JP2021080038A - Elevator control device, elevator control method, machine learning device, machine learning method, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an elevator control device, an elevator control method, a machine learning device, a machine learning method and a program capable of learning and determining an optimum elevator control mode by using a situation generated by a multi-agent simulation. An elevator control device 10 includes a user identification unit 13 that recognizes a user waiting for an elevator, and a usage status recording unit 15 that specifies at least the waiting time of the user as the usage status of the elevator. An external information acquisition unit 17 that acquires external information that affects traffic demand, a machine learning unit 20 that determines an optimum control mode based on at least the waiting time and the external information, and the optimum control mode. It has a car allocation unit 19 that controls the operation of the elevator based on the above. [Selection diagram] Fig. 1

Description

The present invention relates to an elevator control device, an elevator control method, a machine learning device, a machine learning method and a program, and more particularly to a method of learning an optimum control mode of an elevator.

Various elevator operation control methods have been proposed. SCAN is a basic control algorithm. In SCAN, when the car starts to move in a certain direction, it operates without changing the moving direction until all the requests (hall calls) generated on the floor that have not passed and match the moving direction of the car are satisfied. Similar algorithms include LOOK, C-SCAN, C-LOOK and the like.

When a plurality of elevators are operated by a relatively simple algorithm such as SCAN, a plurality of cars may eventually move to accompany each other (dumpling operation). There is group management as a control method that prevents dumpling operation and makes the waiting time on each floor uniform. In group management, when a hall call occurs on a floor, the controller assigns the most suitable car among the multiple cars to that floor. At this time, the control device predicts the traffic demand and selects an algorithm (control mode) for determining the optimum car based on the result.

Patent Document 1 describes that in group management, a traffic demand forecast is created based on an elevator user's behavior pattern. Patent Document 2 and Patent Document 3 describe that a camera provided at a landing recognizes a user's face, measures the waiting time for each user, and allocates a car so as to eliminate long waiting. Has been done.

JP-A-2017-030894 Japanese Unexamined Patent Publication No. 2019-023124 JP-A-2017-178475

However, although the control method as in Patent Document 1 can deal with typical behavior patterns such as going to work and going out at lunch, it is appropriate for traffic demand for sudden behavior due to meetings, events, disasters, etc. There is a problem that the prediction and control mode cannot be selected.

Therefore, it is conceivable to use the actual data that can be obtained by the techniques such as Patent Document 2 and Patent Document 3 so that the control mode suitable for the situation can be selected by a machine learning method such as reinforcement learning. However, learning using actual data requires a lot of time. Further, when machine learning is applied to a control method premised on personal identification as in Patent Document 1, it is inefficient because it is necessary to perform new learning every time the user of the building changes.

Therefore, the present invention provides an elevator control device, an elevator control method, a machine learning device, a machine learning method, and a program capable of learning and determining an optimum elevator control mode using a situation generated by a multi-agent simulation. With the goal.

One aspect of the present invention affects traffic demand with a usage identification unit that recognizes a user waiting for an elevator, a usage status recording unit that specifies at least the waiting time of the user as the usage status of the elevator, and a usage status recording unit that specifies at least the waiting time of the user. An external information acquisition unit that acquires external information to be given, a machine learning unit that determines an optimum control mode based on at least the waiting time and the external information, and an operation control of the elevator based on the optimum control mode. It is an elevator control device having a car allocation unit and a car allocation unit.
In another aspect of the present invention, the machine learning device obtains a situation data including the maximum value of the waiting time and a control mode as state variables by an elevator operation simulation by a multi-agent, and the simulation result. It has a determination unit that outputs determination data indicating the suitability of the above, and a learning unit that associates the situation data with the control mode by using the state variable and the determination data.
In another aspect of the present invention, the learning unit uses a reward calculation unit for obtaining a reward related to the determination data and the reward to update a value function indicating the value of the control mode in the situation data. It has a function update unit.
Another aspect of the present invention shows a simulator that acquires situation data including the maximum value of the waiting time of a user and a control mode as state variables by an elevator operation simulation by a multi-agent, and the suitability of the simulation result. It is a machine learning device having a determination unit that outputs determination data, and a learning unit that associates the situation data with the control mode by using the state variable and the determination data.
Another aspect of the present invention includes a usage identification step in which a computer recognizes a user waiting for an elevator, a usage status recording step for identifying at least the waiting time of the user as the usage status of the elevator, and traffic. An external information acquisition step for acquiring external information that affects demand, a determination step for determining an optimum control mode based on at least the waiting time and the external information, and an elevator based on the optimum control mode. It is a control method of an elevator having a car allocation step for performing operation control.
In another aspect of the present invention, a simulation step in which a computer acquires a situation data including a maximum value of a user's waiting time and a control mode as state variables by an elevator operation simulation by a multi-agent, and the simulation result. It is a machine learning method having a determination step of outputting determination data indicating suitability of the above, and a learning step of associating the situation data with the control mode using the state variable and the determination data.
Another aspect of the present invention is a program for causing a computer to perform the above method.

