JP7294480B1 - Learning device, reasoning device and elevator group control device - Google Patents

Abstract

A learning device, an inference device, and an elevator group management device capable of obtaining a control mode capable of reducing the waiting time of passengers in an elevator where conditions such as crowded time zones are not known in advance are provided. The learning device is traffic flow information in a time zone that satisfies a prescribed condition, out of traffic flow information indicating the number of elevator passengers getting on or off one of a plurality of cars on each floor. A learning data acquisition unit that acquires learning data including information on a certain partial traffic flow, and learning for inferring the control mode of a group control device that controls a plurality of cars from the traffic flow information using the learning data. and a model generation unit that generates the finished model by unsupervised learning. [Selection drawing] Fig. 2

Description

The present disclosure relates to a learning device, a reasoning device, and an elevator group control device.

Patent Literature 1 discloses a group control device for elevators. According to the group control apparatus, characteristics of the traffic flow can be known by inputting the traffic flow indicating the number of passengers getting on and off the elevator to a learning device using a neural network. The group manager may perform control based on characteristics of traffic flow.

JP-A-8-217343

However, in the group management device described in Patent Document 1, the learning device is created by so-called supervised learning, which is learning in which the time zone of input data is set in advance. For this reason, even if it is applied to an elevator in which conditions such as busy time zones are not known in advance, there is a possibility that the waiting time of passengers cannot be reduced.

The present disclosure has been made to solve the above problems. An object of the present disclosure is to provide a learning device, an inference device, and an elevator group management device that can obtain a control mode that can reduce the waiting time of passengers in an elevator where conditions such as busy time zones are not known in advance. is.

The learning device according to the present disclosure calculates the number of passenger occurrences for one day based on traffic flow information indicating the number of elevator passengers who got on or got off from one of a plurality of cars on each floor, Based on the number of passenger occurrences, a time zone that satisfies a specified condition and is recognized as a congested state is determined, and learning data including information on partial traffic flow, which is the information on the traffic flow in the time zone, is acquired. A learning data acquisition unit, and model generation for generating a trained model for inferring a control mode of a group control device for controlling the plurality of cars from the traffic flow information using the learning data by unsupervised learning. and

Further, the inference device according to the present disclosure includes an inference data acquisition unit that acquires traffic flow information indicating the number of elevator passengers getting into or out of one of a plurality of cars on each floor; Based on the information, calculate the number of passengers for one day, based on the number of passengers , determine whether the prescribed conditions are met and the congestion state is recognized, and use the traffic flow information for the time period A model generated by unsupervised learning using information on a certain partial traffic flow as learning data, the model for inferring the control mode of a group control device that controls the plurality of cars from the information on the traffic flow. an inference unit that outputs the control mode of the group control device that controls the plurality of cars from the traffic flow acquired by the inference data acquisition unit, using the estimated model.

The elevator group tube device according to the present disclosure includes a traffic flow creation unit that creates a traffic flow indicating the number of elevator passengers who got on or got off from one of a plurality of cars on each floor, and the information on the traffic flow. Based on the above, the number of passengers generated per day is calculated, based on the number of passengers generated, the time zone that satisfies the prescribed conditions and is recognized as congested is determined, and the traffic flow in that time zone is extracted. A learning data acquisition unit that creates information on partial traffic flow, and unsupervised learning based on the information on the partial traffic flow created by the learning data acquisition unit, thereby determining traffic flow as one of a plurality of traffic patterns. a model generation unit that generates a learned model for classification; a control mode creation unit that determines a plurality of control modes respectively corresponding to the plurality of traffic patterns shown in the learned model; By applying to, a pattern determination unit that determines a traffic pattern in which the traffic flow is classified among the plurality of traffic patterns, and the pattern determination unit among the plurality of control modes created by the control mode creation unit a control mode selector that outputs a control mode corresponding to the determined traffic pattern, and controls operation of the plurality of cars based on the output control mode.

According to the present disclosure, the control mode of the group supervisor can be output based on the trained model that has undergone unsupervised learning. Therefore, it is possible to obtain a control mode capable of reducing the waiting time of passengers in an elevator in which conditions such as busy time zones are not known in advance.

1 is a configuration diagram of a building to which an elevator group control device according to Embodiment 1 is applied; FIG. 1 is a block diagram of an elevator group control device according to Embodiment 1. FIG. FIG. 4 is a diagram showing the transition of the number of generated passengers calculated by the traffic flow creation unit of the elevator group control device according to Embodiment 1; 4 is a flowchart for explaining an overview of learning processing in which the elevator group control device according to Embodiment 1 learns a learned model. 4 is a flow chart for explaining an outline of processing in which the learning device of the elevator group control device according to Embodiment 1 performs learning. 4 is a flow chart for explaining an overview of processing performed by the elevator group control device according to Embodiment 1. FIG. 2 is a hardware configuration diagram of an elevator group control device according to Embodiment 1. FIG. FIG. 10 is a block diagram of an elevator group control device according to Embodiment 2; 10 is a flowchart for explaining an outline of processing for learning a learned model by an elevator group management device according to Embodiment 2;

Embodiments for carrying out the present disclosure will be described with reference to the accompanying drawings. In addition, the same code|symbol is attached|subjected to the part which is the same or corresponds in each figure. Redundant description of the relevant part will be simplified or omitted as appropriate.

Embodiment 1.
FIG. 1 is a block diagram of a building to which an elevator group control system according to Embodiment 1 is applied.

The elevator in FIG. 1 is installed in a building 1 . Each of the plurality of hoistways 2 penetrates each floor of the building 1 . The machine room 3 is provided directly above the plurality of hoistways 2 . A plurality of landings 4 are provided on each floor of the building 1, respectively.

A plurality of cars 5 are provided inside the plurality of hoistways 2, respectively. A weighing device 6 is provided in each of the plurality of cages 5 .

