Recommendation system, recommendation method and program

Abstract

To provide a recommendation system, a recommendation method, and a program capable of recommending an operation mode of a control object corresponding to a change in a demand.SOLUTION: According to an embodiment, a recommendation system includes storage means, acquisition means, extraction means, determination means, and output means. The storage means stores recommendation data including an operation mode to be recommended among a plurality of operation modes of a control object prepared in advance by associating use data about use of the control object with each of a plurality of classes to be classified. The acquisition means acquires first use data about the use of the control object in a first time zone. The extraction means extracts a first feature amount from the acquired first use data. The determination means determines a cluster for classifying the first use data on the basis of the extracted first feature amount. The output means outputs an operation mode included in the recommendation data in association with the determined cluster.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] FIELD Embodiments of the present invention relate to a recommendation system, a recommendation method, and a program. [Background technology]

[0002] It is generally known that the operation of building facilities such as elevators or air conditioners (hereinafter referred to as controlled objects) is controlled in a predetermined operation mode.

[0003] The operation mode for controlling the operation of a controlled object is determined, for example, according to instructions from a manager who manages the controlled object or predetermined conditions (such as a time period), but the demand for the controlled object changes from moment to moment.

[0004] For this reason, an operation mode determined simply in accordance with instructions from a manager or predetermined conditions may not be able to realize an operation of the controlled object that is suited to changes in demand. [Prior art documents] [Patent documents]

[0005] [Patent Document 1] Japanese Patent Application Laid-Open No. 2012-226460 Summary of the Invention [Problem to be solved by the invention]

[0006] Therefore, an object of the present invention is to provide a recommendation system, a recommendation method, and a program that can recommend an operation mode of a controlled object in accordance with changes in demand. [Means for solving the problem]

[0007] According to an embodiment, a recommendation system is provided that recommends an operation mode of a controlled object. The recommendation system includes a storage means, an acquisition means, an extraction means, a determination means, and an output means. The storage means stores recommendation data including a recommended operation mode from among a plurality of operation modes of the controlled object that are prepared in advance and associated with each of a plurality of clusters into which usage data related to the use of the controlled object is classified. The acquisition means acquires first usage data related to the use of the controlled object in a first time period. The extraction means extracts a first feature from the acquired first usage data. The determination means determines a cluster into which the first usage data is classified based on the extracted first feature. The output means outputs the operation mode included in the recommendation data in association with the determined cluster. [Brief explanation of the drawings]

[0008] [Figure 1] FIG. 1 is a diagram showing an example of a network configuration including a recommendation system according to an embodiment. [Figure 2] FIG. 2 is a diagram showing an example of the hardware configuration of a learning device. [Figure 3] FIG. 2 is a diagram showing an example of the hardware configuration of a recommended device. [Figure 4] FIG. 2 is a diagram showing an example of the functional configuration of a learning device. [Figure 5] FIG. 2 is a diagram showing an example of the functional configuration of a recommended device. [Figure 6] 10 is a flowchart showing an example of a processing procedure of a learning device. [Figure 7] FIG. 2 is a diagram showing an example of the data structure of people flow data. [Figure 8] FIG. 4 is a diagram showing an example of a data structure of feature amount data. [Figure 9] FIG. 10 is a diagram showing an example of the data structure of feature amount data to which cluster numbers have been added. [Figure 10] 10 is a flowchart showing an example of a processing procedure for a recommended data generation process. [Figure 11] FIG. 10 is a diagram showing an example of the data structure of feature amount data to which an operation mode is added. [Figure 12]FIG. 4 is a diagram showing an example of a data structure of recommendation data. [Figure 13] 10 is a flowchart showing another example of the recommended data generation process. [Figure 14] 10 is a flowchart showing an example of a processing procedure of a recommended device. [Figure 15] 10 is a flowchart showing an example of a processing procedure for a cluster determination process. [Figure 16] FIG. 10 is a diagram for explaining a modified example of the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] Hereinafter, embodiments will be described with reference to the drawings. Fig. 1 shows an example of a network configuration including a recommendation system according to this embodiment. As shown in Fig. 1, the recommendation system 1 is communicably connected to a people flow system 2 and an elevator system 3 via a network 4 such as the Internet.

[0010] The people flow system 2 is configured to collect (aggregate) and output people flow data (i.e., data regarding people flow based on the use of elevators within the target building) indicating the flow of people using elevators installed in a specified building, such as an office building (hereinafter referred to as the target building), at any time.

[0011] The elevator system 3 includes elevators installed in the target building and elevator control devices that control the operation of the elevators. In this embodiment, the elevator system 3 includes, for example, a group management system. The group management system has the function of group management and control of multiple elevators (cars). With such a group management system, for example, when a user uses an elevator, the user can select the most appropriate elevator (car) from multiple elevators, thereby realizing efficient travel for the user. In the following, the elevator system 3 in this embodiment will be described as a group management system, but the elevator system 3 may also be an elevator system other than a group management system.

[0012] In the elevator system 3, the elevator control device controls the operation of the elevators by switching between, for example, multiple pre-prepared operation modes. When the elevator system 3 is the group control system described above, the multiple pre-prepared operation modes in the elevator system 3 include, for example, a normal operation mode, an off-peak operation mode, a commuter operation mode, a lunch operation mode, and a departure operation mode. The normal operation mode is an operation mode that does not perform special control like the other operation modes. The off-peak operation mode is an operation mode in which multiple elevators are distributed to different floors when, for example, the number of passengers using the elevators is low and there is no particular trend in the departure floors (boarding floors) where passengers board the elevators and the arrival floors (disembarking floors) where passengers disembark. The commuter operation mode is an operation mode that controls the operation of the elevators so that, based on the tendency for many passengers to board from the entrance hall (reference floor) when commuting to work, the elevator is moved to the reference floor when all passengers have disembarked and there are no remaining calls for the elevator. The lunch operation mode is an operation mode in which, based on the tendency that many passengers disembark at a floor with a cafeteria (the cafeteria floor) during the first half of the lunch hour, elevators are allocated to calls originating at a hall that are headed toward the cafeteria floor, with priority given to calls originating at the hall, and based on the tendency that many passengers board at a floor with a cafeteria (the cafeteria floor) during the second half of the lunch hour, elevator operation is controlled so that multiple elevators are allocated to calls originating at a hall on the cafeteria floor.The off-duty operation mode is an operation mode in which, based on the tendency that many passengers disembark at the entrance hall (the reference floor) during off-duty hours, elevator operation is controlled so that if a passenger disembarks at the reference floor, the elevator is moved to a floor other than the reference floor.

[0013] In addition, in the elevator system 3 (group management system), it is also possible to realize control such that, for example, some of the multiple elevators are operated as elevators for the upper floors of the target building, and the others are operated as elevators for the lower floors of the target building.

[0014] The recommendation system 1 of this embodiment is used to recommend to the elevator system 3 an elevator operating mode that corresponds to changes in the flow of people (i.e., changes in demand for elevators) indicated by the people flow data collected in the above-mentioned people flow system 2.

[0015] As shown in FIG. 1 , the recommendation system 1 includes a learning device 11 and a recommendation device 12. In this embodiment, the recommendation system 1 is described as including the learning device 11 and the recommendation device 12 as separate devices, but the learning device 11 and the recommendation device 12 may also be realized as a single device. In this embodiment, one or both of the learning device 11 and the recommendation device 12 may be installed in a target building, or may be a server device that provides a cloud computing service. Furthermore, the recommendation system 1 according to this embodiment may be configured to include only one of the learning device and the recommendation device.

[0016] The following describes the learning device 11 and the recommendation device 12 included in the recommendation system 1 according to this embodiment. First, the hardware configurations of the learning device 11 and the recommendation device 12 will be described.

[0017] Fig. 2 shows an example of the hardware configuration of the learning device 11. As shown in Fig. 2, the learning device 11 includes a CPU 11a, a nonvolatile memory 11b, a RAM 11c, a communication device 11d, and the like.

[0018] The CPU 11a is a processor for controlling the operation of each component in the learning device 11. The CPU 11a may be a single processor or may be configured with multiple processors. The CPU 11a executes various programs loaded from the non-volatile memory 11b to the RAM 11c. In this embodiment, the programs executed by the CPU 11a include a learning program 11e.

[0019] The nonvolatile memory 11b is a storage medium used as an auxiliary storage device. The RAM 11c is a storage medium used as a main storage device. Although only the nonvolatile memory 11b and the RAM 11c are shown in FIG. 2, the learning device 11 may also include other storage devices such as a hard disk drive (HDD) and a solid state drive (SSD).