INDUSTRIAL APPLICABILITY According to the present invention, an elevator control device, an elevator control method, a machine learning device, a machine learning method, and a program capable of learning and determining an optimum elevator control mode using a situation generated by a multi-agent simulation are provided. be able to.

It is a schematic functional block diagram of the control device 10 of an elevator. It is a schematic functional block diagram of the machine learning unit 20. It is a schematic functional block diagram of the learning unit 207.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
FIG. 1 is a schematic functional block diagram of an elevator control device 10. The control device 10 includes landing cameras 11a to 11n provided on each floor (floor number n), in-car cameras 12a to 12m provided in the car (car number m), user identification unit 13, and usage status recording unit. It has 15, an external information acquisition unit 17, a machine learning device 20, and a car allocation unit 19. Each processing unit may be implemented as one function of a CPU (central processing unit), and may be realized by operating the CPU according to software.

The platform cameras 11a to 11n are installed at the elevator platform on each floor so that the faces of waiting users can be seen without exception. The images of the landing cameras 11a to 11n are output to the user identification unit 13.

The cameras 12a to 12m in the car are installed in the car of the elevator so that the user's face can be completely reflected. The images of the cameras 12a to 12m in the car are output to the user identification unit 13.

The user identification unit 13 extracts the feature amount of the face image of the user waiting at the landing on each floor from the images input from the landing cameras 11a to 11n. Further, from the images input from the in-car cameras 12a to 12m, the feature amount of the face image of the user who is in each car is extracted. The feature amount of the user's face obtained here is used as information for identifying the user.

The usage status recording unit 15 identifies on which floor a user got in the car and on which floor he / she got off. There are various known techniques for identifying the boarding / alighting floor. For example, the boarding floor is the stop floor when the user identification unit 13 recognizes that a user has framed in the images of the cameras 12a to 12m in the car. , The stop floor when it is recognized that the frame has been out can be used as the disembarkation floor. Alternatively, the floor on which the landing cameras 11a to 11n first recognized by the user may be set as the boarding floor, and the landing cameras 11a to 11n recognized by the user as the boarding floor may be used as the boarding floor. The usage status recording unit 15 records the determined boarding / alighting floor together with the time and the user's identifier.

In addition, the usage status recording unit 15 specifies the number of people waiting on each floor. Furthermore, the waiting time of each user is specified. There are various known techniques for specifying the waiting time. For example, when a user having any of the landing cameras 11a to 11n is currently captured, the user identification unit 13 uses the landing camera for the user. The elapsed time from the first recognized time in 11a to 11n to the present can be calculated and used as the waiting time.

The external information acquisition unit 17 acquires various external information that is useful for selecting a control mode, that is, that affects traffic demand. External information includes, for example, weather information that can be obtained from a distribution server via the Internet, attendance / leaving information (information about attendance time and leaving time) and event information (information about holding a meeting or event) that can be obtained from groupware, etc. , There is calendar information (information about dates, days of the week and holidays). The weather information may include information on the weather and temperature of each time zone in the location of the building and the surrounding area. The event information may include the floor where the event is scheduled to be held, the scheduled start and end dates, the event name, the organizer name, and the like. The external information acquisition unit 17 can acquire weather information, event information, calendar information, etc. for a predetermined period (for example, today's time, from the current time to 3 hours later, etc.) based on the current time.

The machine learning device 20 includes a trained model that determines an optimum control mode. The trained model includes the number of people waiting on each floor, the maximum waiting time (maximum waiting time) of users waiting on each floor, weather information, attendance / leaving information, event information, calendar information, and the optimum control mode. Corresponds to the model structure that represents the correlation of. That is, the machine learning device 20 inputs situation data S1 (waiting number of people on each floor, maximum waiting time on each floor, weather information, attendance / leaving information, event information, calendar information) indicating the current situation, and the trained model The optimum control mode S2 is output as a determination result according to the model structure.

The car allocation unit 19 controls the operation of the elevator using the control mode S2 determined by the machine learning device 20. That is, operation control (group control) is performed in the control mode S2, which is optimal for the current situation S1, and an appropriate car is assigned to the required floor. Since group control is a known technique, detailed description will not be given here.

