been operated . And the above
In order to solve the problem, the elevator group management system of the present invention
In the control device, at regular intervals obtained in actual use
Teacher data extracting means for sequentially extracting the result data as teacher data, re-learning means for sequentially re-training the neural network with the teacher data extracted at regular intervals ,
The extraction processing time in the teacher data extraction means exceeds a predetermined value
Interrupt the teacher data extraction means and re-learn
If the re-learning processing time in the means exceeds a predetermined value,
Interrupting means for interrupting the learning means.

Further, the elevator according to the second aspect of the present invention.
In the data group management control device, teacher data extracting means for sequentially extracting, as teacher data , result data obtained at regular intervals obtained in the actual use state, and extracted at regular intervals. At the time of the re-learning means for sequentially re-learning the neural network based on the teacher data and the extraction processing by the teacher data extracting means
When the interval exceeds a predetermined value, processing by the teacher data extraction means
Re-learning in re-learning means as well as lowering the level
If the learning processing time exceeds a predetermined value, the processing in the re-learning means
Processing level lowering means for lowering the level.

[0013]

According to the invention corresponding to the first aspect, for example, the result data for each certain period of each day of the week and each time of the building is extracted as teacher data at the determined time. Next, after the process of extracting one result data, it is checked whether or not the operation time exceeds a predetermined value. If not, the process of extracting the next result data is similarly continued. If it exceeds, the calculation is interrupted, and in the next calculation, extraction processing is performed from the next result data. When the extraction process is performed a predetermined number of times, the partial model is re-learned using the extracted data as teacher data.
Each time learning is performed once for all the extracted data, a check is made as to whether a predetermined time has been exceeded, and if not, learning is continued. If it exceeds, suspend the calculation,
When the number of times of learning, error, and the like satisfy a predetermined value, the result of learning is adopted.

According to the invention corresponding to claim 2, for example, the result data for each fixed period of each time of each day of the building and each time is extracted as teacher data at the determined time. Next, after the process of extracting one result data, it is checked whether or not the operation time exceeds a predetermined value. If not, the process of extracting the next result data is similarly continued. If it exceeds, the processing level is lowered, and the calculation is continued until the extraction of a predetermined number of result data is completed. When the extraction process is performed a predetermined number of times, the partial model is re-learned using the extracted data as teacher data. Each time learning is performed once for all the extracted data, a check is made as to whether a predetermined time has been exceeded, and if not, learning is continued. If it exceeds, the processing level is lowered, and if the number of times of learning, the error, etc. satisfy a predetermined value, the result of learning is adopted.

As a result, according to the invention corresponding to claim 1 or claim 2, even if traffic demand or processing data fluctuates, effective result data during system operation can be obtained without affecting other tasks. Extraction and synthesis, and learning / updating of a partial model using the data as teacher data. In addition, reliable data processing can be performed, leading to an improvement in the reliability of the system. By setting the time in advance to a time zone where demand is small, CP
It is possible to obtain an elevator group management control device that leads to the U load distribution.

[0016]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 shows an overall system configuration of an elevator group management apparatus according to an embodiment of the present invention. Reference numeral 1 denotes a group management control unit. This group management control unit 1 controls each unit elevator. And the learning control unit 1-1 are connected via a high-speed transmission line 6 as first transmission control means. The group management control unit 1, the learning control unit 1-1, and the single control units 2-1 to 2-N are configured by a single or a plurality of small computers such as microcomputers, and operate under software management.

Reference numeral 3 denotes a hall call button provided on each floor, and reference numeral 4 denotes a hall call input / output control unit for inputting / outputting a hall call. Then, the group management control unit 1, the learning control unit 1-1, the single control units 2-1 to 2-N, and the hall call input / output control units 4 are connected via the low-speed transmission path 7 as the second transmission control means. Connected.

The high-speed transmission line 6 includes the single control units 2-1 to 2-
This is a transmission control system for performing transmission between N and the group management control unit 1 and the learning control unit 1-1, that is, mainly between control computers in a machine room, and is connected by a high-speed and highly intelligent network. Then, control information required for the group management control is exchanged between the group management control unit 1, the learning control unit 1-1, and each of the unit control units 2-1 to 2-N at high speed.

The low-speed transmission line 7 is a control system for transmitting information mainly via a hoistway, such as a hall call button 3 of each hall and a monitoring panel 5 in a monitoring room. It is a low-speed and long-distance optical fiber cable.
-1 and the single control units 2-1 to 2-N to exchange data.

When the group management control unit 1 is normal, the hall call button 3 is controlled by the group management control unit 1 via the low-speed transmission line 7, and when the hall call button 3 is pressed, the hall call gate is closed. Along with setting the registration lamp, the unit control units 2-1 to 2-N sent via the high-speed transmission line 6
The optimal unit is determined based on the information of
A control command is issued to the corresponding one of .about.2-N. The single control unit that has received the control command performs the single control using the control command as hall call information.

FIG. 2 shows an example of a software system of the group management control section 1 and the single control sections 2-1 to 2-N in the first embodiment of the elevator group management control apparatus of the present invention. Are a single control function task 9 and a group management control main function task 1 by a real-time OS 8 which is an operating system (OS).
0, group management control subfunction task 11, transmission control task 1
In this configuration, the tasks 9 to 12 are activated or held by a scheduler in the real-time OS 8.

The single control function task 9 among these tasks 9 to 12 is a function task for operating the single control units 2-1 to 2-N, and has a higher priority.

The group management control main function task 10 is a central function task of the group management control unit 1. The group management control sub-function task 11 distributed to each of the single control units 2-1 to 2-N is used for each unit. The optimum car is determined by collecting information data of each car and performing a comparison operation, and issues a control command to the car concerned and controls hall call registration.

