Prediction model construction device and demand forecasting device

Abstract

To improve the accuracy of demand prediction for a vehicle in a sharing transportation service.SOLUTION: A prediction model construction apparatus configured to construct a prediction model for predicting demands for lease and return of vehicles in a sharing transportation service includes: a generation unit which generates feature quantities of multiple ports; a classification unit which clusters the ports based on the feature quantities; and a learning unit which trains a prediction model on the basis of past record data on demands in ports belonging to the same cluster as a target port to be subjected to prediction.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present disclosure relates to a prediction model construction device and a demand prediction device. [Background technology]

[0002] In recent years, the number of users of shared transportation services, in which multiple users share vehicles such as bicycles, has been increasing. In shared transportation services, each user rents a vehicle from a desired port and returns the vehicle to a desired port. At each port, the number of rented vehicles does not necessarily match the number of returned vehicles, so there may be a bias in the number of vehicles parked at the port. In this case, vehicles are reallocated from a port with a large number of vehicles to a port with a small number of vehicles.

[0003] Therefore, it is necessary to accurately predict the demand for vehicles at each port. Patent Document 1 describes a method for forecasting demand by performing machine learning using parameters such as time of day and weather, and historical information on points where actual use has occurred as input. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2018-124974 A Summary of the Invention [Problem to be solved by the invention]

[0005] For example, in regional cities, there are not many users of shared transportation services, so a sufficient amount of actual data cannot be obtained compared to large cities. The number of users of shared transportation services is rapidly increasing, and the usage trends of shared transportation services change over time. Therefore, past usage trends may differ from current usage trends, so past actual data may not be able to be used for demand forecasting. For these reasons, there is a risk that actual data suitable for demand forecasting will be insufficient, and the accuracy of the demand forecasting will decrease.

[0006] The present disclosure describes a prediction model construction device and a demand prediction device that can improve the accuracy of demand prediction for vehicles in shared transportation services. [Means for solving the problem]

[0007] A predictive model construction device according to one aspect of the present disclosure is a device that constructs a predictive model for predicting demand for vehicle rental and return in a shared transportation service. This predictive model construction device includes a generation unit that generates feature quantities for each of a plurality of ports, a classification unit that clusters the plurality of ports based on the feature quantities, and a learning unit that trains a predictive model based on performance data on demand at ports that belong to the same cluster as a target port that is the port to be predicted.

[0008] In this predictive model construction device, multiple ports are clustered based on the features of the multiple ports, and a predictive model is trained based on performance data of ports that belong to the same cluster as the target port. With this configuration, ports with similar features can be classified into the same cluster, so even if the target port does not use the shared transportation service frequently, a sufficient amount of performance data can be obtained by using the performance data of ports that belong to the same cluster as the target port. Therefore, since the predictive model can be trained with a sufficient amount of performance data, it is possible to build a predictive model that can improve the accuracy of vehicle demand forecasts in shared transportation services. Effect of the Invention

[0009] According to the present disclosure, it is possible to improve the accuracy of vehicle demand forecasts in shared transportation services. [Brief description of the drawings]

[0010] [Figure 1] FIG. 1 is a schematic configuration diagram of a demand prediction system including a demand prediction device according to an embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of the port information. [Diagram 3] FIG. 3 is a diagram showing an example of weather information. [Figure 4] FIG. 4 is a diagram illustrating an example of population information. [Diagram 5] FIG. 5 is a block diagram illustrating a functional configuration of the demand prediction device illustrated in FIG. [Figure 6] FIG. 6 is a flowchart showing a series of processes of a forecast model building method performed by the demand forecasting device shown in FIG. [Figure 7] FIG. 7 is a diagram for explaining the correlation coefficient between ports. [Figure 8] FIG. 8 is a flowchart showing the learning process of FIG. 6 in detail. [Figure 9] FIG. 9 is a diagram for explaining input and output of a prediction model. [Figure 10] FIG. 10 is a flowchart showing a series of processes of the demand forecasting method performed by the demand forecasting device shown in FIG. [Figure 11] FIG. 11 is a conceptual diagram of demand forecasting using a forecasting model. [Figure 12] FIG. 12 is a diagram illustrating a hardware configuration of the demand prediction device illustrated in FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. In the description of the drawings, the same elements are given the same reference numerals and duplicated description will be omitted.