A plurality of call registration devices 7 are provided in a plurality of halls 4, respectively. A plurality of control devices 8 are provided in the machine room 3 . A plurality of control devices 8 correspond to a plurality of cages 5, respectively. A plurality of controllers 8 control movement of the corresponding cars 5 .

A group management device 9 is provided in the machine room 3 . The group control device 9 controls the operation of the plurality of cars 5 by transmitting commands to the plurality of control devices 8 .

A passenger operates a call registration device 7 at a hall 4 when moving in an elevator. The call registration device 7 transmits a call to the hall 4 to the group control device 9 . The group control device 9 determines the car 5 to which the call to the landing 4 is to be assigned based on the operation status of the plurality of cars 5 . The group control device 9 transmits a call to the hall 4 to the control device 8 corresponding to the determined car 5 . The control device 8 moves the car 5 to the landing 4 when receiving the call to the landing 4 .

When a passenger gets on the car 5 , the weighing device 6 detects a change in weight inside the car 5 . The weighing device 6 transmits information indicating the detected change in weight to the group control device 9 via the control device 8 .

The group control device 9 calculates the number of people getting into the car 5 and the number of people getting off the car 5 at a certain time based on the information from the weighing device 6 . The group control device 9 creates and stores traffic flow information indicating the number of people getting in and out of the car 5 at each of the plurality of halls 4 . For example, the group control device 9 creates traffic flow information every five minutes. For example, the group control device 9 creates information in which a plurality of pieces of traffic flow information created every five minutes are summarized into one day's worth of traffic flow information, and stores the information in itself.

For example, when an event is held on a specific floor of the building 1, the traffic volume, which is the number of people moving in the elevator, increases. At this time, the group control device 9 creates traffic flow information that reflects the increased traffic volume. Based on the created traffic flow information, the group control device 9 determines a traffic pattern that indicates the trend of the traffic flow. The group control device 9 selects a control mode suitable for the determined traffic pattern, and controls operation of the plurality of cars 5 based on the control mode. Specifically, for example, the group control device 9 sets three of the plurality of cars 5 as cars that stop only at the floor where the event is held and the entrance floor based on the control mode. In this case, the group control device 9 sets the other one of the plurality of cars 5 as a car that does not stop at the floor where the event is held.

The group controller 9 selects and models models for judging traffic patterns based on stored traffic flow information when arbitrary conditions such as when a command to learn is received, when a prescribed period of time has passed, etc. are satisfied. Each control mode is created by machine learning.

Next, the group management device 9 will be explained using FIG.
FIG. 2 is a block diagram of an elevator group control apparatus according to the first embodiment. 2, illustration of the car 5 and the weighing device 6 is omitted.

As shown in FIG. 2, the group management device 9 includes a traffic flow storage unit 10, a traffic flow creation unit 11, a learning device 12, and an inference device 13.

The traffic flow storage unit 10 stores information on traffic flow. For example, the traffic flow storage unit 10 stores traffic flow information for a plurality of days with one day's worth of traffic flow information as one unit. The traffic flow storage unit 10 stores the latest traffic flow information starting from a specified time. Table 1 below shows an example of traffic flow information.

The traffic flow information is information in which the number of passengers and the number of alighters on each floor at each time are associated with each other. Table 1 shows traffic flow information summarized for one day. For example, "3 people" in the column "number of people getting off on the 1st floor" in the row where the "time" is "23:57" indicates that 3 people got off from the car 5 on the 1st floor at the time 23:57. indicates that

The traffic flow generation unit 11 collects information related to elevator control such as information from a plurality of control devices 8 . The traffic flow creation unit 11 creates traffic flow information based on the collected information. The traffic flow creation unit 11 stores the created traffic flow information in the traffic flow storage unit 10 . At this time, the traffic flow creation unit 11 adds the newly created traffic flow information to the traffic flow information already stored in the traffic flow storage unit 10, so that the traffic flow information stored in the traffic flow storage unit 10 is changed. The information may be updated with the latest traffic flow information.

Specifically, the traffic flow generator 11 calculates the traffic flow based on information indicating the current position of the car 5, the running state of the car 5, and changes in the weight inside the car 5. FIG. When the weight inside the car 5 increases while the car 5 is stopped, the traffic flow generation unit 11 determines that a passenger has boarded the car 5 . At this time, the traffic flow generation unit 11 estimates the number of passengers who boarded the vehicle. The traffic flow generation unit 11 divides the increased weight by a preset average weight value to estimate the number of passengers on board. Similarly, the traffic flow generator 11 estimates the number of passengers who have exited the car 5 when the weight inside the car 5 has decreased.

The learning device 12 includes a learning data acquisition unit 14 , a model storage unit 15 , a model generation unit 16 , a control mode storage unit 17 and a control mode creation unit 18 .

The learning data acquisition unit 14 acquires learning data, which is information to be learned by a learning model for determining traffic flow patterns. The learning data includes information on multiple partial traffic flows. For example, the learning data acquisition unit 14 creates learning data by creating partial traffic flow information from the traffic flow information stored in the traffic storage unit 10 as an operation for acquiring learning data. First, the learning data acquisition unit 14 acquires traffic flow information from the traffic flow storage unit 10 . The learning data acquisition unit 14 extracts information on partial traffic flow, which is suitable for learning, from the acquired information on traffic flow. For example, the learning data acquisition unit 14 extracts the traffic flow during the time when the elevator is in a congested state from the traffic flow for one day, and uses it as the partial traffic flow information. The learning data acquisition unit 14 creates information on a plurality of partial traffic flows.

When extracting the partial traffic flow, the learning data acquisition unit 14 calculates the transition of the number of passengers generated for one day based on the traffic flow for one day. The learning data acquisition unit 14 calculates the number of passengers at a certain time by adding the number of passengers in a specified time range. For example, when calculating the number of passengers at 23:57, the learning data acquisition unit 14 adds the number of passengers on all floors at each time from 23:48 to 23:57. The learning data acquisition unit 14 calculates the number of passenger occurrences for one day, which indicates the number of passenger occurrences at each hour of the day.