Next, a method of generating a trained model in the machine learning device 20 will be described. As shown in FIG. 2, the machine learning device 20 in the learning mode includes a condition generation unit 201 and a simulator 203 for performing an elevator operation simulation by a multi-agent, a determination unit 205 for calculating the suitability determination result D of the elevator operation simulation, and situation data S1. It has a learning unit 207 that learns the relationship between the device and the control mode S2.

The condition generation unit 201 and the simulator 203 generate a state variable S (situation data S1 and control mode S2) by multi-agent simulation. Multi-agent simulation uses a mechanism in which the smallest unit (agent) that can autonomously make decisions and act determines its own actions while recognizing environmental information such as the existence information of other agents. It is a simulation technology. In the present embodiment, each agent determines his / her behavior according to the following constraints, for example.
・ The agent who arrives at the landing will register to call in the direction toward the target floor and enter the waiting state. ・ When the car arrives, the agent will board and get off at the target floor unless the passenger capacity is reached. ・ The car will be disembarked depending on the weather. The number of passengers that can be boarded varies (the number of passengers that can be boarded decreases in rain or low temperatures)
・ Agents move to the specified floor until the time of work other than holidays and after the time of leaving work ・ By the time the event starts, the specified agent moves to the floor where the event is held ・ In addition, a certain number of agents randomly move between floors Moving

The condition generation unit 201 randomly determines parameters (weather information, attendance / leaving information, event information, calendar information) that specifically define such constraint conditions. The condition generation unit 201 also randomly determines the control mode S2 of the elevator that affects the action result of the agent. Since the control mode S2 can be defined as a set of a plurality of control parameters, the condition generation unit 201 randomly determines a combination of control parameters.

The simulator 203 executes the multi-agent simulation on the premise of various conditions generated by the condition generation unit 201. Each agent acts autonomously while complying with the constraints defined by the parameters determined by the condition generation unit 201. Elevator operation control is performed according to a known group control algorithm, but control mode S2 is used. The simulator 203 measures the number of people waiting on each floor, which is the result of the agent's actions, and the maximum waiting time on each floor at regular intervals.

The simulator 203 records the measured number of people waiting on each floor and the maximum waiting time on each floor in a storage area (not shown) together with weather information, attendance / leaving information, event information, calendar information, and control mode S2. In this way, the state variables S (status data S1 and control mode S2) are collected.

The determination unit 205 calculates the suitability determination result D, which is an evaluation index of the simulation result by the condition generation unit 201 and the simulator 203. In the present embodiment, it is considered that the shorter the maximum waiting time of the user is, the better the control is, and the maximum value of the waiting time generated in the simulation (the maximum value in the trial of the maximum waiting time on each floor) is regarded as the judgment data D. To do.

The learning unit 207 learns the optimum control mode S2 according to an arbitrary machine learning algorithm. The learning unit 207 repeats learning using the state variable S and the determination data D by using the results of a plurality of simulations performed by the condition generation unit 201 and the simulator 203. By repeating the learning cycle, the learning unit 207 can gradually identify the correlation between the situation data S1 and the control mode S2 and approach the optimum solution.

The learning algorithm used by the learning unit 207 is not particularly limited, but in the present embodiment, an example of using reinforcement learning is shown. In reinforcement learning, the state indicating what the environment is now is s, the action that the agent can take is a, and the reward obtained when the agent takes action in a certain state is r, and the agent makes trial and error. The purpose is to maximize the state action value Q (Equation 1) when the action is repeated. Note that the agent referred to here is a virtual subject for searching for the optimum control mode S2 (consisting of a combination of a plurality of control parameters) in reinforcement learning, unlike the agent in the above-mentioned multi-agent simulation.

FIG. 3 is a diagram showing a configuration of a learning unit 207 when performing reinforcement learning. The learning unit 207 has a reward calculation unit 2071 that calculates the reward r for the action a in the state s, and a value function update unit 2073 that updates the function Q based on the reward r.

The reward calculation unit 2071 is positive when, for example, the determination data D under the state variable S is determined to be appropriate (for example, when the maximum value of the waiting time generated in the simulation is less than a predetermined threshold value). When it is determined that the reward r is inappropriate (for example, when the maximum value of the waiting time generated in the simulation exceeds a predetermined threshold value), a negative reward r is output. Here, the absolute values of the positive reward r and the negative reward r may be the same or different.