The group management control sub-function task 11 is a function task for processing information for each unit of the group management control unit 1, and performs information processing under the control of the group management control main function task 10. In other words, the computer having the group management control main function task 10 is configured to manage the activation and termination of the task via the high-speed transmission line 6.
In response to a command from the server, distributed processing is performed for each unit, and data is conveyed to the group management control main function task 10 when the processing is completed.

The transmission control task 12 controls transmission and reception of data of the high-speed transmission line 6 and activation and termination of the group management control subfunction task 11.

FIG. 3 is a block diagram showing a system configuration of the high-speed transmission line 6 shown in FIG. 1. The transmission control is performed by using a microprocessor 13.
Data transmission using a data link controller 14 and a media access controller 15 configured by hardware as a part for controlling a data link layer of a network model layer of a LAN (local data communication network) proposed by (International Organization for Standardization). And the ratio of transmission control software managed by the microprocessor 13 for high-speed transmission control is reduced.

For example, as the data link controller 14 which is a controller for realizing the above-mentioned high intelligent transmission control, L of Intel (INTEL) is used.
The i82586, which is an SI, and the i82501, which is also manufactured by Intel Corporation as the media access controller 15, have been put into practical use.
A high-speed transmission function such as 0 Mbit / sec can be performed relatively easily with a reduced support ratio of the microprocessor 13.

In FIG. 3, 16 is a system bus, 17 is a control line, and 18 is a serial transmission system.

FIG. 4 shows a learning control unit 1-1 according to the present invention.
FIG. 5 is a functional block diagram showing the flow of input / output signals of FIG.
6 shows an example of the system configuration of the learning control unit 1-1 in FIG. 4. FIGS. 6 and 7 show the detailed system configuration of the inference unit 21 and the partial model unit 22 in FIG. 5, respectively.

In FIG. 4, the group management control unit 1 includes the single control units 2-1 to 2 of the elevator group system 2 as described above.
-N cooperates with the group management control sub-function task 11 to execute the hall call assignment control function.

The evaluation algorithm used for this hall call allocation control evaluates each evaluation index on the group management performance,
Comprehensive evaluation is performed by weighting and adding each evaluation value.

The learning control unit 1-1 includes an elevator group system 2 transmitted through the group management control unit 1 at regular intervals, that is, a single unit control unit 2-1 to 2-N and a hall call input / output control unit. The traffic demand and the group management control response result are generated based on the information from No. 4, the optimal control parameters are set from the traffic demand, and the group management control response result is used as base data for online learning.

Then, the learning control unit 1-1 transmits the control parameters set for each fixed time period at each time to the group management control unit 1 as optimal evaluation index weight values.

The group management control unit 1 is composed of unit control units 2-1 and 2-2.
-N, calculate the evaluation value of each evaluation index based on the information from N, perform the optimum weighting on the evaluation value by the control parameter transmitted from the learning control unit 1-1, and give a control command to the optimum unit. .

Next, the detailed operation of the learning control section 1-1 will be described with reference to FIG.

The functional configuration of the learning control unit 1-1 will now be described. As shown in FIG.
, A synthesis unit 23, an inference result evaluation unit 24, a system adjustment unit 25, and an operation monitoring unit 26.

The evaluation calculation in the group management control unit 1 is generally performed for a plurality of evaluation indices l, and for the i-th unit, g ₁ (i), g ₂ (i),..., _Gl (i) And the overall evaluation value is obtained by weighting and adding the evaluation evaluation value for each evaluation. The overall evaluation value E _i for the _{i-th unit} is

[0039]

(Equation 1) Is expressed as

Here, α _j is a weight value for each evaluation index j, and is a control parameter transmitted from the learning control unit 1-1 to the group management control unit 1.

[0041] Therefore, the inference result evaluation unit 24 calculates the traffic demand for each predetermined period of another each time period, and outputs the inference unit 21 generates a combination of control parameters alpha _j in a predetermined range, Output to the partial model unit 22, and as a result, a group management control response inference result corresponding to the combination of the control parameters α _j is evaluated according to the use of the building in which the elevator is installed, and optimal control parameters are set.

Assuming that the group management control response inference result is y and the input to the learning control unit 1-1 is u, the operation in the combining unit 23 can be expressed as y = F (u) (1). Note _{that, y = (y 1, y} 2, ......, y n) T u = (u s, u e) and ^{T = (C, α) T} .

In the group management control response inference result y,
_{y 1, y 2, ......,} y n incidence of hall call response time at a certain time, average vans rate, represents the inference results, such as average service time, and evaluation reference data in determining the group management performance Become.

Further, C in the input u represents traffic demand, and if C = (c ₁ , c ₂ , c ₃ ), c ₁ , c ₂ , c ₃ are respectively the total average passenger generation interval [s / person], It is data representing the average passenger occurrence interval at the reference floor, the average passenger occurrence interval toward the reference floor,
It indicates the status of the system such as the congestion degree of the system and the flow of people.

Further, α is a weighting value (control parameter) for each evaluation index.
Α = (α ₁ , α ₂ ,..., Α _l ).

Therefore, the inference unit 21, the partial model unit 22,
The target model composed of the synthesis unit 23 is expressed by synthesis of m partial system models f _i (α), (i = 1, 2,..., M), and the expression (1) can be rewritten as the following expression. Will be.

[0047]

(Equation 2) Here, a _i (C) indicates the activity of the partial system model f _i (α) in the traffic demand C, and the group management response inference result y indicates the state of the system obtained from the traffic demand C and the part in the partial model unit 22. It is determined by the connection with the system model.