The learning data acquisition unit 14 determines a time period during which the train is congested, based on the number of passenger occurrences for one day. A criterion for judging a time zone in which the learning data acquiring unit 14 is congested is set arbitrarily.

In the first example of the criterion, the learning data acquisition unit 14 determines that a time zone in which the number of passengers is equal to or greater than a preset threshold is a time zone in a congested state. For example, the threshold is set to a value obtained by multiplying the number of people living in the building in which the elevator is installed by a prescribed percentage.

In the second example of the criterion, the learning data acquisition unit 14 uses the amount of increase in the moving average value of the number of passenger occurrences as the determination criterion. Specifically, the learning data acquiring unit 14 determines that the time when the increase in the moving average value of the number of passenger occurrences per 5 minutes becomes larger than the prescribed increase threshold is the start time of the congestion state. The learning data acquisition unit 14 determines that the time at which the number of passengers becomes less than or equal to the moving average value of the number of passengers at the start time is the end time of the congestion state. The learning data acquisition unit 14 determines that the period from the start time to the end time is the time period in which the traffic is congested.

In the third example of the criterion, the learning data acquiring unit 14 determines that the time when the short-term moving average value of the number of passenger occurrences exceeds the long-term moving average value is the start time of the congestion state. The learning data acquisition unit 14 determines that the time when both the short-term moving average value and the long-term moving average value of the number of passenger occurrences decrease is the end time of the congestion state. The time of interest in the short-term moving average is shorter than the time of interest in the long-term moving average.

The model storage unit 15 stores information on learned models. A trained model is a model that classifies a traffic flow or a partial traffic flow, which is input data, into one of a plurality of traffic patterns.

The model generation unit 16 generates a trained model by performing unsupervised learning using the learning data acquired by the learning data acquisition unit 14 . Any suitable method is selected for the unsupervised learning method. For example, the model generation unit 16 performs unsupervised learning based on the k-means method, which is a clustering method. The model generation unit 16 generates a learned model at arbitrary timing, such as when a learning command is received or when a set period elapses. The model generation unit 16 stores information on the generated trained model in the model storage unit 15 . When information is already stored in the model storage unit 15 , the model generation unit 16 updates the information of the learned model stored in the model storage unit 15 .

The control mode storage unit 17 stores information on a plurality of control modes. A control mode is a condition for controlling the car 5 when the car 5 is operated. A plurality of control modes respectively correspond to a plurality of traffic patterns shown in the learned model. For example, the control mode storage unit 17 stores information in which a plurality of control modes are summarized. Table 2 below provides an example of information for multiple control modes.

In Table 2, "traffic pattern No." indicates identification information of a plurality of traffic patterns. The "whether or not to dispatch car" indicates whether or not it is necessary to dispatch the car 5 to which no call is assigned to the set floor. The "delivery floor" indicates the floor to which the car 5 is delivered. "Distributed number of cars" indicates the number of cars 5 to be distributed. The control mode may be associated with "traffic pattern No." information that limits the floors where the car 5 stops.

For example, the second row of Table 2 indicates No. 1 shows a control mode corresponding to one traffic pattern. No. In the control mode corresponding to one traffic pattern, the group control device 9 always controls three cars 5 to be dispatched to the first floor.

The control mode creation unit 18 creates information on a plurality of control modes using a machine learning technique. Specifically, first, the control mode creation unit 18 acquires information on the learned model from the model storage unit 15 . The control mode creation unit 18 creates control modes corresponding to each of the plurality of traffic patterns shown in the learned model.

At this time, the control mode creating unit 18 may create the control mode using a so-called reinforcement learning method. For example, the control mode generator 18 sets a feature quantity of traffic flow indicated by a certain traffic pattern. The control mode creation unit 18 sets an arbitrary control mode as a provisional control mode for the feature amount of the traffic flow. The control mode creation unit 18 sets the conditions of a traffic simulation for transporting passengers based on the feature amount of the traffic flow and the provisional control mode that is the set provisional control mode. The control mode creation unit 18 performs a traffic simulation in which a virtually generated call is repeated under set conditions to transport passengers, and calculates an evaluation value of the traffic simulation. For example, the control mode creating unit 18 calculates the average waiting time of passengers in the traffic simulation as an evaluation value.

After that, the control mode creating unit 18 repeats the traffic simulation under the conditions in which the temporary control mode is changed for the same feature quantity. Specifically, after calculating the evaluation value, the control mode generator 18 sets another condition in which the temporary control mode is changed to another temporary control mode while the feature quantity of the traffic flow remains constant. The control mode creation unit 18 calculates similar traffic simulation evaluation values under different conditions.

After repeating the traffic simulation a specified number of times, the control mode creating unit 18 selects the traffic simulation with the best evaluation value. For example, the control mode creation unit 18 selects the traffic simulation with the smallest calculated evaluation value. The control mode creation unit 18 determines the temporary control mode in the traffic simulation having the best evaluation value as the control mode corresponding to the traffic pattern, and creates information on the control mode. Note that the control mode creation unit 18 may perform traffic simulations for all control modes and select the traffic simulation with the best evaluation value.

Note that the control mode creation unit 18 determines, as the control mode corresponding to the traffic pattern, not the traffic simulation with the best evaluation value, but the control mode in the traffic simulation in which the evaluation value is equal to or greater than a predetermined score. good too.

The inference device 13 includes an inference data acquisition unit 19 and an inference unit 20 .

The inference data acquisition unit 19 acquires traffic flow information as data for inference from the traffic flow storage unit 10 at arbitrary timing. At this time, the inference data acquisition unit 19 may acquire the latest traffic flow information stored in the traffic flow storage unit 10 . In addition, the inference data acquisition unit 19 may acquire information on traffic flow at the latest time.

The inference section 20 includes a pattern determination section 21 and a control mode selection section 22 .