[0020] The communication device 11d is a device configured to communicate with external devices. The external devices with which the learning device 11 communicates include, for example, the people flow system 2 and the recommendation device 12.

[0021] Fig. 3 shows an example of the hardware configuration of the recommendation device 12. As shown in Fig. 3, the recommendation device 12 includes a CPU 12a, a nonvolatile memory 12b, a RAM 12c, a communication device 12d, and the like.

[0022] The CPU 12a is a processor for controlling the operation of each component in the recommendation device 12. The CPU 12a may be a single processor or may be configured with multiple processors. The CPU 12a executes various programs loaded from the non-volatile memory 12b to the RAM 12c. In this embodiment, the programs executed by the CPU 12a include a recommendation program 12e.

[0023] The nonvolatile memory 12b is a storage medium used as an auxiliary storage device. The RAM 12c is a storage medium used as a main storage device. Although only the nonvolatile memory 12b and the RAM 12c are shown in Fig. 3, the recommendation device 12 may include other storage devices such as a hard disk drive (HDD) and a solid state drive (SSD).

[0024] The communication device 12d is a device configured to communicate with external devices. The external devices with which the recommendation device 12 communicates include, for example, the people flow system 2, the elevator system 3, and the learning device 11.

[0025] Next, we will explain the functional configuration of the learning device 11 and the recommendation device 12. Fig. 4 shows an example of the functional configuration of the learning device 11. As shown in Fig. 4, the learning device 11 includes a data acquisition unit 111, a feature extraction unit 112, a clustering unit 113, a recommended data generation unit 114, and a storage unit 115.

[0026] It should be noted that some or all of the data acquisition unit 111, feature extraction unit 112, clustering unit 113, and recommended data generation unit 114 are realized by software, for example, when the CPU 11a (i.e., the computer of the learning device 11) executes the learning program 11e. This learning program 11e may be stored in a computer-readable storage medium and distributed, or may be downloaded to the learning device 11 via a network.

[0027] Here, it has been described that some or all of the units 111 to 114 are realized by software, but some or all of the units 111 to 114 may also be realized by hardware such as an IC (Integrated Circuit), or may be realized by a combination of software and hardware.

[0028] The storage unit 115 is realized by, for example, the nonvolatile memory 11b or other storage devices.

[0029] The data acquisition unit 111 acquires (receives) from the people flow system 2 a plurality of pieces of people flow data collected in the people flow system 2, for example, for each predetermined period (time period). The plurality of pieces of people flow data acquired by the data acquisition unit 111 is stored (accumulated) in the storage unit 115. The plurality of pieces of people flow data stored in the storage unit 115 corresponds to a collection of people flow data indicating the past flow of people within the target building (i.e., past people flow data). Here, it has been described that the plurality of pieces of people flow data acquired by the data acquisition unit 111 is stored in the storage unit 115, but the plurality of pieces of people flow data may also be stored in the storage unit 115 in advance.

[0030] The feature extraction unit 112 acquires from the storage unit 115 a plurality of pieces of people flow data accumulated (stored) in the storage unit 115, and extracts, from each of the plurality of pieces of people flow data, features that represent the characteristics of the flow of people indicated by the people flow data. Data including the features extracted by the feature extraction unit 112 (hereinafter referred to as feature data) is stored in the storage unit 115.

[0031] The clustering unit 113 classifies the plurality of people flow data stored in the storage unit 115 into a plurality of clusters (i.e., performs clustering on the plurality of people flow data) based on the feature data (feature amounts included in the feature data) stored in the storage unit 115. The clustering results by the clustering unit 113 are stored in the storage unit 115.

[0032] The recommended data generation unit 114 samples, as sample data, some of the people flow data classified into each of the clusters by the clustering unit 113. Based on the sampled sample data, the recommended data generation unit 114 determines, for each cluster into which the above-mentioned people flow data is classified, one of the pre-prepared driving modes as a driving mode corresponding to the people flow indicated by the people flow data classified into the cluster (i.e., a driving mode to be recommended). The recommended data generation unit 114 generates recommended data indicating the driving mode determined for each cluster (i.e., recommended data including the driving mode determined for the cluster in association with the cluster). The recommended data generated by the recommended data generation unit 114 in this manner is output to the recommendation device 12 and is used by the recommendation device 12 when recommending an elevator driving mode.

[0033] That is, in this embodiment, the learning device 11 operates to learn past people flow data (feature data including features extracted from the data) and execute a process (learning process) to generate recommended data used to recommend an elevator operation mode.

[0034] 5 shows an example of the functional configuration of the recommendation device 12. As shown in FIG. 5, the recommendation device 12 includes a data acquisition unit 121, a feature extraction unit 122, a cluster determination unit 123, a driving mode determination unit 124, and a storage unit 125.

[0035] It should be noted that some or all of the data acquisition unit 121, the feature extraction unit 122, the cluster determination unit 123, and the driving mode determination unit 124 are realized by, for example, the CPU 12a (i.e., the computer of the recommendation device 12) executing the recommendation program 12e, that is, by software. This recommendation program 12e may be stored in a computer-readable storage medium and distributed, or may be downloaded to the recommendation device 12 via a network.

[0036] Here, it has been described that some or all of the units 121 to 124 are realized by software, but some or all of the units 121 to 124 may be realized by hardware such as an IC (Integrated Circuit), or may be realized by a combination of software and hardware.

[0037] The storage unit 125 is realized by, for example, the nonvolatile memory 12b or other storage devices.

[0038] The data acquisition unit 121 acquires (receives) the recommended data output (transmitted) from (the recommended data generation unit 114 included in) the learning device 11. The recommended data acquired by the data acquisition unit 121 is stored in the storage unit 125.

[0039] Furthermore, the data acquisition unit 121 acquires (receives) people flow data collected in the people flow system 2 from the people flow system 2. The people flow data acquired by the data acquisition unit 121 corresponds to people flow data (i.e., current people flow data) indicating the current flow of people (movement of users) within the target building, for example, and is output to the feature extraction unit 122.

[0040] The feature extraction unit 122 extracts features that represent the characteristics of the flow of people indicated by the people flow data from the people flow data output from the data acquisition unit 121. The feature data including the features extracted by the feature extraction unit 122 is output to the cluster determination unit 123.

[0041] The cluster determination unit 123 determines a cluster into which the people flow data acquired by the data acquisition unit 121 is classified, based on the features output from the feature extraction unit 122. In this case, the cluster determination unit 123 determines (discriminates) one cluster from among the multiple clusters into which the multiple people flow data have been classified in the learning device 11 described above.

[0042] Based on the recommendation data stored in the storage unit 125 and the cluster determined by the cluster determination unit 123, the operation mode determination unit 124 determines the operation mode included in the recommendation data in association with the cluster as the elevator operation mode to be recommended to the elevator system 3 (i.e., the operation mode according to the flow of people indicated by the current people flow data). The operation mode determined by the operation mode determination unit 124 is output (transmitted) to the elevator system 3 and used to control the operation of the elevators in the elevator system 3.

[0043] That is, in this embodiment, the recommendation device 12 operates to determine a cluster into which the pedestrian flow data is classified based on feature data including features extracted from the current pedestrian flow data, and to perform a process (recommendation process) of recommending to the elevator system 3 an operating mode corresponding to the determined cluster.

[0044] The operation of the recommendation system 1 (the learning device 11 and the recommendation device 12) according to this embodiment will be described below.

[0045] First, an example of the processing procedure of the learning device 11 (the learning process executed by the learning device 11) will be described with reference to the flowchart in Fig. 6. Note that the processing shown in Fig. 6 may be executed, for example, according to an instruction (i.e., an external request) from an administrator of the recommendation system 1 (learning device 11), or may be executed when a predetermined condition, such as reaching a predetermined time, is met.

[0046] The feature extraction unit 112 acquires past people flow data stored in the storage unit 115 from the storage unit 115 (step S1). Note that the past people flow data is a collection of people flow data (plural people flow data) previously collected in the people flow system 2, as described above. Here, it has been described that the past people flow data stored in the storage unit 115 is acquired, but the past people flow data may also be acquired (received), for example, via the data acquisition unit 111 from an external device that manages the past people flow data.

[0047] Here, an example of the data structure of (one of) the past people flow data acquired in step S1 will be described with reference to FIG.

[0048] As shown in Figure 7, the people flow data includes the date and time when the people flow data was collected. In the example shown in Figure 7, the people flow data includes the date and time "2020 / 02 / 18 10:30." As described above, people flow data is collected every predetermined period. If the predetermined period is five minutes, the people flow data shown in Figure 7 indicates the flow of people (movement of users using elevators) within the target building for five minutes from 10:30 to 10:35 on February 18, 2020.