Next, the system configuration of each of the inference unit 21, partial model unit 22, inference result evaluation unit 24, system adjustment unit 25, and operation monitoring unit 26 will be described.

The inference unit 21 is, as shown in FIG.
1-1, storage unit 21-2, output unit 21-3, and gate 21
-4, which receives traffic demands C from the inference result evaluation unit 24, and _ai (C), (i = 1,
2,..., M).

The input unit 21-1 has a state vector V k-dimensional of k neurons constituting the traffic demand C inputted by passing the membership functions phi _i, each element in its membership grade is the partial input vectors c _i, and outputs the (i = 1,2, ......, M ). The M partial input vectors c _i are collectively input to the input state vector V as an input vector C.

The storage unit 21-2 is composed of an r-dimensional state vector X composed of r neurons.
-1 and a storage unit that associates the output unit 21-3.

The output unit 21-3 comprises an m-dimensional state vector Z composed of m neurons, and each element Z _i ,
(I = 1, 2,..., M) correspond to the partial system model f _i (α) of the partial model unit 22.

The input unit 21-1 and the storage unit 21-2, the storage unit 21-2 and the output unit 21-3 each have a mutual loop, and each of the units 21-1 to 21-3 has a self-loop.

This relationship is in a discrete form and is expressed as follows.

C (k) = φ (u (k)) (3.1) V (k + 1) = φ (W _VC × C (k) + W _VV × V (k) + W _VX × X (k)) … (3.2) X (k + 1) = φ (W _XV · V (k + 1) + W _XX · X (k) + W _XZ · Z (k))… (3.3) Z (k + 1) = φ (W _ZX · X (k + 1) + W ZZ · Z (k)) ... (3.4) V (0) = V 0, X (0) = X 0, Z (0) = Z 0, k ≧ 0 here, W _VC is a matrix representing a load from the vector C to the vector V, and is a synapse load on the vector C of the neuron constituting the vector V. Also,
The same applies to W _VV , W _VX , W _XV , W _XX , W _ZX , and W _ZZ .

Φ is a j-dimensional membership function,
φ is a sigmoid function corresponding to each dimension, and performs an operation of f (x) = 1 ÷ [1 + exp (−x)] (3.5) for each input element.

Further, k is a parameter representing time, and the unit time elapses each time it increases by one.

The traffic demand C (u) input by appropriately setting each W in the above equations (3.1) to (3.4)
(K)) for the partial system model f _i (α), (i
.., M) appear in the state vector Z of the output unit 21-3 over time.

[0059] The gate 21-4 is opened after a lapse of time is set T Z _i (T) the partial system model f
_It is output as the activity a _i (C) of _i (α).

Next, partial model 22 is a system configuration shown in FIG. 7, the control parameter group supervisory control response results for each partial system model by inputting the alpha f _i
(Α) is output.

Each of the partial system models f _i (α), (i = 1, 2,..., M) in the partial model unit 22 is constituted by a multilayer neural network as shown in FIG. Corresponds to a particular demand C _i , each of which is a control parameter α for that demand
And the input / output data i of the group management control response result of the real system. And the partial system model f
_i (α) is learned using the back-propagation method using the input / output data i as teacher data.

[0062] As shown in detail in Figure 8, each partial system model f _i when the input u is given to perform the calculation of _{y (k) = f i (} u (k)). This calculation processing is: net (k) = W _hu (k) · u (k) (4.1) h (k) = φ (net (k) + θ _h (k)) (4.2) nety (K) = W _yh (k) · h (k) + W _yu (K) · u (k) (4.3) y (k) = φ (nety (k) + θ _y (k)) (4) ... 4) k ≧ 0 where W _hu , W _yh and W _yu are matrices representing synaptic loads. Θ _h and θ _y are vectors representing bias values for the intermediate layer h and the output layer y, respectively.

Each partial system model f _i (u), (i =
1,2, ..., m) are different synapse loads,
Operate with bias.

The synthesizing unit 23 outputs the partial system model outputs f ₁ (α), f
₂ (α),..., F _m (α) and the activity a for each partial system model input from the inference unit 21
₁ (C), a ₂ (C),..., _Am (C) are synthesized according to the equation (2), and output to the inference result evaluation unit 24 as the group management control response inference result y.

In the inference result evaluation unit 24, as described in detail later, the inference unit 21, the partial model unit 2
2. For the target model composed of the synthesis unit 23, input u
In (= (C, α) ^T ), a combination of control parameters α is generated within a predetermined range with respect to the traffic demand of the actual system, and given as an input u. The corresponding group management control response result is evaluated, an optimal control parameter α is set, and the result is transmitted to the group management control unit 1.

Next, the traffic demand from the group management control unit 1 and the group management control response result obtained online, and the control unit 25 based on the control parameters from the inference result evaluation unit 24.
The online learning performed by the inference unit 21 and the partial model unit 22 will be described.

In the inference unit correction unit 25-2, the loop from the output unit 21-3 to the storage unit 21-2, that is, the matrix W _XZ , among the loops for correcting the certainty in the inference unit 21 shown in FIG. limit. At this time, the inference unit correction unit 25-2 outputs the (i, j) element w of the matrix W _XZ.
ij is expressed as follows: w _ii = p _i , (i = 1, 2,..., m) (5.1) W _ji = −p _i , (j and i are not equal) (5.2) To fix.

Here, pi ≧ 0 is the partial system model f
_i is a parameter indicating the certainty of storage for (α), and is calculated by the following equation.