The pattern determination unit 21 acquires information on learned models stored in the model storage unit 15 . The pattern determination unit 21 determines the current traffic pattern of traffic flow by applying the traffic flow information acquired by the inference data acquisition unit 19 to the learned model. For example, the pattern determination unit 21 determines the current traffic pattern of the traffic flow every hour.

When the pattern determination unit 21 determines the traffic pattern of the current traffic flow, the control mode selection unit 22 selects the traffic pattern determined by the pattern determination unit 21 from among the plurality of control modes stored in the control mode storage unit 17. Select the corresponding control mode.

The control mode selection unit 22 applies the selected control mode to the plurality of control devices 8 as the output of the control mode, thereby controlling the operation of the plurality of cars 5 . Note that the control mode selector 22 may simply output information on the control mode. In this case, the group control device 9 may control the operation of the plurality of cars 5 based on the control mode information output from the control mode selection section 22 .

As described above, the inference unit 20 outputs the control mode corresponding to the traffic flow by using the learning model for the traffic flow information.

For example, the group management device 9 uses the average value of each value of traffic flow as a feature quantity of traffic flow indicated by the traffic pattern in the learned model. At this time, an example of the learning process in which the group control device 9 performs learning based on the k-means method will be described below.

In the k-means-based learning example, it is assumed that a plurality of traffic flow information is used, including the traffic flow information shown in Table 1. In this case, 18 values representing the number of passengers and the number of passengers getting off on the 1st to 9th floors are selected as feature quantities. The feature quantity of the partial traffic flow is a value obtained by averaging the feature quantity at each time. For example, a partial traffic flow is represented by an 18-dimensional vector having each feature amount as a component. In the following, mean of partial flows means mean vectors of partial flows.

When performing learning, the model generator 16 sets the number K of traffic patterns as an initial value. Specifically, the model generation unit 16 may set the number of assumed congestion states as an initial value. The model generator 16 may also determine the initial values based on installation conditions such as the number of cars 5 provided in the building 1 and the number of floors of the building 1 .

After that, the model generator 16 randomly assigns one of the plurality of traffic patterns to each of the plurality of partial traffic flows. At this time, the traffic pattern is represented by any number from 1 to K.

After that, the model generator 16 calculates the center of gravity of the traffic pattern for each of the plurality of traffic patterns. The centroid of a traffic pattern is equal to the average of the partial flows belonging to that traffic pattern.

After that, the model generator 16 calculates distances from the center of gravity of a plurality of traffic patterns for a certain partial traffic flow. For example, the distance from the center of gravity is the Euclidean distance between the point indicating the partial traffic flow and the center of gravity of the traffic pattern in an 18-dimensional space. The model generator 16 identifies the center of gravity that is closest to the partial traffic flow among the centers of gravity of the plurality of traffic patterns. The model generator 16 reassigns the traffic pattern of the partial traffic flow to the traffic pattern closest to the center of gravity. The model generation unit 16 performs calculations to reassign traffic patterns for all partial traffic flows.

After that, the model generating unit 16 repeats the same calculation of calculating the center of gravity of a plurality of traffic patterns and reassigning the traffic patterns in a state where the traffic patterns have been reassigned.

When the allocation of traffic patterns does not change, the model generation unit 16 determines whether or not there is a traffic pattern in which the maximum distance between the partial traffic flow belonging to the traffic pattern and the center of gravity of the traffic pattern is equal to or greater than a specified threshold. determine whether

If there is a traffic pattern whose maximum distance value is equal to or greater than the specified threshold, the model generator 16 increases the number of traffic patterns by one from the initial value, and repeats the subsequent calculations.

If there is no traffic pattern that is equal to or greater than the specified threshold, that is, if the maximum value of the distance is less than the specified threshold for all traffic patterns, the model generator 16 terminates learning. When the learning is finished, the model generating unit 16 creates information in which the identification information of the plurality of traffic patterns and the centroids of the plurality of traffic patterns are associated with each other as the information of the learned model. Table 3 below shows an example of the information of the trained model.

The control mode creation unit 18 creates a control mode corresponding to the traffic pattern based on the condition indicated by the center of gravity of the traffic pattern, which is a characteristic quantity. For example, in a traffic simulation for transporting passengers, the control mode creation unit 18 calculates the average waiting time of passengers by performing calculations such as creating an OD table (Origin Destination table) shown in Table 4 below. .

The pattern determination unit 21 calculates the average of the current traffic flow when determining the traffic pattern. The pattern determination unit 21 calculates the Euclidean distance between the average of the current traffic flow and each of the centroids of the plurality of traffic patterns. The pattern determination unit 21 determines the traffic pattern having the center of gravity with the closest Euclidean distance as the traffic pattern of the current traffic flow.

By performing such processing, the group control device 9 can select and output an appropriate control mode according to the type of congestion that has occurred. For example, when boarding congestion occurs in which the number of passengers on the main floor is large, the group control device 9 performs control to allocate the car 5 to the main floor in advance, control to limit the floors where the car 5 stops, Select a control mode such as Further, for example, when congestion occurs on the main floor where many passengers get off and the car 5 stays on the main floor for a long time, the group control device 9 moves the car 5 to the farthest floor from the main floor in advance. or control to allocate the call to the car 5 so that the stop floor of the car 5 is as close to the floor as possible. Therefore, the waiting time of passengers can be reduced.

The type of congestion need not be set in advance. For example, a flexible situation that fits the current situation rather than a typical situation set based on human experience can be considered.

Next, the number of passengers generated will be described with reference to FIG.
FIG. 3 is a diagram showing changes in the number of generated passengers calculated by the traffic flow generation unit of the elevator group control device according to the first embodiment.

FIG. 3 shows a graph representing the number of passenger occurrences per day. The horizontal axis of the graph is time. For example, the horizontal axis is the time from the opening time of the building 1 to the closing time. The vertical axis of the graph is the number of passengers at each time. For example, the number of passenger occurrences is calculated every minute.