[0049] As shown in FIG. 7 , the people flow data includes the number of users who traveled from the departure floor (the floor where the user boarded the elevator) to the arrival floor (the floor where the user disembarked from the elevator) of each of multiple users who used the elevators installed in the target building. In the example shown in FIG. 7 , the people flow data includes, for example, a "4" associated with the departure floor "2F" and the arrival floor "9F." This indicates that the number of users who traveled from the 2F to the 9F during the five minutes from 10:30 to 10:35 on February 18, 2020, is four. In the example shown in FIG. 7 , the people flow data includes, for example, a "7" associated with the departure floor "9F" and the arrival floor "1F." This indicates that the number of users who traveled from the 9F to the 1F during the five minutes from 10:30 to 10:35 on February 18, 2020, is seven.

[0050] The people flow data shown in Fig. 7 is called an OD (Origin-Destination) table, and is generated by aggregating user movements that occur for each specified period by floor and direction. In order to generate such people flow data, it is necessary to recognize the departure and arrival floors of users who use elevators, and these can be recognized by using security gate passage data (data indicating that users have passed through the security gate) obtained from the security gates installed on each floor where the elevator lands.

[0051] Furthermore, in this embodiment, it is assumed that the data acquisition unit 111 acquires people flow data (OD table) from the people flow system 2 and stores the people flow data in the storage unit 115 (i.e., the people flow data is generated on the people flow system 2 side), but the data acquisition unit 111 may also be configured to acquire data for generating people flow data (i.e., security gate passage data) and generate people flow data based on that data. In other words, this embodiment may also be configured so that the people flow data generated by the data acquisition unit 111 (i.e., generated on the learning device 11 side) is stored in the storage unit 115.

[0052] Furthermore, although we have explained here that people flow data is generated using security gate passage data, the departure and arrival floors of the above-mentioned users may also be recognized using, for example, beacons or cameras (i.e., facial recognition).

[0053] Although the people flow data has been described in FIG. 7 as being in the format of an OD table, the people flow data may be in another format.

[0054] Although the explanation here has been given mainly on one piece of people flow data included in the past people flow data, as described above, multiple pieces of people flow data collected every predetermined period (for example, 5 minutes) are stored in the storage unit 115 as past people flow data, and the multiple pieces of people flow data are acquired in step S1. Note that the predetermined period for collecting people flow data may be other than 5 minutes.

[0055] Returning to FIG. 6 again, the feature extraction unit 112 extracts feature amounts from each piece of past people flow data acquired in step S1 (step S2).

[0056] The feature amounts extracted in step S2 will be specifically described below. Here, it is assumed that five feature amounts (hereinafter referred to as feature amounts 1 to 5) are extracted in step S2.

[0057] First, feature 1 includes the number of users moving in the upward direction (number of users moving in the upward direction). Feature 2 includes the number of users moving in the downward direction (number of users moving in the downward direction). Feature 3 includes the number of users whose departure and arrival floors are other than a predetermined reference floor (e.g., an entrance hall, etc.) (inter-floor traffic volume). Feature 4 includes the number of users moving in the upward direction that exceeds the number of users moving in the downward direction (excess number of users moving in the upward direction). Feature 5 includes the number of users moving in the downward direction that exceeds the number of users moving in the upward direction (excess number of users moving in the downward direction).

[0058] Feature 1 is extracted (identified) by calculating the sum of the number of users associated with the departure and arrival floors corresponding to the upbound direction in each of the above-mentioned past people flow data (i.e., the sum of the values ​​in the upper right half of the diagonal in the OD table shown in FIG. 7). Feature 2 is extracted (identified) by calculating the sum of the number of users associated with the departure and arrival floors corresponding to the downbound direction in each of the above-mentioned past people flow data (i.e., the sum of the values ​​in the lower left half of the diagonal in the OD table shown in FIG. 7). While features 1 and 2 have been described here, the other features 3 to 5 can also be identified (extracted) based on each of the past people flow data (the departure and arrival floors of each user included in the data). For example, if the number of users moving downbound is greater than the number of users moving upbound, feature 4 is set to 0. Similarly, if the number of users moving upbound is greater than the number of users moving downbound, feature 5 is set to 0.

[0059] Furthermore, the above-described features 1 to 5 may be normalized. In this case, features 1 to 5 can be normalized, for example, by dividing them by the total number of users (total traffic volume) calculated (specified) based on the people flow data from which features 1 to 5 were extracted. When normalized in this way, feature 1 becomes the ratio of people going up, feature 2 becomes the ratio of people going down, feature 3 becomes the inter-floor traffic volume ratio, feature 4 becomes the ratio of excess people going up, and feature 5 becomes the ratio of excess people going down.

[0060] Furthermore, although features 1 to 5 are extracted from each of the past people flow data acquired in step S1, the features 1 to 5 may be normalized by dividing them by the maximum value of the total number of users (maximum traffic volume) calculated based on each of the past people flow data. Furthermore, features 1 to 5 may be normalized by dividing them by the maximum value of each of the features 1 to 5 extracted from each of the past people flow data or the maximum value of the maximum values ​​of each of the features 1 to 5.

[0061] The parameter values ​​(e.g., total traffic volume, etc.) for normalizing the above-described feature quantities 1 to 5 may be determined (calculated) by the feature quantity extraction unit 112 at the time the processing of step S2 is executed, or may be stored in advance as set values ​​(fixed values) in the storage unit 115. The parameter values ​​for normalizing the feature quantities 1 to 5 described here are merely examples, and may be values ​​calculated by a formula combining the maximum value, minimum value, or average value of the number of users identified from the above-described people flow data based on a predetermined viewpoint.

[0062] When the process of step S2 is executed, the feature extraction unit 112 stores in the storage unit 115 feature data including the feature extracted in step S2.

[0063] Here, Fig. 8 shows an example of the data structure of feature data. As shown in Fig. 8, the feature data includes the date and time included in the people flow data and feature amounts 1 to 5 extracted from the people flow data. Note that the feature data can be considered as vector data having each of feature amounts 1 to 5 as components, which are associated with the people flow data based on the date and time, for example.

[0064] The date and time indicates the date and time (time period) when the people flow data from which Features 1 to 5 included in the feature data were extracted was collected. Features 1 to 5 are features extracted from the people flow data corresponding to the feature data (people flow data including the same date and time as the feature data). Note that normalized Features 1 to 5 are shown in FIG. 8.

[0065] Although only one piece of feature data is shown in Figure 8, the feature data is stored in storage unit 115 for each piece of people flow data acquired in step S1 (i.e., storage unit 115 stores multiple pieces of feature data corresponding to each piece of past people flow data).

[0066] Returning to Figure 6 again, the clustering unit 113 performs clustering on the past people flow data using multiple feature data stored in the storage unit 115 (i.e., feature data including features extracted from each of the past people flow data in step S2) (step S3).

[0067] The clustering in step S3 is performed, for example, based on k-means, a non-hierarchical clustering algorithm. According to k-means, feature vector data that is close to the center value (hereinafter referred to as the centroid) of a cluster is assigned to that cluster, and by repeating a process of recalculating the centroid, it is possible to classify multiple people flow data corresponding to multiple feature vector data with similar trends in the features corresponding to the components of the vector data into the same cluster. While an example in which clustering is performed based on k-means has been shown here, other algorithms may also be used.

[0068] As described above, a machine learning model (hereinafter referred to as a cluster model) that clusters feature data (people flow data) is stored in the storage unit 115. The cluster model holds data necessary for clustering, such as a cluster number assigned to each of a plurality of clusters into which past people flow data is classified, and the centroid of the cluster.

[0069] Furthermore, the number of clusters (e.g., N), which is a hyperparameter in the k-means algorithm, may be determined by the clustering unit 113 at the time the process of step S3 is executed, or may be stored in advance as a set value (fixed value) in the storage unit 115. In this embodiment, the number of clusters is the number of elevator operation modes (the number of candidate recommended operation modes), but it may be any positive number unrelated to the number of elevator operation modes, and may be determined, for example, using the maximum, minimum, or average value of the number of users identified from people flow data based on a predetermined perspective, or a value calculated by a formula combining these.

[0070] The result of the processing in step S3 (i.e., the clustering result) is stored in the storage unit 115. The clustering result may be stored in the storage unit 115 in a manner that allows for understanding of the correspondence between the clusters and the people flow data (feature data) classified into the clusters. As an example, as shown in FIG. 9, a cluster number assigned to a cluster into which people flow data has been classified is added to the feature data corresponding to the people flow data. In the example shown in FIG. 9, the cluster number assigned to the cluster into which people flow data corresponding to the feature data shown in FIG. 9 has been classified is 1. Although not shown, for example, the cluster number assigned to the cluster into which the people flow data has been classified may be added to each of the past people flow data stored in the storage unit 115.