P _i = {· R _i (k + 1) + ζ (6.1) R _i (k + 1) = 1-exp [−β (N _i (k + 1) + γ)] (6.2) N _i ( k + 1) = N _i (k) + δN _i (6.3) δN _i = min ｛1, η × [a _i (R _i (k))]｝ (6.4) where η, β , Γ, ξ, ζ are constants, R _i , N
_i is the proficiency and learning progress of the partial system model f _i (α), respectively. Then, the learning progress N _i (k)
Represents the progress of the learning after learning the partial system model f _i (α) k times, and is proportional to the activity a _i ,
It changes with a degree δN _i (maximum 1) inversely proportional to the current proficiency level R _i (k).

Then, for the new learning progress N _i (k + 1) calculated by the equation (6.3), the new proficiency R
_i (k + 1) is determined by the equation (6.2), and the finally corrected certainty p _i is obtained by the equation (6.1).

On the other hand, the modification of the partial system model f _i (α) is performed by the input / output data correction unit 25-3, the input / output data group 2
5-41-4m and partial system model correction unit 25-
At 2, input and output data correction unit 25-3 is first modified degree .delta.N _i, control parameter alpha _o, output data group having from each partial system model f _i (alpha) is the group management response result y _o 25-41 Rewrite ~ 4m. The input and output data 25-4i was Tsu rewritten as teacher data, back propagation method by subjected to additional or relearn a partial system model correction unit 25-5, core part system model after learning the old part system model f _i Replace with f _i ′.

Next, a method of correcting input / output data will be described. The input / output data correction unit 25-3 includes the operation monitoring unit 2
When the command of No. 6 is input, the group management control response result is calculated based on the response result in the time zone on the predetermined day of the week, and the input / output data _Do = ( _uo , _yo ) together with the control parameters in the time zone. , U _o = (C _o ^* , α _o ) ^T.

At this time, the partial system model f _i (α)
Rewrite the input / output data (D ₁ , D ₂ ,..., D _L ) possessed by.

All the input / output data are scanned from the input / output data groups 25-41 to 4m, and the square of the distance between α and α _o dα = | α−α _o | ² (7.1) is obtained. , Two data D ^(1st) and D ^(2nd) are extracted in order from the closest (d _α is smaller).

Here, if C ^(1st) = C _o and d ^(1st) α> 0 (7.1 ′), D _o is additionally registered as input / output data D _{L + 1} in exceptional cases. .

If the expression (7.1 ') does not hold, the square of the distance between C and _Co of the two data D ^(1st) and D ^(2nd) d _C = | C−C _o | ² ... (7.2) is obtained, and then D ^(1st) and D ^(2nd) are corrected by the following equation.

Here, y ′ ^(1st) and y ′ ^(2nd) are output data after correction.

Y ′ ^(1st) = ρ ₁ · _yo + (1−ρ ₁ ) · y ^(1st) (7.3) y ′ ^(2nd) = ρ ₂ · _yo + (1−ρ ₂ ) · Y ^(2nd) ··· (7.4) κ = 1 ÷ [γ · D ^(1st) α + 1] · · (7.5) ρ ₁ = δN _i · 1 ÷ [λD ^(1st) C + 1] · κ · (7 .6) ρ ₂ = (1−κ) · δN _i × {1} [λD ^(2nd) c + 1]} × {1} [γD ^(2nd) α + 1]} (7.7) The above (7.3) ) And (7.4), the input / output data in the partial system model is rewritten, and the input / output data after rewriting is used as teacher data in the partial system model correcting unit 25-5 by the back propagation method to load the partial system model. The matrix is modified by the following procedure.

Δ _y (k) = φ ′ (nety (k) + θ _y (k)) * (y ^* (k) −y (k)) (8.1) δ _h (k) = φ ′ ( net (k) + θ _h (k)) * W ′ _yh · δ _y (k) (8.2) ΔW _yh (k + 1) = η _yh (k) · δ _y (k) · h ′ (k) + α _yh · ΔW _yh (k) (8.3) ΔW _hu (k + 1) = η _hu (k) · δ _h (k) · u ′ (k) + α _hu · ΔW _hu (k) (8.4) ΔW _yu (k + 1) = η _yu (k) · δ _y (k) · u ′ (k) + α _yu · ΔW _yu (k) (8.5) W _yh (k + 1) = W _yh (k) + ΔW _yh (k + 1) ... (8.6 ) W hu (k + 1) = W hu (k) + ΔW hu (k + 1) ... (8.7) W yu (k + 1) = W yu (k) + ΔW yu (k + 1) ... (8 .8) ΔW _yh (0) = 0, ΔW _hu (0) = 0, ΔW
_yu (0) = 0, Δθ _y (0) = 0, Δθ _h (0) = 0 where y ^* is the teacher data and ｜| y ^* −y | ² <ε is all y ^* Is incremented by k and repeated.

Here, A * B represents a product for each element of the matrices A and B, and η and α are learning parameters.

By the above operation, the partial system model correcting section 25-5 corrects the load matrix of the partial system model every time the input / output data group 25-41 to 4m is rewritten by the input / output data correcting section 25-3. Do and re-learn.

An outline of the configuration and operation of the operation monitoring unit 26 will be described with reference to FIGS.

The operation monitoring unit 26 comprises an operation start / interruption command unit 26-1 and a time management unit 26-2.
At a certain time, for example, at midnight, the time management unit 26-
2 outputs a calculation start command to the system adjustment unit 25 via the calculation start / interruption command unit 26-1. When the command is input, the system adjustment unit 25 starts processing in each of the units, and outputs a calculation continuation request to the calculation monitoring unit 26 each time the processing in a predetermined unit is completed. When a predetermined number of processes have been completed, an operation completion signal is output to the operation monitoring unit 26.