As shown in FIG. 3, the number of passenger occurrences is 0 at the opening time. The number of passengers generated gradually increases from 0. After that, the number of passenger occurrences repeatedly increases or decreases, and then gradually decreases toward zero. The number of passengers generated becomes 0 at the closing time.

For example, in the transition of the number of passengers generated in FIG. The occurrence of congestion greatly affects the waiting time of passengers. For this reason, the group management device 9 applies a special control mode during the time period when the traffic is congested.

On the other hand, in the actual operation of elevators, the ratio of busy hours in a day is not high. That is, most of the day is in a quiet state with few passengers. Traffic flow information that is quiet most of the time is regarded as unevenly distributed traffic flow information. When unsupervised learning of a trained model is performed based on information on traffic flows with an imbalanced distribution, there is a risk that originally congested traffic flows will be classified into the same pattern as quiet traffic flows. For this reason, information on traffic flow with imbalanced distribution is not effective as information used for unsupervised learning.

The learning data acquisition unit 14 extracts the congested time zones from the traffic flow information for one day, excluding the quiet time zones, and creates partial traffic flow information. At this time, the learning data acquisition unit 14 counts a group of several consecutive traffic flows as one state. For example, as shown in FIG. 3, the learning data acquisition unit 14 extracts quiet time periods Q1 and Q2 and congested time periods C1, C2, C3, C4, C5, C6, and C7. . The learning data acquisition unit 14 extracts the traffic flows in the congested time periods C1 to C7, respectively.

By extracting the traffic flow during the congested time period and acquiring it as learning data, the model generation unit 16 can perform unsupervised learning based on information that is as effective as possible. As a result, different control modes can be created for traffic patterns that indicate congested conditions and traffic patterns that indicate non-congested conditions.

Next, an example of learning processing in which a learned model is learned will be described with reference to FIG.
FIG. 4 is a flow chart for explaining an outline of learning processing for learning a learned model by the elevator group control device according to the first embodiment.

The processing shown in FIG. 4 is learning processing based on the k-means method. The group control device 9 starts the learning process at an arbitrary timing, such as when receiving a learning instruction or when a specified period has passed since the previous learning process.

In step S101, the group controller 9 initializes the number K of traffic patterns by setting the number K of traffic patterns to an initial value.

After that, the operation of step S102 is performed. In step S102, the group control device 9 randomly assigns one of the plurality of traffic patterns to each of the plurality of partial traffic flows.

After that, the operation of step S103 is performed. In step S103, the group control device 9 calculates the center of gravity of the traffic pattern.

After that, the operation of step S104 is performed. In step S104, the group control device 9 reassigns the traffic pattern of the partial traffic flow.

After that, the operation of step S105 is performed. In step S105, the group control device 9 determines whether or not the traffic pattern allocation has changed in any of the partial traffic flows.

If it is determined in step S105 that the allocation of traffic patterns has changed in any of the partial traffic flows, the group control device 9 performs operations from step S103 onward.

If it is determined in step S105 that the allocation of traffic patterns has not changed in any of the partial traffic flows, the group control device 9 performs the operation of step S106. In step S106, the group control device 9 determines whether or not there is a traffic flow in which the maximum value of the distance between the partial traffic flow belonging to the traffic pattern and the center of gravity of the traffic pattern is equal to or greater than a prescribed threshold value.

If it is determined in step S106 that there is no traffic pattern in which the maximum value of the distance is equal to or greater than the specified threshold value, the group control device 9 terminates the learning process.

If it is determined in step S106 that there is a traffic pattern in which the maximum value of the distance is equal to or greater than the specified threshold value, the group control device 9 performs the operation of step S107. In step S107, the group control device 9 sets the number K of traffic patterns to a number incremented by 1 from the current number. After that, the group management device 9 repeats the operations after step S102.

Next, the processing when the learning device 12 performs learning will be described with reference to FIG.
FIG. 5 is a flow chart for explaining an outline of processing for learning by the learning device of the elevator group control device according to the first embodiment.

As shown in FIG. 5, the learning device 12 acquires learning data from the traffic flow storage unit 10 in step S201.

After that, the operation of step S202 is performed. In step S202, the learning device 12 executes learning processing.

After that, the operation of step S203 is performed. In step S<b>203 , the learning device 12 stores the learned model generated in the learning process in the model storage unit 15 .

After that, the learning device 12 ends the process.

Next, with reference to FIG. 6, the process when the group control device 9 infers the control mode based on the traffic flow information and performs control according to the control mode will be described.
FIG. 6 is a flow chart for explaining an outline of processing performed by the elevator group control device according to the first embodiment.

As shown in FIG. 6, in step S301, the inference device 13 acquires traffic flow information from the traffic flow storage unit 10 as data for inference.

After that, the operation of step S302 is performed. In step S302, the inference device 13 inputs traffic flow information to the learned model.

After that, the operation of step S303 is performed. In step S303, the inference device 13 infers a traffic pattern corresponding to the input traffic flow.

After that, the operation of step S304 is performed. In step S304, the inference device 13 selects and outputs a control mode corresponding to the inferred traffic pattern.

After that, the operation of step S305 is performed. At step S305, the group control device 9 controls operation of the plurality of cars 5 based on the control mode output at step S304.

After that, the inference device 13 terminates its operation.

According to the first embodiment described above, the group control device 9 includes the traffic flow generation unit 11, the learning data acquisition unit 14, the model generation unit 16, the control mode generation unit 18, the pattern determination unit 21, and the control mode selection unit. 22. A group control device 9 extracts a partial traffic flow from the traffic flow. The group controller 9 creates a learned model by performing unsupervised learning based on the extracted partial traffic flow. The group control device 9 selects a control mode based on the traffic pattern of the traffic flow classified by the learned model, and controls the operation of the car 5 according to the control mode. For example, it is not installed in buildings such as buildings where boarding and alighting congestion may occur on other floors due to multiple uses, or where it is difficult to predict in advance when boarding and alighting congestion will occur. In the elevator, the group control device 9 can output a control mode corresponding to the congestion state that has occurred. Therefore, the group control device 9 can obtain a control mode capable of reducing the waiting time of passengers in an elevator in which conditions such as busy time zones are not known in advance.