[0071] After the process of step S3 is executed, the recommended data generating unit 114 executes a process of generating recommended data based on the clustering result (hereinafter referred to as a recommended data generating process) (step S4).

[0072] An example of the processing procedure of the above-mentioned recommended data generation process will be described below with reference to the flowchart in Fig. 10. Note that when the recommended data generation process is executed, it is assumed that the past people flow data (and the feature data corresponding to the people flow data) have been organized (grouped) for each cluster based on the results of the clustering performed in step S3 shown in Fig. 6.

[0073] First, the recommended data generation unit 114 selects one cluster from among the clusters into which the above-mentioned past people flow data has been classified (step S11). Hereinafter, the cluster selected in step S11 will be referred to as cluster i.

[0074] Here, the recommended data generation unit 114 includes a sampling unit that samples some of the people flow data classified into cluster i. As a result, the recommended data generation unit 114 acquires sample data (e.g., one of the people flow data classified into cluster i) sampled by the sampling unit from the people flow data classified into cluster i (step S12). When the clustering result (cluster number) is added to the feature data as described above, in step S12, sample data is acquired from the people flow data corresponding to the feature data to which the cluster number assigned to cluster i is added (people flow data with the same date and time as the feature data).

[0075] The sample data may be any of the people flow data classified into cluster i, but may also be people flow data that satisfies predetermined conditions (hereinafter referred to as filtering conditions) among the plurality of people flow data. The filtering conditions may include, for example, the total traffic volume (total number of users) being greater than a predetermined value (or not being zero). Such filtering conditions can prevent people flow data that does not adequately represent the flow of people in the target building from being acquired as sample data. The filtering conditions may be any conditions other than those described here, as long as they exclude people flow data that is not suitable for generating recommended data (determining an operating mode) as outliers. The filtering conditions may be, for example, set values ​​stored in the storage unit 115. Hereinafter, the sample data acquired in step S12 will be referred to as sample data j.

[0076] When the process of step S12 is executed, the recommended data generation unit 114 selects one operation mode from among a plurality of operation modes of the elevator (step S13). Note that the list of operation modes of the elevator (data indicating the list) is assumed to be stored in advance in the storage unit 115 as, for example, set values. Hereinafter, the operation mode selected in step S13 will be referred to as operation mode k.

[0077] Here, the recommended data generation unit 114 has a simulation unit that simulates elevator operation when operation mode k is applied to sample data j (the flow of people indicated by the sample data j). The simulation unit holds, for example, a simulator (elevator model) that simulates elevator system 3 (group control system), and operates the elevator in operation mode k in the simulator (i.e., simulates elevator operation) so as to realize the flow of people indicated by sample data j (people flow data) (i.e., the movement of each user from the departure floor to the arrival floor). The recommended data generation unit 114 acquires the simulation results from the simulation unit (step S14).

[0078] The simulation results obtained in step S14 include, for example, at least one of energy consumption, a performance index, and a comfort index as indices obtained by simulating elevator operation. The energy consumption is calculated, for example, based on the power consumption of the elevator when all passengers included in the sample data are transported from the departure floor to the arrival floor. The performance index is calculated, for example, based on the waiting time of each passenger and the service time for each passenger when all passengers included in the sample data are transported from the departure floor to the arrival floor. The service time refers to the time from when a passenger arrives at the boarding area (a call is made at the boarding area) to when they disembark at the destination floor (when the elevator arrives at the destination floor) (i.e., waiting time + riding time). The comfort index is calculated, for example, based on the load capacity in the elevator (car) (i.e., the number of passengers) when all passengers included in the sample data are transported from the departure floor to the arrival floor. Here, the simulation results have been described as including at least one of energy consumption, performance index, and comfort index, but the simulation results may also include an index that is useful for determining whether or not operation mode k should be applied to the flow of people indicated by sample data j.

[0079] The simulation results obtained in step S14 are stored inside the recommended data generating unit 114.

[0080] After the process of step S14 is executed, it is determined whether or not the process has been executed for all the operation modes (that is, whether or not the simulation results for all the operation modes have been acquired) (step S15).

[0081] If it is determined that the process has not been executed for all the operation modes (NO in step S15), the process returns to step S13 and is repeated. In this case, an operation mode for which the process has not been executed is selected in step S13, and the process of step S14 is executed.

[0082] On the other hand, if it is determined that the processing has been performed for all the operation modes (YES in step S15), the recommended data generator 114 determines the operation mode of the sample data j (step S16).

[0083] Here, when it is determined that the processes of steps S13 and S14 have been executed for all operation modes as described above (i.e., a round-robin simulation has been performed for all operation modes), the simulation results for each operation mode are stored inside the recommendation data generation unit 114. The simulation results include at least one of the energy consumption, performance index, and comfort index as indices obtained by the simulation.

[0084] In this case, for example, if the energy consumption included in the simulation results of a specific driving mode is low, the driving mode can be said to have a high energy-saving effect for the flow of people (people flow situation) indicated by sample data j and to be an driving mode suitable for that flow of people. Also, if the performance index included in the simulation results of a specific driving mode is high, the driving mode can be said to be able to efficiently move users for the flow of people (people flow situation) indicated by sample data j and to be an driving mode suitable for that flow of people. Furthermore, if the comfort index included in the simulation results of a specific driving mode is high, the driving mode can be said to be able to move users in a comfortable state for the flow of people (people flow situation) indicated by sample data j and to be an driving mode suitable for that flow of people.

[0085] In step S16, the driving mode that provides the best simulation results (at least one of energy consumption, performance index, and comfort index) based on the above-mentioned perspectives is determined as the driving mode for sample data j (the driving mode suitable for the flow of people indicated by the sample data j).

[0086] The driving mode of the sample data j determined in step S16 (i.e., the driving mode determined for the sample data j) is stored in the storage unit 115. Specifically, the driving mode of the sample data j is added to the feature data corresponding to the sample data j (people flow data), for example, as shown in FIG. 11. In the example shown in FIG. 11, the driving mode of the sample data j is shown to be driving mode X. Although not shown, for example, (data indicating) the driving mode of the sample data j may be added to the sample data j (people flow data) stored in the storage unit 115.

[0087] When the process of step S16 is executed, it is determined whether or not the number of sample data has reached a predetermined number (that is, whether or not the operation mode for the predetermined number of sample data has been determined) (step S17).

[0088] If it is determined that the number of sample data has not reached the predetermined number (NO in step S17), the process returns to step S12 and is repeated. In this case, people flow data different from the sample data acquired in step S12 is acquired (sampled) as new sample data, and the process from step S13 onwards is executed.

[0089] On the other hand, if it is determined that the number of sample data has reached the predetermined number (YES in step S17), the recommended data generating unit 114 determines the operation mode of cluster i (step S18).

[0090] In step S18, the driving mode of each of a predetermined number of sample data sampled from the plurality of people flow data classified into cluster i is acquired from the feature data corresponding to the sample data stored in the storage unit 115, and the driving mode of cluster i is determined based on the acquired driving mode. In this case, the driving mode of cluster i can be, for example, the driving mode with the highest appearance frequency among the driving modes of the predetermined number of sample data (i.e., determined by majority vote).

[0091] Here, we have explained that the processing of step S18 is executed when it is determined that the number of sample data has reached a predetermined number. However, if, for example, the number of people flow data classified into cluster i is less than the predetermined number (i.e., the amount of data of the people flow data is less than the predetermined amount), the processing of steps S12 to S16 can be executed for all of the people flow data, and then the processing of step S18 can be executed.

[0092] After the process of step S18 is executed, it is determined whether or not the process has been executed for all clusters (that is, whether or not the operation modes for all clusters have been determined) (step S19).

[0093] If it is determined that the process has not been executed for all clusters (NO in step S19), the process returns to step S11 and is repeated. In this case, a cluster for which the process has not been executed is selected in step S11, and the process from step S12 onwards is executed.

[0094] On the other hand, if it is determined that the processing has been performed for all clusters (YES in step S19), the recommendation data generation unit 114 generates recommendation data indicating the operation modes of each of the clusters determined in step S18 (step S20).

[0095] FIG. 12 shows an example of the data structure of the recommended data generated by executing the above-described recommended data generation process.

[0096] As shown in FIG. 12, the recommendation data is, for example, data in a table format indicating the relationship between the cluster number assigned to a cluster and the operation mode of the cluster, and includes the cluster number and the operation mode in association with each other.