The operation monitoring unit 26 determines whether or not a predetermined time limit has been exceeded. If the time limit has not been exceeded, a calculation continuation command is transmitted to the system adjustment unit 25 if an operation continuation request is received, and the operation completion If it is a signal, it does not output a calculation continuation command to the system adjustment unit 25. If it has exceeded, an operation interruption command is transmitted to the system adjustment unit 25, and the process of extracting the result data of the system such as the group management control response result or the optimal control parameter for each time zone and the extracted result data are transmitted to the teacher data. The learning process is interrupted. For the extraction process, the calculation is started from the time when the previous extraction process was interrupted by the next calculation start command, and for the learning process, if the number of learning or the error satisfies a predetermined value when the learning is interrupted,
The partial model is updated and the inference unit is corrected. If the partial model is not satisfied, the learning data is invalidated and learning is performed again in the next calculation.

Next, the detailed configuration of the inference result evaluation unit 24 and the operation of optimizing the control parameter α will be described with reference to FIGS. 9 and 10.

As shown in FIG. 9, the inference result evaluation unit 24 includes a control parameter combination generation unit 24-1, a traffic demand detection unit 24-2, and an inference result evaluation parameter setting unit 24.
-3, an inference result evaluation operation unit 24-4, and a control parameter setting unit 24-5.

As described above, the inference unit 2 of the learning control unit 1-1
1. The relationship between the control parameter and the group management control response inference result for an arbitrary traffic demand for each building is configured by the partial model unit 22 and the synthesis unit 23, and the group management control response result y is calculated by the equation (1). It could be estimated by association.

Therefore, the inference result evaluation unit 24 sets the group management control response result y to the optimum value of the control parameter with respect to the “optimum” reference for each building according to each building use and customer needs. Do.

For this purpose, the inference result evaluation section 24 detects traffic demand in a predetermined time zone and outputs the detected traffic demand to the inference section 21, as well as inference result evaluation parameters set in advance according to the use of each building and customer needs. Then, the inference result evaluation parameters in the current time zone are selected.

Here, the inference result evaluation parameter is a parameter table for evaluating the group management control response result y, and is set for each building for each traffic demand. The group management control response result y is an index indicating the group management control performance, and is evaluation reference data such as the occurrence rate of hall call response time, the average multiplication rate, and the average service time, as described above.

When the group management control performance is evaluated, the evaluation is performed based on the evaluation reference data. In general, the weighting of the evaluation reference data differs depending on the building use, customer needs, and traffic demand. In such a case, items such as hall call response time and average service time have a high weight, and in hotels, on the contrary, items having a low average crossing ratio have a high weight. Similarly, even in buildings of the same use, the specific gravity of the items changes depending on the time zone and customer specifications. Therefore, the inference result evaluation parameter is configured as a weight value for evaluating each index of the group management control response result y using the traffic demand and the time zone as array elements in accordance with the use of each building.

The inference result evaluation parameter β selected based on the predetermined traffic demand C is calculated by the inference result evaluation operation unit 2
4-2, where the group management control response result y output from the synthesizing unit 23 is evaluated, and the performance evaluation value PE, which is the evaluation result, is transmitted to the control parameter setting unit 24-5, and the optimal The control parameter α _o is set.

The operation of the inference result evaluation unit 24 will be described in detail with reference to the flowchart of FIG. 10. First, in the control parameter combination generation unit 24-1, each control parameter α is changed by a small change amount Δα within a possible range. And a finite number of combinations P for each α
_max (Step S (hereinafter referred to as S) 1, S
2).

Next, the control parameter P (α _1P , α _2P ,..., Α _lP ) for this combination P is input together with the traffic demand parameter C of the current time zone by the traffic demand detector 24-2. (= (C, α) ^T ) is input to the inference unit 21 and the partial model unit 22, and as a result, the synthesis unit 2
3 to obtain the group management control response inference result y _P (S
3).

Based on the inference result y _P , the performance evaluation value PE is modeled as a mathematical model as an objective function indicating the group management control performance evaluation, and the performance evaluation is quantified (S4).

When the inference result is evaluated using the evaluation value for the control parameter combination P as PE _P , the following results are obtained.

[0097]

(Equation 3) Here, β (= (β ₁ , β ₂ ,..., Β _n )) is an inference result evaluation parameter, which is data set in advance for each building use and traffic demand and time zone according to customer specifications. Is an array.

When the performance evaluation value PE _P ; (P = 0 to P _max ) is calculated for all the combinations P, (S
5), the control parameter P _o that the combination P of the PE _P is minimized and P _o, corresponding to the combination P _{o
(Α 1Po, α 2Po, ......} , α lP o) was set to the optimal control parameters are transmitted to the group management control unit 1 (S
6).

As shown in FIG. 11, the system adjustment unit 25 includes a difference detection unit 25-1, an inference correction unit 25-2, an input / output data correction unit 25-3, and input / output data groups 25-41 to 25.
-4m and a partial system model correction unit 25-5. When a calculation start command is input from the calculation monitoring unit 26, the group management control inference result y for each time zone and the group management control unit 1 based on the difference of the response result, it serves to fix the "certainty" p _i and related parts system model f _i of the inference section 21. The correction amount is made in direct proportion to the activity a _i and in inverse proportion to the proficiency R _i (k).

[0100] When the difference detector 25-1 difference inference result y and response result y _o satisfies the equation _{(10), | y-y o | /} y o ≧ D efo ... (10) fixes the amount of the original the degree .delta.N _i to be (6.4) is calculated from the equation, and outputs the inference section modification portion 25-2 and the input and output data correction unit 25-3.