The learning device 12 also includes a learning data acquisition unit 14 and a model generation unit 16 . The learning device 12 generates a trained model for inferring a control mode from traffic flow information by unsupervised learning. Since the trained model is generated by unsupervised learning based on the information of the partial traffic flow, it is possible to take into consideration the current situation rather than the typical situation set based on human experience. As a result, it is possible to output a control mode capable of reducing the waiting time of passengers in an elevator where conditions such as busy time zones are not known in advance.

Note that the control mode storage unit 17 may store information on some control modes in advance. The control mode creation unit 18 assigns a control mode corresponding to each of a plurality of traffic patterns shown in the learned model from among the control modes already stored in the control mode storage unit 17, thereby matching the traffic pattern. You may determine the control mode to use. The control mode creation unit 18 may update the control mode information stored in the control mode storage unit 17 to the control mode information to which the traffic pattern is assigned. That is, the learning device 12 may create information that enables inference of the control mode corresponding to the traffic flow by applying the learned model from the traffic flow information.

In addition, the learning device 12 creates a plurality of control modes corresponding to a plurality of traffic patterns classified by unsupervised learning by repeating the traffic simulation. Multiple traffic patterns can be traffic patterns that are difficult to predict in advance because they are classified by models using unsupervised learning. The learning device 12 can determine the most appropriate control mode even for such traffic patterns.

In addition, the learning device 12 extracts a time zone in which the moving average value of the number of passenger occurrences in the entire elevator is equal to or greater than the threshold as a partial traffic flow. Therefore, even if the duration of the congestion state is short, a corresponding control mode can be created. As a result, control modes that reduce passenger waiting times can be provided for a variety of traffic patterns.

In addition, the learning device 12 sets the time at which the increase in the moving average value of the number of passenger occurrences in the entire elevator becomes larger than the specified increase threshold as the start time, and the number of passenger occurrences is the moving average value of the number of passenger occurrences at the start time. The end time is the time when: The learning device 12 extracts the time period from the start time determined in this way to the end time as a partial traffic flow. Therefore, the conditions for extracting partial traffic flows can be set in more detail.

In addition, the learning device 12 creates a learned model based on clustering with input information of the number of passengers and the number of alighters in a specified time period on an arbitrary floor. Therefore, it is possible to accurately reproduce the actual traffic flow of the elevator.

For example, a group management apparatus that classifies traffic patterns with a high probability of occurrence by unsupervised learning can more flexibly deal with congestion than a group management apparatus that classifies by supervised learning. However, in a group management device that classifies traffic patterns with a high occurrence probability by unsupervised learning, it is not possible to perform control considering traffic patterns with a low occurrence probability. Congestion can occur that has a low probability of occurrence but can significantly increase waiting times for passengers. The learning data acquiring unit 14 of the group management device 9 according to Embodiment 1 creates learning source information without depending on the occurrence probability. Therefore, the group management device 9 can create a learned model that does not depend on the occurrence probability of traffic patterns. Therefore, the group management device 9 can more flexibly cope with congestion.

The traffic flow creating unit 11 may use call information acquired from the call registration device 7 when creating traffic flow information. In this case, the call registration device 7 accepts the call including the destination floor. The information on the call includes information on the departure floor and destination floor of the call. Based on the information of the call, the traffic flow creation unit 11 may total the number of passengers and the number of people getting off at regular time intervals to create traffic flow information.

In addition, in the group control device 9, a floor may be preset in consideration of boarding congestion and alighting congestion. In this case, the group control device 9 may treat the number of passengers and the number of passengers alighting at the set floor as the feature quantity of the traffic flow. For example, if it is possible to predict in advance which floors will be congested with boarding and getting off, this setting is made. Therefore, the memory of the group control device 9 and the resource required for arithmetic processing can be reduced.

In the group control device 9, the evaluation value used in the traffic simulation is not only the average waiting time of passengers, but also the power consumption of the entire elevator, the average number of passengers in the car 5, and the like. good. For example, the evaluation value may be the sum of the power consumption, the average number of passengers in the car 5, and the waiting time, which are weighted arbitrarily. Therefore, the power consumption of the elevator can be reduced. The number of passengers inside the car 5 can be reduced.

Next, an example of hardware constituting the group management apparatus 9 will be described with reference to FIG.
FIG. 7 is a hardware configuration diagram of the elevator group control apparatus according to the first embodiment.

Each function of the group management device 9 can be realized by a processing circuit. For example, the processing circuitry comprises at least one processor 100a and at least one memory 100b. For example, the processing circuitry comprises at least one piece of dedicated hardware 200 .

When the processing circuit includes at least one processor 100a and at least one memory 100b, each function of the group control device 9 is realized by software, firmware, or a combination of software and firmware. At least one of software and firmware is written as a program. At least one of software and firmware is stored in at least one memory 100b. At least one processor 100a realizes each function of the group control device 9 by reading and executing a program stored in at least one memory 100b. The at least one processor 100a is also referred to as a central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, DSP. For example, the at least one memory 100b is a nonvolatile or volatile semiconductor memory such as RAM, ROM, flash memory, EPROM, EEPROM, magnetic disk, flexible disk, optical disk, compact disk, mini disk, DVD, or the like.

Where the processing circuitry comprises at least one piece of dedicated hardware 200, the processing circuitry may be implemented, for example, in single circuits, multiple circuits, programmed processors, parallel programmed processors, ASICs, FPGAs, or combinations thereof. be. For example, each function of the group control device 9 is implemented by a processing circuit. For example, each function of the group control device 9 is collectively implemented by a processing circuit.