[0097] 12, the recommended data includes a cluster number "1" and a driving mode "driving mode X." This indicates that the driving mode suitable for the people flow indicated by the people flow data classified into the cluster to which cluster number "1" is assigned is driving mode X (i.e., driving mode X should be recommended for the people flow data classified into that cluster).

[0098] The recommended data also includes a cluster number "2" and a driving mode "driving mode Y." This indicates that the driving mode suitable for the flow of people indicated by the people flow data classified into the cluster to which cluster number "2" is assigned is driving mode Y (i.e., driving mode Y should be recommended for the people flow data classified into that cluster).

[0099] Furthermore, the recommendation data includes a cluster number "3" and a driving mode "driving mode Z." This indicates that the driving mode suitable for the people flow indicated by the people flow data classified into the cluster to which cluster number "3" is assigned is driving mode Z (i.e., driving mode Z should be recommended for the people flow data classified into that cluster).

[0100] Here we have explained the driving modes of clusters assigned cluster numbers "1" to "3", but the recommended data also includes cluster numbers assigned to other clusters in which past pedestrian traffic data has been classified, and the driving modes of those clusters.

[0101] As described above, the driving mode of a cluster is determined by majority vote of the driving modes of the sample data sampled from the cluster (among the plurality of people flow data classified into the cluster), so the recommended data may further include additional information indicating the bias of the driving modes of the sample data (the degree to which the driving modes are dispersed), etc. In the example shown in Fig. 12, the additional information indicates, for example, the proportion of driving modes determined as the driving mode of the cluster out of the driving modes of a predetermined number of sample data.

[0102] The recommendation data generated by executing the recommendation data generation process in this manner is output (transferred) from the learning device 11 to the recommendation device 12 and stored in the storage unit 125 included in the recommendation device 12.

[0103] Incidentally, in the recommended data generation process shown in FIG. 10, the process of step S18 is executed when it is determined in step S17 that the number of sample data has reached a predetermined number (i.e., the operation mode of the cluster is determined based on the predetermined number of sample data), but the number of sample data used to determine the operation mode of the cluster (i.e., the timing at which sampling of the sample data is ended) may be changed.

[0104] Specifically, the processing of step S18 may be executed, for example, even if the number of sample data has not reached a predetermined number, when the processing time for cluster i (the processing time for determining the operation mode of cluster i) exceeds a predetermined time or when the ratio of the number of sample data to the number of multiple people flow data classified into cluster i (hereinafter referred to as the ratio of sample data to cluster) reaches a predetermined ratio.

[0105] Furthermore, the number of sample data used to determine the operating mode of a cluster may be changed based on the accuracy of the operating mode of the cluster.

[0106] Here, an example of the processing procedure for the recommended data generation process when the number of sample data is changed based on the accuracy of the operation mode of the cluster as described above will be described with reference to Fig. 13. Here, only the differences from the recommended data generation process shown in Fig. 10 will be described.

[0107] 13, when the process of step S18 is executed, it is determined whether the accuracy of the driving mode of cluster i determined in step S18 is high (step S21). If it is determined that the accuracy of the driving mode of cluster i is high (YES in step S21), the process of step S19 described above is executed. On the other hand, if it is determined that the accuracy of the driving mode of cluster i is not high (NO in step S21), the process returns to step S12 and is repeated.

[0108] In step S21, for example, when the multiple driving modes determined in step S18 by repeatedly executing the processes of steps S12 to S18 are the same driving mode or a predetermined percentage or more of the multiple driving modes are the same driving mode (that is, the deviation of the driving modes of the multiple clusters is small and the degree of agreement is high), it is determined that the accuracy of the driving mode of cluster i is high (the driving mode of cluster i has converged).When it is determined that the accuracy of the driving mode of cluster i is high in this way, the driving mode determined to be high accuracy (the converged driving mode) is adopted as the driving mode of cluster i.

[0109] Here, the conditions for determining when to end sampling of sample data (hereinafter referred to as sampling end conditions) have been explained as follows: the sample data reaches a predetermined number (hereinafter referred to as the first condition), the processing time exceeds a predetermined time (hereinafter referred to as the second condition), the ratio of sample data to the cluster exceeds a predetermined ratio (hereinafter referred to as the third condition), and the accuracy of the cluster's operating mode is high (hereinafter referred to as the fourth condition).However, the sampling end condition may be any one of the first to fourth conditions, or may be a combination of at least two of the first to fourth conditions.

[0110] In the recommended data generation process shown in FIG. 10, steps S11 to S18 are repeatedly executed for all clusters. However, because the people flow data classified into each cluster and the feature data corresponding to the people flow data (data within the cluster) are independent, steps S11 to S18 may be parallelized. This configuration reduces the processing time (computation time) required for the recommended data generation process. While the parallelization of steps S11 to S18 for each cluster has been described, steps S12 and S13 for each driving mode may also be parallelized. Furthermore, in the recommended data generation process shown in FIG. 10, steps S12 to S16 are repeatedly executed for a predetermined number of sample data. However, a configuration is also possible in which a predetermined number of sample data are acquired in step S12, and steps S13 to S16 are executed for each of the sample data.

[0111] The above-mentioned learning process (the process shown in Figure 6) only needs to be executed at least once to generate recommended data, but if, for example, people flow data is added to the storage unit 115, the recommended data may be updated by executing the learning process again using the people flow data (i.e., the learning process may be executed periodically).

[0112] Next, an example of the processing procedure of (recommended processing executed by) the recommended device 12 will be described with reference to the flowchart in Fig. 14. Note that the processing shown in Fig. 14 may be executed, for example, according to an instruction (i.e., an external request) from an administrator of the recommendation system 1 (recommended device 12), or may be executed when a predetermined condition, such as reaching a predetermined time, is met.

[0113] Here, it is assumed that the recommended data output from the learning device 11 (the recommended data generating unit 114 included in the learning device 11) is stored in the storage unit 125.

[0114] First, the data acquisition unit 121 acquires current people flow data (people flow data indicating the current flow of people) collected in the people flow system 2 (step S31).

[0115] The current people flow data acquired in step S31 may be people flow data collected in the people flow system 2 at a time later than the time when the past people flow data used in the learning process performed by the learning device 11 described above was collected, and is not limited to people flow data collected in the people flow system 2 at the current time, for example.

[0116] The data structure of the current people flow data acquired in step S31 is the same as that explained in Fig. 7, and therefore a detailed explanation thereof will be omitted here. However, while multiple people flow data are acquired as past people flow data in step S1 shown in Fig. 6, for example, one piece of people flow data is acquired as current people flow data in step S31.

[0117] In addition, in step S31, it has been described that people flow data (OD table) is acquired from the people flow system 2, but the data acquisition unit 121 may be configured to acquire data for creating the people flow data from the people flow system 2 and generate (acquire) the people flow data based on the acquired data. Note that, since the recommendation process does not require a simulation as described below (i.e., it is not necessarily required to acquire people flow data), it is sufficient that data from which features can be acquired (extracted) is acquired in step S31. This may enable the amount of data acquired by the data acquisition unit 121 to be reduced.

[0118] When the process of step S31 is executed, the feature extractor 122 extracts features from the data acquired in step S31 (for example, current people flow data) (step S32).

[0119] 6, feature amounts are extracted from each piece of past people flow data, but the process of step S32 is the same as the process of step S2 except that the past people flow data in step S2 is used as the current people flow data. Therefore, a detailed description of the process of step S32 will be omitted here.

[0120] Furthermore, when normalizing the feature in step S32, the parameter value for the normalization (e.g., total traffic volume, etc.) may be determined (calculated) by the feature extraction unit 122 at the time the processing of step S32 is executed, or may be stored in the storage unit 125 as a preset value (fixed value) in advance.

[0121] When the process of step S32 is executed, the cluster determination unit 123 executes a process of determining a cluster into which the current people flow data acquired in step S31 is classified based on feature data including the feature extracted in step S32 (hereinafter referred to as a cluster determination process) (step S33). The cluster determination process executed in step S33 corresponds to a process of determining a cluster that is closest to the current people flow data from among multiple clusters into which past people flow data was classified during the learning process. Details of this cluster determination process will be described later.

[0122] Next, the operation mode determination unit 124 determines an operation mode suitable for the flow of people indicated by the current people flow data acquired in step S31 (i.e., an operation mode recommended for the elevator system 3) based on the recommended data stored in the storage unit 125 and the cluster determined in step S33 (cluster determination process) (step S34). In step S34, the operation mode included in the recommended data in association with the cluster number assigned to the cluster determined in the cluster determination process is determined as the operation mode suitable for the flow of people indicated by the current people flow data.