Here, _Defo is a constant and serves as a criterion for determining whether or not to modify. The input / output data correction unit 25-3 outputs the response result and control parameters suitable for the demand in the time zone to the input / output data i25-4i corresponding to the demand, and the output data and the input / output data i25-4i are similar. If there is any data, it is synthesized and input / output data i
25-4i is updated. Then, the input / output data i25-
When 4i is updated a predetermined number of times, it requests the partial system model correction unit 25-5 to re-learn using the input / output data i25-4i as teacher data. Partial system model correction unit 2
And re-learning based on 5-5 in input and output data I25-4i, update the partial system model f _i, with respect to inference unit correcting unit 25-2 instructs the correction of the "certainty" of the inference section 21.

Next, the operation of the system adjusting unit 25 will be described with reference to the flowcharts of FIGS. FIG. 12 shows the input / output data extraction / synthesis processing of the difference detection unit 23-1 and the input / output data correction unit 25-3, and FIG. 13 shows the parts of the partial system model correction unit 25-5 and the inference unit correction unit 25-2. This represents learning, modification of the system model, and modification processing of the inference unit.

First, the input / output data extraction / synthesis processing will be described.

[0104] When the operation start command from the arithmetic monitoring unit 26 is input, the response result y _o from the group management control inference result y and the group management control unit 1 of the time zone previously saved in the difference detection unit 25-1 difference the degree to _/ y _o ≧ D efo when satisfying (S21), the response result y _o of the time period, output the control parameter data correction unit 25-3, which is the basis of the correction amount | but | y-y _o the .delta.N _i (6.4) is calculated from the equation, and outputs the respective inference section modification portion 25-2 (S22).

Here, _Defo is a constant, which serves as a criterion for determining whether or not to modify. The input / output data correction unit 25-3 outputs the response result and control parameters suitable for the demand in the time zone to the input / output data i25-4i corresponding to the demand, and the output data and the input / output data i25-4i are similar. If there is such data, it is synthesized to update the input / output data i25-4i (S23).

Next, after the day of the week and the time zone to be calculated next are saved (S24), it is checked whether or not a predetermined number of time zones have been calculated (S25). When the calculated number of time zones does not satisfy the predetermined value, a calculation continuation request is output to the calculation monitoring unit 26 (S26), and when the calculated time zone number satisfies the predetermined value, a calculation completion signal is output to the calculation monitoring unit 26 (S2).
7).

Next, learning, modification of the partial system model, and modification of the inference unit will be described. The partial system model correction unit 25-5 determines whether the instruction from the operation monitoring unit 26 is operation interruption (whether the operation is other than operation start or operation continuation), and if not, the learning is started or continued.
If the instruction is an interruption instruction, the process shifts to updating or discarding the partial model based on the learning data (S31). In the partial system model 25-5, the learning state of the partial model such as the number of times of learning at the end of the previous operation and the inter-unit weight and bias of the partial model is read (S32). From input / output data i to T
The teacher data of the set is extracted and input to the partial model (S34). The error is calculated by the P method (S35),
This is repeated from the 0th set to the _Tmax set (S
33, S36). When the error calculation using all data of the input / output data i as the teacher data is completed, the learning number counter is incremented by 1 (S37), and the learning number at that time and the learning of the partial model such as the inter-unit weight and bias of the partial model are performed. Save the data representing the state (S3
8).

Whether the number of times of learning is equal to or more than a predetermined value,
Or an error is the learning end condition is less than the predetermined value is satisfied (S39), the old partial model f replacing _i new partial model f _i 'that learning is completed (S312) simultaneously with the inference unit correcting unit 25-2 The "certainty" of the inference unit 21
Issuing the correction, corrects the "certainty" of the inference unit 21 by the degree .delta.N _i which is the basis of the correction amount of the inference unit correction unit 25-2 in the differencing unit 25-1 (S313). Then, an operation completion signal is output to the operation monitoring unit 26 (S314). When the learning end condition is not satisfied, it is determined whether or not the command from the operation monitoring unit 26 is the interruption of the operation. When the instruction is not the operation interruption instruction, an operation continuation request instruction is output to the operation monitoring unit 26 (S311). If it is an interruption command, an operation completion signal is output to the operation monitoring unit 26 (S31).
4).

Next, the operation of the operation monitoring section 26 will be described with reference to the flowchart of FIG.

At a certain time, for example, at midnight (S51), the time management unit 26-2 starts a timer (S51).
52), difference detecting section 25-1 and input / output data correcting section 2
An operation start command is transmitted from the operation start / interruption command unit 26-1 to 5-3, and data extraction / synthesis processing is executed. Difference detection unit 25-1 and input / output data correction unit 25-
After performing the data extraction / synthesis processing in step 3, the time management unit 2
An operation completion signal or an operation continuation request is output to 6-2 (S53). When the signal is input to the time management unit 26-2, it is first checked whether or not a predetermined time has elapsed (S
54). If the predetermined time has elapsed, the processing is interrupted. In the next calculation, processing is performed from the next data that has been interrupted. If the predetermined time has not elapsed, the difference detection unit 25
-1 to input / output data correction unit 25-3 to time management unit 2
Based on the signal input to 6-2, it is checked whether or not the operations of the difference detection unit 25-1 to the input / output data correction unit 25-3 have been completed (S55). The interruption command unit 26-1 transmits a calculation continuation command to the difference detection unit 25-1 and the input / output data correction unit 25-3,
The data extraction / synthesis processing is executed again. If the calculation has been completed, it is checked whether the input / output data i has been updated a predetermined number of times (S56). If the input / output data i has been updated a predetermined number of times, it is further checked whether a predetermined time has elapsed. I do. If the predetermined time has not elapsed, a calculation start command is transmitted to the partial system model correction unit 25-5 and the inference unit correction unit 25-2, and learning of the partial system model, correction, and correction processing of the inference unit are executed ( S58). If the predetermined time has elapsed, the operation start / interruption command unit 26-1 transmits an operation interruption instruction to the difference detection unit 25-1 and the input / output data correction unit 25-3 (S510), and the difference detection unit 25- 1
Then, the processing at the time of interruption of the operation is performed in the input / output data correction unit 25-3 (S58). If the update has not been performed the predetermined number of times, the processing is interrupted.