A part of each function of the group management device 9 may be realized by dedicated hardware 200 and the other part may be realized by software or firmware. For example, the function of controlling the operation of the car 5 based on the control mode is realized by a processing circuit as dedicated hardware 200, and the functions other than the function of controlling the operation of the car 5 based on the control mode are at least one It may be realized by reading and executing a program stored in at least one memory 100b by one processor 100a.

Thus, the processing circuit implements each function of the group control device 9 with hardware 200, software, firmware, or a combination thereof.

Although not shown, the functions of the learning device 12 and the inference device 13 are also implemented by processing circuits equivalent to the processing circuits that implement the functions of the group management device 9 . At least one of the learning device 12 and the reasoning device 13 may not be provided in the group management device 9 . For example, at least one of the learning device 12 and the reasoning device 13 may be installed in a building other than the building 1 in which the group management device 9 is installed. Specifically, at least one of the learning device 12 and the inference device 13 may be provided in a cloud server or the like, and each function may be implemented by being implemented on the cloud server.

Embodiment 2.
FIG. 8 is a block diagram of an elevator group control apparatus according to the second embodiment. In addition, the same code|symbol is attached|subjected to the part which is the same as that of Embodiment 1, or an equivalent part. Description of this part is omitted.

As shown in FIG. 8 , in the second embodiment, the reasoning device 13 of the group control device 9 further includes an unlearned traffic flow storage section 30 and a standard control mode storage section 31 .

The unlearned traffic flow storage unit 30 stores information on traffic flows determined to be unlearned. The traffic flow information determined to be unlearned has the same data structure as the traffic flow information stored in the traffic flow storage unit 10 . Table 5 below shows an example of traffic flow information determined to be unlearned.

As shown in Table 5, for example, traffic flow information determined to be unlearned is traffic flow information at a certain time. The traffic flow information determined to be unlearned may be traffic flow information in a certain time period.

The standard control mode storage unit 31 stores standard control mode information. Standard control modes are preset. Note that the standard control mode may be periodically updated from the outside. Table 6 below provides an example of standard control mode information.

Standard control mode is a control mode in which no special control is performed. For example, as shown in Table 6, in the standard control mode, the car 5 is not pre-allocated to a specific floor.

In the second embodiment, the pattern determination unit 21 of the inference unit 20 determines whether or not the traffic flow is an unlearned traffic flow when determining a traffic flow pattern. When determining that the traffic flow does not apply to any of the plurality of traffic patterns shown in the learned model, the pattern determination unit 21 determines that the traffic flow is an unlearned traffic flow. Specifically, for example, after the inference data acquisition unit 19 acquires the current traffic flow information stored in the traffic flow storage unit 10, the pattern determination unit 21 determines the current traffic flow average and a plurality of traffic patterns. Compute the Euclidean distance to each of the centroids of . If the pattern determination unit 21 determines that the distance between the average of the current traffic flow and the center of gravity is greater than or equal to the specified threshold for all traffic patterns, the pattern determination unit 21 determines that the current traffic flow is an unlearned traffic flow.

When the pattern determination unit 21 determines that the current traffic flow is an unlearned traffic flow, the pattern determination unit 21 stores information on the current traffic flow in the unlearned traffic flow storage unit 30 as an unlearned traffic flow. If the pattern determining unit 21 determines that the current traffic flow is not an unlearned traffic flow, the pattern determining unit 21 determines the current traffic pattern of the traffic flow in the same manner as in the first embodiment.

In the second embodiment, when the pattern determination unit 21 determines that the traffic flow is unlearned, the control mode selection unit 22 of the inference unit 20 selects the standard control mode stored in the standard control mode storage unit 31. do. The control mode selector 22 applies standard control modes to the plurality of controllers 8 to control the operations of the plurality of cars 5 . Note that the group control device 9 may control the operations of the plurality of cars 5 based on the control mode information output from the control mode selection unit 22 .

When the pattern determination unit 21 determines that the traffic flow is not an unlearned traffic flow, the control mode selection unit 22 performs the same operation as in the first embodiment.

When the inference unit 20 determines that the traffic flow is an unlearned traffic flow, the model generation unit 16 performs learning based on the information on the plurality of partial traffic flows, the information on the unlearned traffic flow, and the information on the learned model. Retrain the model by unsupervised learning. Note that the model generation unit 16 may re-learn at any timing. The model generation unit 16 updates the learned model in the model storage unit 15 with re-learned information. For example, by re-learning, the model generation unit 16 newly generates information of a learned model in which an unlearned traffic flow is added to one of the traffic patterns.

Next, an example of processing for re-learning a trained model in the second embodiment will be described with reference to FIG. 9 .
FIG. 9 is a flow chart for explaining an outline of processing for learning a learned model by the elevator group control device according to the second embodiment.

The processing shown in FIG. 9 is the learning processing in Embodiment 2 based on the k-means method. The model generation unit 16 starts learning processing when the inference unit 20 determines that the traffic flow is an unlearned traffic flow.

In step S108, the model generation unit 16 determines whether the unlearned traffic flow information is stored in the unlearned traffic flow storage unit 30 or not.

When it is determined in step S108 that no unlearned traffic flow information is stored, the model generator 16 performs steps S101 to S107. Steps S101 to S107 are the same as steps S101 to S107 in the flowchart of FIG. 4 of the first embodiment.

If it is determined in step S108 that unlearned traffic flow information is stored, the model generator 16 performs the operation of step S109. In step S109, the model generator 16 calculates the average of unlearned traffic flow information. The model generation unit 16 sets the center of gravity of a plurality of traffic patterns with the average of unlearned traffic flows as one of the centers of gravity of the plurality of traffic patterns. That is, the model generator 16 adds the average of unlearned traffic flows to one of the initial solutions of the centroids of the traffic patterns.

After step S108, the model generating unit 16 performs the operations after step S104.

According to the second embodiment described above, when the inference device 13 determines that the current traffic flow is an unlearned traffic flow, the learning device 12 uses the information on the unlearned traffic flow to re-learn the traffic flow. A new learned model is generated by learning. Therefore, it is possible to prevent the current traffic from being forcibly classified into the traffic that has already been learned. In addition, it is possible to operate in a control mode suitable for various traffic flows.