[0123] The operation mode (data indicating the operation mode) determined in step S34 is output (transmitted) to the elevator system 3 (step S35).

[0124] By executing the recommendation process shown in FIG. 14 described above, the elevator system 3 (elevator control device) can control the operation of the elevator in the operation mode output from the recommendation device 12.

[0125] Here, it is assumed that the operation mode determined in step S34 is automatically applied to the elevator system 3 (i.e., the operation mode of the elevator is automatically changed), but the operation mode may be presented to an administrator who manages the elevator system 3. In this case, the administrator can refer to the operation mode output from the recommendation device 12 (the operation mode determination unit 124 included in the recommendation device 12) and determine whether or not to change the operation mode for controlling the operation of the elevator.

[0126] Note that, when additional information is included in the recommended data as described above, the additional information included in the recommended data may be output (presented to the administrator) together with the operation mode to the elevator system 3 in association with the cluster number assigned to the cluster determined in the cluster determination process described above. In this case, the administrator can determine whether to change the operation mode by further referring to the additional information output from the recommendation device 12 (for example, bias in the operation modes of the sample data used when the operation mode of the cluster was determined in the recommended data generation process described above).

[0127] Next, an example of the processing procedure of the above-mentioned cluster determination processing will be described with reference to the flowchart of FIG.

[0128] The cluster determination unit 123 selects one cluster from among the multiple clusters into which past people flow data has been classified in the above-mentioned learning process (step S41). As described above, the storage unit 115 included in the learning device 11 stores a cluster model that holds data necessary for clustering, such as multiple cluster numbers and centroids of clusters to which the cluster numbers are assigned. However, it is assumed that the storage unit 125 included in the recommendation device 12 stores a cluster model identical to the cluster model. In step S41, one cluster to which one of the multiple cluster numbers held in such a cluster model is assigned is selected. Hereinafter, the cluster selected in step S41 will be referred to as cluster i.

[0129] Next, the cluster determination unit 123 calculates the distance between the current people flow data acquired in step S31 shown in FIG. 14 and the cluster i (step S42).

[0130] Here, the above-described cluster model holds the centroid of each cluster. If the feature data corresponding to the current people flow data is assumed to be vector data having, for example, feature quantities 1 to 5 as components as described above in FIG. 8, the centroid of cluster i is assumed to be data indicating the center value of cluster i using the same number of components as feature quantities 1 to 5. In step S42, the distance between the current people flow data corresponding to the feature data and cluster i is calculated based on the difference between corresponding components of such feature data and the centroid of cluster i. Specifically, the distance between the current people flow data and cluster i is calculated using, for example, the L2 norm.

[0131] The distance between the current people flow data and cluster i calculated in step S42 is held inside the cluster determination unit 123.

[0132] When the process of step S42 is executed, it is determined whether or not the process has been executed for all clusters (that is, whether or not the distance from the current people flow data has been calculated for all clusters) (step S43).

[0133] If it is determined that the process has not been executed for all clusters (NO in step S43), the process returns to step S41 and is repeated. In this case, a cluster for which the process has not been executed is selected in step S41, and the process of step S42 is executed.

[0134] On the other hand, if it is determined that the processing has been performed for all clusters (YES in step S43), the processing of steps S41 and S42 is repeatedly performed, and the distances to the current people flow data for all clusters are stored inside the cluster determination unit 123. In this case, the cluster determination unit 123 determines the cluster that is the shortest distance from the current people flow data based on the distances stored inside the cluster determination unit 123 (step S44).

[0135] In the cluster determination process shown in Figure 15, the cluster closest to the current people flow data can be determined using feature data including features extracted from the current people flow data and the centroids of each cluster held in the cluster model.

[0136] In this embodiment, as described above, an operating mode suitable for the flow of people indicated by the current people flow data is recommended to the elevator system 3, and the operating mode that controls the operation of the elevator can be automatically changed. However, in order to achieve stable elevator operation in the elevator system 3, it is not desirable to frequently change the operating mode.

[0137] However, if the above-mentioned recommended processing is performed at relatively short intervals (i.e., in real time), the operating mode may change frequently (i.e., fluctuate over time), which may hinder stable operation of the elevator.

[0138] Therefore, in this embodiment, a configuration may be adopted in which a process (hereinafter referred to as chattering removal process) is executed to suppress such frequent changes in the driving mode (to reduce fluctuations in the driving mode over time).

[0139] One example of chattering removal processing is to adjust the distance between the current people flow data calculated in step S42 and cluster i based on, for example, the current elevator operation mode. According to this, if the current elevator operation mode would be changed when cluster i is determined in step S44, for example, a penalty is applied to the distance calculated for cluster i (for example, by adding a predetermined distance to the distance), thereby making it possible to prevent the current elevator operation mode from being easily changed.

[0140] Specifically, for example, if the distance between the current people flow data and cluster 1 (the cluster with cluster number 1) is distance 1, the distance between the current people flow data and cluster 2 (the cluster with cluster number 2) is distance 2, and the current elevator operation mode is operation mode X (the operation mode of cluster 1), distance α is added to distance 2 as a penalty. According to this, even if distance 1 is longer than distance 2, for example, if distance 1 is shorter than distance 2 + α, cluster 1 is determined as the cluster closest to the current people flow data (i.e., operation mode X, which is the operation mode of cluster 1, is recommended to the elevator system 3), thereby preventing the elevator operation mode from being changed. Note that the penalty is assumed to be stored in advance as a set value in storage unit 125.

[0141] Furthermore, as another example of chattering removal processing, it is possible to configure the system so that the distance calculated for the cluster corresponding to the current elevator operation mode (i.e., the cluster to which the cluster number included in the recommendation data is assigned in correspondence with that operation mode) is accumulated each time the recommendation processing is executed, and if the accumulated distance exceeds a threshold value, the current elevator operation mode is changed (i.e., an operation mode different from the current elevator operation mode is recommended).

[0142] The current elevator operation mode described above can be obtained from the elevator system 3 (the elevator control device provided therein), but the current elevator operation mode may be, for example, an operation mode determined in the past (immediately before) by the recommendation device 12 (an operation mode output to the elevator system 3).

[0143] Furthermore, the chattering removal process may be performed using, for example, the distance between multiple clusters (the distance calculated based on the centroid of each of the multiple clusters), the distance between the current people flow data and each of the multiple clusters, the ratio of the distances, the sum of the distances or ratios, or the integral of the distances or ratios.

[0144] 14 assumes a case where a driving mode is recommended in real time based on the current people flow data (or feature data extracted from the people flow data) acquired in step S31, but the recommendation process may also be executed for each of a day's worth of people flow data (a plurality of people flow data). According to this recommendation process, by determining a driving mode for each of a day's worth of people flow data, it becomes possible to create (recommend) a driving mode schedule for the entire day (i.e., a driving mode schedule for the next day, etc.).

[0145] As described above, in this embodiment, the recommendation device 12 stores in the storage unit 125 recommended data including the operation mode to be recommended in association with each of a plurality of clusters into which people flow data indicating the flow of people using the elevator (movement of users) is classified, acquires current people flow data (first people flow data collected in the people flow system 2 in the first time period), extracts features (first features) from the acquired people flow data, determines the cluster into which the current people flow data is classified based on the extracted features (feature data including the features), and outputs the operation mode included in the recommended data in association with the determined cluster.

[0146] The recommendation data stored in the storage unit 125 is generated by the learning device 11. Specifically, the learning device 11 extracts features (second features) from each piece of past people flow data (second people flow data collected in the people flow system 2 in a second time slot prior to the first time slot), classifies each piece of past people flow data into multiple clusters based on the extracted features (feature data including the features), samples some of the people flow data classified into each of the multiple clusters as sample data, and determines a driving mode to be recommended for each cluster based on the sample data, thereby generating recommendation data including the driving mode in association with the cluster.

[0147] In this embodiment, such a configuration makes it possible to recommend an appropriate operation mode in response to changes in demand for elevators (changes in the flow of people indicated by the current people flow data).

[0148] Incidentally, for example, a configuration for recommending an elevator operation mode (hereinafter referred to as a first comparative example of this embodiment) may involve, before recommending the elevator operation mode, simulating elevator operation when each of the operation modes is applied and determining the operation mode to be recommended based on the results of the simulation. However, in this first comparative example of this embodiment, it is necessary to simulate elevator operation each time a recommended operation mode is determined, which raises concerns about increased computational costs and computational time associated with the process of recommending the operation mode (recommendation process). When actually operating the system, it is preferable to minimize computational costs and computational time in accordance with constraints imposed by control time and hardware computational resources.