As described in detail above, the learning control unit 1-1 includes an operation monitoring unit 26 added to the inference unit 21, the partial model unit 22, the synthesis unit 23, the inference result evaluation unit 24, and the system adjustment unit 25. By extracting and synthesizing effective traffic demand during operation of the system, optimal evaluation index weight values and group management control response result data, learning / updating of partial models using the data as teacher data, inference unit network In the correction processing, even if there is an excess of the processing time due to the increase of the data to be processed or traffic demand, the other task processing is not affected, and an autonomous system applied to each building by online learning. Construction is possible. In particular,
In the learning of the partial model, it is effective in terms of the reliability of the system to provide a time limit because there is a large variation in the calculation time until the learning end condition is satisfied.

In the above embodiment, the start time of the calculation is set to midnight.
By setting the time limit in a time zone where the demand is small, the load distribution of the CPU can be realized, and the reliability of the system is further improved.

Further, in the above embodiment, the learning result is discarded if the number of times of learning or the error is not satisfied when a predetermined time has elapsed in the learning process of the partial model, but the learning state is saved when the predetermined time has elapsed. Each time, the learning is continued from the state of being interrupted again in the next calculation, and the calculation is executed until the learning end condition is satisfied, whereby the efficiency of the process can be improved.

According to the present invention, in the above embodiment,
Effective traffic demand during system operation, extraction and synthesis of optimal evaluation index weights and group management control response result data, learning and updating of partial models using the data as teacher data,
Although the correction process of the inference unit net is performed sequentially, the same effect can be obtained by executing each process independently.

(Second Embodiment) In the first embodiment, the result data extraction / synthesis processing and the partial model learning / correction (correcting the connection state of the neural network according to the actual use state) processing are within the time limit. If the learning is not completed, the processing is interrupted. For the learning and correction of the partial model, whether the learning model at the time of the interruption satisfies a predetermined value is used to adjust the partial model by using the learning data, Had decided to discard.

In the second embodiment, when the result data extraction / synthesis processing and the partial model learning / correction (correction of the connection state of the neural network according to the actual use state) processing are not completed within the time limit, the processing is performed. Rather than interrupting, the processing level (task level) is lowered and the calculation is continued until a predetermined number of data or a predetermined condition is satisfied.

The operation of extracting and synthesizing the result data is the same as that of the first embodiment, and the operation of learning and modifying the partial system model and the modification of the inference unit are the same as those of the first embodiment except for the interruption processing. Therefore, here, the processing operation of the calculation monitoring unit 26 will be described based on the flowcharts of FIGS.

At a certain time, for example, at midnight (S71), the time management section 26-2 starts a timer (S71).
72), difference detection unit 25-1 and input / output data correction unit 2
An operation start command is transmitted from the operation start / interruption command unit 26-1 to 5-3, and the data extraction / synthesis processing unit is executed. Difference detection unit 25-1 and input / output data correction unit 25
After performing the data extraction / synthesis processing at -3, an operation completion signal or an operation continuation request is output to the time management unit 26-2 (S73). When the signal is input to the time management unit 26-2, it is first checked whether a predetermined time has elapsed (S74). If the predetermined time has elapsed, the processing level is lowered (command to lower the task level managed by the OS), and the calculation is continued (S7).
5). This low level state continues until the operation is completed. If the predetermined time has not elapsed, the difference detection unit 25
-1 to input / output data correction unit 25-3 to time management unit 2
Based on the signal input to 6-2, it is checked whether or not the operations of the difference detection unit 25-1 to the output data correction unit 25-3 are completed (S76). A command continuation command is transmitted from the command unit 26-1 to the difference detection unit 25-1 and the input / output data correction unit 25-3, and the data extraction / synthesis process is executed again. If the operation has been completed, it is determined whether the processing level has dropped (S77). If the processing level is low, the processing level is returned to the original processing level (the OS is returned to the original processing level). (S78). Next, it is checked whether or not the input / output data i has been updated a predetermined number of times (S79). If the input / output data i has been updated a predetermined number of times, it is further checked whether or not a predetermined time has elapsed (S710). If not, a calculation start command is transmitted to the partial system model correction unit 25-5 and the inference unit correction unit 25-2 to execute learning, correction of the partial system model, and correction processing of the inference unit (S
712). If the predetermined time has elapsed, the processing level is lowered (command to lower the task level managed by the OS), and the calculation is continued (S711).
This low level state continues until the operation is completed. Partial system model correction unit 25-5 and inference unit correction unit 25
-2 from the signal input to the time management unit 26-2,
Partial system model correction unit 25-5 to inference unit correction unit 2
It is checked whether the operation of 5-2 is completed (S71).
3) If not completed, start / interrupt command unit 2
6-1 sends an operation continuation command to the partial system model correction unit 25-5 and the inference unit correction unit 25-2 to execute the learning, correction of the partial system model, and the correction process of the inference unit again (S712). If the calculation has been completed, it is determined whether or not the processing level has dropped (S71).
4) If the level is low, the processing level is returned to the original processing level (the OS is instructed to return to the original processing level), and the processing is terminated (S715).