Alternatively, the reward calculation unit 2071 may calculate the reward r according to the value of the determination data D based on a predetermined evaluation function, evaluation table, or the like. For example, when D is less than the threshold value, the value of the positive reward r is increased as D becomes smaller, and when D exceeds the threshold value, the value of the negative reward r is increased as D becomes larger. An evaluation function, an evaluation table, or the like can be used. As a result, the reward r can be set more precisely.

The value function update unit 2073 uses a method such as Q-learning, Sarasa, or the Monte Carlo method to perform iterative trials (execution of the next action at + 1 in the state st brought about by the previous action at). Meanwhile, the function Q can be continuously updated based on the reward r. Since these methods are known, specific description thereof will be omitted here.

That is, the learning unit 207 updates the function Q by repeatedly executing the elevator operation control simulation. This process can be carried out, for example, by the following procedure. One trial here is, for example, the execution of an elevator operation control simulation over a predetermined time.
(1) In the first trial, a certain parameter of the control mode S2 is randomly determined as the action a under the condition given by the condition generation unit 201, and the simulator 203 performs the simulation. The determination unit 205 outputs D as a trial result. The reward calculation unit 2071 calculates the reward r based on D, and the value function update unit 2073 updates the function Q based on r. (2) In the next trial, as the next action a, a certain parameter in the control mode S2 is changed according to a predetermined rule, and the function Q is updated.
(3) The same trial as in (2) above is repeated a certain number of times.
(4) The state is reset to the state of (1) above, and the set of (2) to (3) above is repeated a certain number of times.
(5) While randomly changing the conditions given by the condition generation unit 201, the above sets (1) to (4) are repeated a certain number of times.

Although the embodiments of the present invention have been described above, the present invention is not limited to the above embodiments and can be appropriately modified without departing from the spirit. Within the scope of the present invention, it is possible to modify any component of the embodiment or omit any component of the embodiment.

For example, the machine learning method shown in the above-described embodiment is only an example, and if the effect of learning the relationship between the situation data S1 and the control mode S2 can be exhibited, another machine learning method can be used instead. Can be done. For example, in reinforcement learning, a method using a neural network can be used. You may also use other machine learning methods, including supervised learning.

Further, the information processing of the present invention may be realized by hardware, or may be realized by the CPU executing a computer program. The computer program may be supplied to the computer by various types of non-transitory computer readable medium or temporary computer readable medium.

10 Elevator control device 11a to 11n Landing camera 12a to 12m In-car camera 13 User identification unit 15 Usage status recording unit 17 External information acquisition unit 19 Car allocation unit 20 Machine learning device 201 Condition generation unit 203 Simulator 205 Judgment unit 207 Learning Department 2071 Reward calculation department 2073 Value function update department

Claims (7)

A user identification unit that recognizes users waiting for the elevator,
As the usage status of the elevator, at least a usage status recording unit that specifies the waiting time of the user, and
The external information acquisition department that acquires external information that affects traffic demand,
A machine learning unit that determines the optimum control mode based on at least the waiting time and the external information.
An elevator control device having a car allocation unit that controls the operation of the elevator based on the optimum control mode. The machine learning device
A simulator that acquires the status data including the maximum value of the waiting time and the control mode as state variables by the elevator operation simulation by the multi-agent.
A judgment unit that outputs judgment data indicating the suitability of the simulation result, and a judgment unit.
The elevator control device according to claim 1, further comprising a learning unit that associates the situation data with the control mode using the state variable and the determination data. The learning unit
A reward calculation unit that obtains rewards related to the judgment data, and
The control device for an elevator according to claim 2, further comprising a value function update unit that updates a value function indicating the value of the control mode in the situation data using the reward. A simulator that acquires status data including the maximum value of the user's waiting time and control mode as state variables by simulating elevator operation by a multi-agent system.
A judgment unit that outputs judgment data indicating the suitability of the simulation result, and a judgment unit.
A machine learning device having a learning unit that associates the situation data with the control mode using the state variable and the determination data. The computer
A user identification step that recognizes a user waiting for an elevator,
As the usage status of the elevator, at least a usage status recording step for specifying the waiting time of the user and
External information acquisition step to acquire external information that affects traffic demand,
A determination step for determining the optimum control mode based on at least the waiting time and the external information.
An elevator control method comprising a car allocation step for controlling the operation of the elevator based on the optimum control mode. The computer
A simulation step to acquire the status data including the maximum value of the waiting time of the user and the control mode as state variables by the elevator operation simulation by the multi-agent.
A judgment step that outputs judgment data indicating the suitability of the simulation result, and
A machine learning method having a learning step of associating the situation data with the control mode using the state variable and the determination data. A program for causing a computer to perform the method according to claim 5 or 6.

Images