The inference device 13 further includes a standard control mode storage unit 31 that stores information on preset standard control modes. If the inference device 13 determines that the current traffic flow is an unlearned traffic flow, it outputs the standard control mode. The group control device 9 controls the operation of the car 5 according to the output standard control mode. Therefore, when an unlearned traffic flow occurs, the car 5 can be operated in the standard control mode. For example, if a control mode corresponding to a learned traffic flow is set for an unlearned traffic flow, there is a risk that the waiting time of passengers will rather increase. The group control device 9 can suppress an increase in passenger waiting time when such an unlearned traffic flow occurs.

1 building, 2 hoistway, 3 machine room, 4 platform, 5 car, 6 weighing device, 7 call registration device, 8 control device, 9 group control device, 10 traffic flow storage unit, 11 traffic flow creation unit, 12 learning Apparatus 13 Inference device 14 Learning data acquisition unit 15 Model storage unit 16 Model generation unit 17 Control mode storage unit 18 Control mode creation unit 19 Inference data acquisition unit 20 Inference unit 21 Pattern determination unit 22 Control mode selection unit 30 Unlearned traffic flow storage unit 31 Standard control mode storage unit 100a Processor 100b Memory 200 Hardware

Claims (10)

Calculate the number of passengers per day based on traffic flow information that indicates the number of elevator passengers boarding or exiting one of the multiple cars on each floor, and based on the number of passengers generated, A learning data acquisition unit that determines a time zone that satisfies a specified condition and is recognized as a congested state , and acquires learning data including information on partial traffic flow, which is the information on the traffic flow in the time zone;
a model generation unit that uses the learning data to generate a trained model by unsupervised learning for inferring a control mode of a group control device that controls the plurality of cars from the traffic flow information;
A learning device with The learned model generated by the model generation unit classifies the traffic flow information into a plurality of traffic patterns,
a control mode creation unit that determines the control mode corresponding to the traffic pattern classified based on the learned model;
The learning device of claim 1, further comprising: The control mode creation unit sets the temporary control mode, which is the temporary control mode, for the traffic patterns classified based on the learned model, and performs a traffic simulation for transporting passengers. 3. The learning device according to claim 2, wherein the control mode is determined by calculating an evaluation value. The learned model generated by the model generation unit classifies the traffic flow information into a plurality of traffic patterns,
The model generation unit uses the generated learned model to classify the traffic flow into the plurality of traffic patterns, and determines whether the traffic flow input to the inference device is one of the plurality of traffic patterns. When it is determined that the unlearned traffic flow does not apply to the above, the learned model is newly generated by re-learning based on the learning data and the unlearned traffic flow. The learning device according to any one of claims 1 to 3. The learning data acquisition unit includes the partial traffic flow information, which is traffic flow information obtained by extracting a time period in which a moving average value of the number of passenger occurrences in the entire elevator is equal to or greater than a threshold based on the traffic flow information. 5. The learning device according to any one of claims 1 to 4, wherein the learning data is acquired. The learning data acquisition unit, based on traffic flow information, sets the time at which the increase in the moving average value of the number of passenger occurrences in the entire elevator becomes larger than a prescribed increase threshold as the start time, and the number of passenger occurrences is set to the start time. The partial traffic flow information, which is traffic flow information obtained by extracting a time period from the start time to the end time of the traffic flow, with the end time set to the time when the number of passengers generated at that time is equal to or less than the moving average value. 5. The learning device according to any one of claims 1 to 4, wherein the learning data including The model generation unit converts the partial traffic flow into a traffic pattern based on clustering using, as the partial traffic flow, the number of passengers and the number of people getting off in a specified time period on an arbitrary floor of the building in which the elevator is installed as input information. 7. The learning device according to any one of claims 1 to 6, which generates the learned model classified into . an inference data acquisition unit that acquires traffic flow information indicating the number of elevator passengers getting into or out of one of a plurality of cars on each floor;
Based on the information on the traffic flow, the number of passengers generated per day is calculated, based on the number of passengers generated , the time zone that satisfies the specified conditions and is recognized as being congested is determined. A control mode of a group control device that is a model generated by unsupervised learning using partial traffic flow information, which is the traffic flow information, as learning data, and that controls the plurality of cars from the traffic flow information. an inference unit that outputs the control mode of the group control device that controls the plurality of cars from the traffic flow acquired by the inference data acquisition unit, using a learned model for inferring the
A reasoning device with The learned model classifies the traffic flow information into a plurality of traffic patterns,
When the inference unit determines that the traffic flow acquired by the inference data acquisition unit is an unlearned traffic flow that does not apply to any of the plurality of traffic patterns, the preset standard control is performed. 9. A reasoning apparatus according to claim 8, which outputs a mode. a traffic flow creation unit that creates a traffic flow indicating the number of elevator passengers getting on or off one of a plurality of cars on each floor;
Based on the information on the traffic flow, the number of passengers generated per day is calculated, based on the number of passengers generated, the time zone that satisfies the specified conditions and is recognized as being congested is determined. a learning data acquisition unit that extracts traffic flow and creates partial traffic flow information;
a model generation unit that performs unsupervised learning based on the partial traffic flow information generated by the learning data acquisition unit to generate a trained model that classifies the traffic flow into one of a plurality of traffic patterns;
a control mode creation unit that determines a plurality of control modes respectively corresponding to the plurality of traffic patterns indicated by the learned model;
a pattern determination unit that determines a traffic pattern into which the traffic flow is classified among the plurality of traffic patterns by applying the traffic flow to the learned model;
Out of the plurality of control modes created by the control mode creation unit, a control mode corresponding to the traffic pattern determined by the pattern determination unit is output, and operation of the plurality of cars is controlled based on the output control mode. a control mode selector for
elevator group control device.

Images