As described above, the effective traffic demand, the optimum evaluation index weight value and the group management control response result data are extracted and synthesized when the system of the second embodiment is operated, and the learning of the partial model using the data as the teacher data is performed. -In the update and inference unit net correction processing, even if the data to be processed or the processing time is exceeded due to the increase in traffic demand, lowering the task level and continuing the calculation will affect other task processing. Without affecting, online learning can be efficiently performed. Further, since the calculation is reliably continued until a predetermined condition is satisfied, the reliability of the system is improved.

[0120]

As described above, according to the present invention, the following functions and effects can be obtained. In other words, in the invention corresponding to claim 1, the time limit is given to the result data extracting means and the re-learning means for re-learning the extracted data as teacher data, and if the calculation is not completed within that time, the processing of the means is performed. The processing is interrupted, and it is determined whether or not to use the calculation result according to the processing status. Further, in the invention corresponding to claim 2, when the processing of the extracting means and the re-learning means does not end within the time limit, the processing levels of the two means are reduced so as not to affect other processing. It is intended to be continued. As a result, even if traffic demand or processing data fluctuates, it does not affect other tasks and extracts and outputs effective result data during system operation.
Processing such as synthesis and learning / updating of a partial model using the data as teacher data becomes possible, and furthermore, data processing can be performed reliably, leading to an improvement in the reliability of the system. By setting the time period during which the demand is small, it is possible to provide an elevator group management control device that leads to the load distribution of the CPU.

[Brief description of the drawings]

FIG. 1 shows a first embodiment of an elevator group management control device according to the present invention.
FIG.

FIG. 2 is a configuration diagram of a software system of a group management control unit and a single control unit in the embodiment of FIG. 1;

FIG. 3 is a system configuration diagram of a high-speed transmission line according to the embodiment of FIG. 1;

FIG. 4 is a functional block diagram showing a flow of input / output signals of a learning control unit of the embodiment of FIG. 1;

FIG. 5 is a system configuration diagram of a learning control unit according to the embodiment of FIG. 1;

FIG. 6 is a system configuration diagram of an inference unit in FIG. 5;

FIG. 7 is a system configuration diagram of a partial model unit of FIG. 5;

8 is a configuration diagram of each partial system model of the partial model of FIG. 7;

FIG. 9 is a system configuration diagram of an inference result evaluation unit in FIG. 5;

FIG. 10 is a flowchart showing the operation of the inference result evaluation unit in FIG. 9;

FIG. 11 is a system configuration diagram of a system adjustment unit in FIG. 5;

FIG. 12 is a flowchart showing the operation of the inference result evaluation unit in FIG. 5;

FIG. 13 is a flowchart showing the operation of the inference result evaluation unit in FIG. 5;

FIG. 14 is a system configuration diagram of an operation monitoring unit in FIG. 5;

FIG. 15 is a flowchart showing the operation of the operation monitoring unit in FIG. 14;

FIG. 16 is a flowchart showing the operation of the inference result evaluation unit of the second embodiment of the elevator group management control device of the present invention.

FIG. 17 is a flowchart showing the operation of the operation monitoring section of the second embodiment of the elevator group management control device of the present invention.

FIG. 18 is a flowchart showing the operation of the operation monitoring unit of the second embodiment of the elevator group management control device of the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Group management control part, 1-1 ... Learning control part, 2 ... Elevator group system, 2-1 to 2-N ... Single control part, 3 ... Hall call button, 4 ... Hall call input / output control part, 5 ... Monitoring panel, 6: high-speed transmission line, 7: low-speed transmission line, 21: inference unit,
Reference numeral 22: partial model unit, 23: synthesis unit, 24: inference result evaluation unit, 25: system adjustment unit, 26: operation monitoring unit.

Continuation of front page (58) Field surveyed (Int.Cl. ⁷ , DB name) B66B 1/18

Claims (2)

(57) [Claims] In order to assign a plurality of elevators to a plurality of floors and assign an optimum car to a generated common hall call by a predetermined evaluation operation, a neural network learning function is used for assignment control. In the elevator group management control device that applies and corrects the connection state of the neural network according to the actual use state after the building has been operated, the result data at regular intervals obtained in the actual use state is sequentially extracted as teacher data. and the teacher data extraction means, and sequentially performs retraining means relearning of the New <br/> Rarunetto by the teacher data extracted for each of the constant period, the extraction processing time in the teaching data extraction means exceeds a predetermined value
Is exceeded, the teacher data extraction means is interrupted.
The re-learning processing time in the re-learning means is set to a predetermined value.
An elevator group management control device comprising: interruption means for interrupting the re-learning means when the number exceeds the limit . 2. A method in which a plurality of elevators are put into service on a plurality of floors, and a learning function of a neural network is used for assignment control in allocating an optimum car to a generated common hall call by a predetermined evaluation operation. In the elevator group management control device that applies and corrects the connection state of the neural network according to the actual use state after the building has been operated, the result data at regular intervals obtained in the actual use state is sequentially extracted as teacher data. and the teacher data extraction means, and sequentially performs retraining means relearning of the New <br/> Rarunetto by the teacher data extracted for each of the constant period, the extraction processing time in the teaching data extraction means exceeds a predetermined value
Exceeds the processing level in the teacher data extraction means.
Re-learning by the re-learning means.
If the processing time exceeds a predetermined value, the processing by the
An elevator group management control device comprising: a processing level lowering unit that lowers a processing level .