[0149] In contrast, in this embodiment, the driving mode to be recommended is determined using the recommendation data generated in advance by the learning device 11 as described above (i.e., no simulation is required in the recommendation process for recommending a driving mode), so the calculation cost and calculation time related to the recommendation process can be reduced.

[0150] Furthermore, as a second comparative example of this embodiment, it is conceivable to generate recommended data in advance by simulating elevator operation when each operation mode is applied to all past people flow data. However, if the number of past people flow data is large, the calculation cost and calculation time (i.e., simulation cost) related to the process of generating recommended data (learning process) will increase.

[0151] In contrast, in this embodiment, simulations are performed by thinning out past people flow data (i.e., sampling sample data from past people flow data), thereby reducing the number of simulations and, as a result, reducing the computational costs and time involved in the process of generating recommended data.

[0152] Furthermore, if the computational cost and computational time were reduced by, for example, randomly sampling past people flow data to generate recommended data, there is a concern that the sampled people flow data would be biased, reducing the accuracy of the generated recommended data. However, in this embodiment, clustering is performed to pre-group people flow data sets with similar feature trends, and recommended data is generated using sample data (people flow data with similar feature trends) sampled from each of the groups (clusters). This makes it possible to reduce computational cost and computational time while maintaining the accuracy of the recommended data.

[0153] In this embodiment, the sample data is data that satisfies predetermined conditions among the plurality of people flow data classified into each of the plurality of clusters. In this embodiment, such a configuration makes it possible to exclude people flow data that corresponds to an outlier, for example, from sampling, thereby improving the accuracy of the driving mode determined based on the sample data (recommended data generated based on the sample data).

[0154] Furthermore, in this embodiment, the operating mode included in the recommendation data (i.e., the operating mode to be recommended to the elevator system 3) is determined based on, for example, a predetermined number (predetermined number) of sample data, but the number of sample data may be changed (determined) based on at least one of a predetermined processing time, the ratio of the number of sample data to the number of people flow data classified into clusters (ratio of sample data to clusters), and the accuracy of the recommended operating mode. With such a configuration, even if the number of sample data has not reached the predetermined number, for example, if the processing time exceeds a predetermined time (predetermined processing time), the ratio of sample data to clusters exceeds a predetermined ratio, or the accuracy of the recommended operating mode (determined operating mode) is high, it becomes possible to terminate sampling of the sample data, which may further reduce calculation costs and calculation time.

[0155] Furthermore, in this embodiment, a portion of the people flow data classified into each of the multiple clusters is sampled as sample data, but this sampling may be performed when the amount of people flow data classified into the cluster is greater than a predetermined amount (a predetermined value).

[0156] In this embodiment, clustering is performed on past people flow data to classify the past people flow data into multiple clusters, and the number of clusters is determined based on the number of elevator operation modes. In this case, for example, it is possible to set the number of clusters to the same number as the number of elevator operation modes. With this configuration, for example, recommendation data including different operation modes associated with each of the multiple clusters is generated, making it possible to recommend an appropriate operation mode for each cluster.

[0157] Although the case where the number of clusters is the same as the number of elevator operation modes has been described here, the number of clusters does not have to match the number of elevator operation modes.

[0158] Here, the above-described recommendation data generation process may result in, for example, two clusters having common driving modes (i.e., recommendation data including the same driving mode is generated in association with the two clusters), but in this case, if the number of clusters is the same as the number of driving modes, a situation may arise in which at least one driving mode among the multiple driving modes is not recommended. On the other hand, if the number of clusters is greater than the number of driving modes, it may be possible to generate recommendation data that can recommend all driving modes even when the driving modes of two clusters are common.

[0159] Even if two clusters share the same driving mode, the additional information included in the recommended data associated with each cluster (bias in the driving modes of the sample data used when the driving modes of the clusters were determined in the recommended data generation process described above) is likely to be different. For this reason, this embodiment can be configured to recommend the same driving mode with different additional information depending on the current people flow data (the cluster determined based on the data).

[0160] In addition, in this embodiment, the recommended operating mode (the operating mode included in the recommendation data in association with a cluster) is determined based on an index (e.g., at least one of energy consumption, performance index, and comfort index) obtained by simulating elevator operation based on sample data sampled from the cluster (among multiple people flow data classified into the cluster). Note that, for example, when multiple sample data (e.g., a predetermined number of sample data) are sampled from a cluster, the recommended operating mode is determined based on the operating mode determined for each of the multiple sample data based on the above-mentioned index (e.g., the result of a majority vote for the operating mode). In this embodiment, with this configuration, an appropriate operating mode corresponding to each cluster can be determined.

[0161] In this embodiment, the recommended operation mode is determined based on an index (simulation result) obtained by simulating elevator operation when each of a plurality of operation modes is applied to sample data. However, the recommended operation mode may be determined based on an index obtained by actually operating the elevators provided in the elevator system 3 based on the sample data (i.e., an index obtained by measurement using the actual equipment) instead of the simulation result, or an index obtained from performance data showing the past operational performance of the elevator system 3 searched based on the sample data (operational performance of the elevators in response to a flow of people of a similar level to that of the sample data). The recommended operation mode may also be determined by combining these indexes.

[0162] Furthermore, in this embodiment, the distance between the current people flow data and each of the multiple clusters is calculated based on the features (including feature data) extracted from the current people flow data and the centroids of each of the multiple clusters (the centroids for each cluster based on the features extracted from each of the people flow data classified into each of the multiple clusters), and the cluster into which the current people flow data is classified (the cluster closest to the current people flow data) is determined based on the calculated distance. However, the distance may be adjusted based on, for example, the current operating mode of the elevator (i.e., a chattering removal process may be performed). With this configuration, it is possible to suppress fluctuations in the operating mode over time, such as when the elevator operating mode is frequently changed.

[0163] In this embodiment, it is assumed that an operation mode of an elevator provided in the elevator system 3 is recommended, and the people flow data collected by the people flow system 2 includes, for example, the departure floor and arrival floor of each of multiple users who used the elevator. Furthermore, in this embodiment, the feature values ​​extracted from the people flow data include, for example, at least one of the following: the number of users moving in the upward direction identified based on the departure floor and arrival floor of each of the multiple users included in the people flow data; the number of users moving in the downward direction; the number of users whose departure and arrival floors are other than a predetermined reference floor; the number of users moving in the upward direction exceeding the number of users moving in the downward direction; and the number of users moving in the downward direction exceeding the number of users moving in the upward direction. In this embodiment, by extracting the feature values ​​(feature value data including the feature values) described above, it is possible to appropriately grasp changes in demand for elevators and recommend elevator operation modes in accordance with the changes in demand.

[0164] The above-mentioned features (at least one of the number of users moving in the upward direction, the number of users moving in the downward direction, the number of users whose departure and arrival floors are other than a predetermined reference floor, the number of users moving in the upward direction exceeding the number of users moving in the downward direction, and the number of users moving in the downward direction exceeding the number of users moving in the upward direction) may be normalized using a parameter value such as total traffic volume. Using feature data including such normalized features can improve the accuracy of clustering for people flow data corresponding to the feature data. Furthermore, recommendation data generated using feature data including normalized features can also be used to recommend elevator operation modes for buildings, for example, of a different size from the target building where people flow data (past people flow data) corresponding to the feature data was collected.

[0165] In this embodiment, the recommendation system 1 has been mainly described as recommending the operating mode of an elevator provided in the elevator system 3, but the recommendation system 1 may also be applied to recommending, for example, the operating mode of an air conditioner.

[0166] A recommendation system that recommends an operation mode for an air conditioner will be briefly described below as a modification of this embodiment with reference to FIG.

[0167] In the above-mentioned Figure 1, the recommendation system 1 is described as being communicatively connected to the people flow system 2 and the elevator system 3 via the network 4, but the recommendation system 1 in a modified example of this embodiment is communicatively connected to the people flow system 2 and the air conditioning system 5 via the network 4.

[0168] In a modified example of this embodiment, the people flow system 2 is configured to collect (aggregate) and output people flow data indicating the flow of people using the air conditioners installed in the target building at any timing. Note that, as described above, the people flow data collected by the people flow system 2 when the recommendation system 1 recommends an elevator operation mode has been explained as including the departure floor and arrival floor of each user using the elevator, but the people flow data collected by the people flow system 2 when the recommendation system 1 recommends an air conditioner operation mode may be data including, for example, the number or density of users for each floor (or area) where the air conditioner is installed.

[0169] The air conditioning system 5 includes air conditioners installed in the target building and air conditioner control devices that control the operation of the air conditioners. In the air conditioning system 5, the air conditioner control devices control the operation of the air conditioners by switching between, for example, multiple operation modes prepared in advance.

[0170] The air conditioners provided in the air conditioning system 5 have a ventilation function in addition to an air conditioning function for adjusting the temperature of the air in the room, and the air conditioning system 5 is assumed to have a plurality of operating modes with different ventilation ratios in the ventilation function (for example, a plurality of operating modes each assigned with a discrete ventilation ratio such as 0%, 10%, 20%, ...). The ventilation ratio is assumed to be, for example, the ratio of the amount of air flowing into the room per unit time to the volume of the room in which the air conditioner is installed, but may be another value.

[0171] Furthermore, if the air conditioning system 5 includes a plurality of air conditioners, a plurality of operation modes may be provided in which a share (discretized share) of the ventilation function of each of the plurality of air conditioners is assigned.

[0172] The operation of the recommendation system 1 relating to a modified example of this embodiment differs from that of the present embodiment described above in terms of people flow data and operation mode (contents), but other points are generally the same as those described in this embodiment, so detailed explanations will be omitted here.

[0173] In addition, in a modified example of this embodiment, the feature extracted from the people flow data may be the number or density of users included in the people flow data as described above, or may be a value identified based on the number or density of users (a value calculated using a specified calculation formula, etc. to which the number or density of users is applied).

[0174] That is, the recommendation system 1 according to a modified example of this embodiment is capable of recommending the operating mode of the air conditioner provided in the air conditioning system 5 based on the recommendation data generated by the learning device 11 as described in the above-described embodiment.

[0175] Although the present embodiment and the modified example of the present embodiment have described the recommendation system 1 that recommends operation modes for elevators and air conditioners, the recommendation system 1 can also be used to recommend operation modes for controlled objects including building facilities other than elevators and air conditioners. In other words, the elevators and air conditioners described in the present embodiment and the modified example of the present embodiment are examples of controlled objects whose operation is controlled in the operation mode recommended by the recommendation system 1.

[0176] Furthermore, in this embodiment and the modified example of this embodiment, the operation modes of elevators and air conditioners (i.e., the operation modes of the controlled object) have been described as being recommended based on people flow data, but the people flow data may be data that can grasp changes in demand for the controlled object (i.e., usage data related to the use of the controlled object). That is, in this embodiment and the modified example of this embodiment, instead of the people flow data described above, data such as an operation log of the controlled object (e.g., operation data) may be used, or data related to the environment inside or outside the target building when the controlled object is used (e.g., temperature data, etc.) may be used.

[0177] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their modifications are included within the scope and spirit of the invention, as well as within the scope of the invention described in the claims and their equivalents. [Explanation of symbols]

[0178] 1...Recommendation system, 2...People flow system, 3...Elevator system, 4...Network, 5...Air conditioning system, 11...Learning device, 11a...CPU, 11b...Non-volatile memory, 11c...RAM, 11d...Communication device, 11e...Learning program, 12...Recommendation device, 12a...CPU, 12b...Non-volatile memory, 12c...RAM, 12d...Communication device, 12e...Recommendation program, 111...Data acquisition unit, 112...Feature extraction unit, 113...Clustering unit, 114...Recommended data generation unit, 115...Storage unit, 121...Data acquisition unit, 122...Feature extraction unit, 123...Cluster determination unit, 124...Operating mode determination unit.

Claims

1. In a recommendation system that recommends an operating mode of a controlled object, a storage means for storing recommendation data including a recommended operation mode among a plurality of operation modes of the control object, the operation modes being prepared in advance in association with each of a plurality of clusters into which usage data relating to the use of the control object is classified; an acquisition means for acquiring first usage data relating to usage of the controlled object during a first time period; extraction means for extracting a first feature amount from the acquired first usage data; a determining means for determining a cluster into which the first usage data is classified based on the extracted first feature amount; an output means for outputting the operation mode included in the recommendation data in association with the determined cluster; A recommendation system comprising:

2. The recommendation system of claim 1, wherein the recommendation data is generated by extracting a second feature from each of a plurality of second usage data relating to the use of the controlled object during a second time period prior to the first time period, classifying each of the plurality of second usage data into the plurality of clusters based on the extracted second feature, sampling a portion of the second usage data classified into each of the plurality of clusters as sample data, and determining the recommended operating mode for each of the clusters based on the sample data.

3. 3. The recommendation system according to claim 2, wherein the sample data is data that satisfies a predetermined condition among the second usage data classified into each of the plurality of clusters.

4. 4. The recommendation system according to claim 3, wherein the recommended operation mode is determined based on a predetermined number of sample data.

5. 5. The recommendation system of claim 4, wherein the number of sample data is changed based on at least one of a processing time for determining the operation mode to be recommended, a ratio of the number of sample data to the number of second usage data, and accuracy of the operation mode to be recommended.

6. The recommendation system of claim 2, wherein when the amount of second usage data classified into each of the plurality of clusters is greater than a predetermined amount, a portion of the second usage data classified into that cluster is sampled as sample data.

7. The recommendation system according to claim 2 , wherein the number of clusters is determined based on the number of operation modes of the controlled object.

8. The recommendation system of claim 2, wherein the recommended operating mode is determined based on an index obtained by simulating the operation of the control object based on the sample data, an index obtained by actually operating the control object based on the sample data, or an index obtained from performance data showing past operational performance of the control object according to the sample data.

9. The recommendation system of claim 8, wherein when a plurality of sample data are sampled from the second usage data classified into each of the plurality of clusters, the recommended operating mode is determined based on the operating mode determined for each of the plurality of sample data based on the index.

10. the determining means calculates a distance between the first usage data and each of the plurality of clusters based on the extracted first feature amount and a centroid for each cluster based on the second feature amount extracted from each of the second usage data classified into each of the plurality of clusters, and determines a cluster into which the first usage data is classified based on the calculated distance; The distance is adjusted based on the current operating mode of the controlled object. The recommendation system of claim 2.

11. the controlled object includes an elevator, the first and second usage data include a departure floor and an arrival floor of each of a plurality of users who have used the elevator; The first and second feature amounts include at least one of the number of users moving in an upward direction specified based on the departure floor and arrival floor of each of the plurality of users included in the first and second usage data, the number of users moving in a downward direction, the number of users whose departure floor and arrival floor are other than a predetermined reference floor, the number of users moving in an upward direction exceeding the number of users moving in a downward direction, and the number of users moving in a downward direction exceeding the number of users moving in an upward direction. The recommendation system of claim 2.

12. The recommendation system according to claim 11 , wherein the first and second feature amounts are normalized.

13. the controlled object includes an elevator, The index includes at least one of an energy consumption index calculated based on the power consumption of the elevator when all users are moved based on the sample data, a performance index calculated based on the waiting time of each user and the service time for each user, and a comfort index calculated based on the load inside the elevator. The recommendation system of claim 8.

14. A recommendation method executed by a recommendation system that recommends an operation mode of a controlled object, comprising: storing recommendation data in a storage device, the recommendation data including a recommended operation mode from among a plurality of operation modes of the control object, the operation modes being prepared in advance in association with each of a plurality of clusters into which usage data relating to the use of the control object is classified; acquiring first usage data relating to usage of the controlled object during a first time period; extracting a first feature amount from the acquired first usage data; determining a cluster into which the first usage data is classified based on the extracted first feature amount; outputting the operation mode included in the recommendation data in association with the determined cluster; A recommended method includes:

15. The recommendation method described in claim 14, wherein the recommended data is generated by extracting a second feature from each of a plurality of second usage data related to the usage of the controlled object during a second time period prior to the first time period, classifying each of the plurality of second usage data into the plurality of clusters based on the extracted second feature, sampling a portion of the second usage data classified into each of the plurality of clusters as sample data, and determining the recommended operating mode for each of the clusters based on the sample data.

16. A program executed by a computer of a recommendation system that recommends an operation mode of a control object, The computer, storing recommendation data in a storage device, the recommendation data including a recommended operation mode from among a plurality of operation modes of the control object, the operation modes being prepared in advance in association with each of a plurality of clusters into which usage data relating to the use of the control object is classified; acquiring first usage data relating to usage of the controlled object during a first time period; extracting a first feature amount from the acquired first usage data; determining a cluster into which the first usage data is classified based on the extracted first feature amount; outputting the operation mode included in the recommendation data in association with the determined cluster; A program to execute.

17. The program of claim 16, wherein the recommended data is generated by extracting a second feature from each of a plurality of second usage data relating to the use of the controlled object in a second time period prior to the first time period, classifying each of the plurality of second usage data into the plurality of clusters based on the extracted second feature, sampling a portion of the second usage data classified into each of the plurality of clusters as sample data, and determining the recommended operating mode for each cluster based on the sample data.

Images