[0012] The configuration of a demand prediction system including a demand prediction device according to an embodiment will be described with reference to Figs. 1 to 4. Fig. 1 is a schematic diagram of a demand prediction system including a demand prediction device according to an embodiment. Fig. 2 is a diagram showing an example of port information. Fig. 3 is a diagram showing an example of weather information. Fig. 4 is a diagram showing an example of population information.

[0013] The demand forecasting system 1 shown in FIG. 1 is a system for forecasting demand for vehicles in a shared transportation service. The shared transportation service is a sharing service in which a vehicle is used jointly by multiple users, for example, a one-way MEV (Micro Electric Vehicle) sharing system. Examples of vehicles include bicycles, cars, motorcycles, and scooters. In the shared transportation service, multiple ports are provided. A port is a space (parking lot) for parking vehicles. A port is also called a station.

[0014] Each user rents a vehicle at a desired port within the area, and returns the vehicle at a desired port within the area. The area is an area where a shared transportation service is provided and where vehicles can be used jointly. Examples of areas include cities, wards, towns, and villages. The port to which the vehicle is returned may be different from the port from which the vehicle was rented, or may be the same. Therefore, since the number of vehicles rented out (rented number) and the number of vehicles returned (returned number) do not necessarily match at each port, a bias may occur in the number of vehicles parked at the port (parked number). In the demand forecasting system 1, the number of vehicles rented out and the number of vehicles returned at each port are forecast as demand. This forecast result is used, for example, to reallocate vehicles between ports. The reallocation is performed, for example, by a transport vehicle such as a truck transporting the vehicles between ports.

[0015] The demand forecasting system 1 includes a demand forecasting device 10 and a plurality of servers 30. The demand forecasting device 10 and the plurality of servers 30 are configured to be able to communicate with each other via a network NW. The network NW may be configured as either a wired or wireless network. Examples of the network NW include a mobile communication network, the Internet, and a WAN (Wide Area Network).

[0016] The demand prediction device 10 is a device that predicts demand for rental and return of vehicles in a sharing transportation service. An example of the demand prediction device 10 is an information processing device such as a server.

[0017] Each server 30 is a device that manages information used for demand forecasting. The server 30 stores, for example, port information, weather information, population information, and map image data.

[0018] As shown in FIG. 2, the port information (each record) includes an area ID (identifier), a port ID, date and time information, the number of rented vehicles, the number of returned vehicles, and a mesh code. The area ID is information that can uniquely identify an area and indicates the area to which the port identified by the port ID belongs. In other words, each port belongs to one of multiple areas. The port ID is information that can uniquely identify a port. The date and time information is information that indicates a date and time. The time is expressed in a predetermined time unit (e.g., 10 minutes). In other words, the time indicates a time period. The number of rented vehicles is the number of vehicles rented out from the port identified by the port ID at the date and time indicated by the date and time information. The number of returned vehicles is the number of vehicles returned to the port identified by the port ID at the date and time indicated by the date and time information.

[0019] The mesh code is information that can uniquely identify the mesh to which the port identified by the port ID belongs. A mesh is a divided area obtained by dividing a map according to certain rules. A mesh indicates a lattice-like space that is narrower than an area. For example, the top record in Figure 2 indicates that seven vehicles were loaned and one vehicle was returned between 0:00 and 0:10 on March 1, 2021 at a port identified by port ID "Pa" in an area identified by area ID "A1", and that the port belongs to a mesh identified by mesh code "533910001".

[0020] As shown in FIG. 3, the weather information (each record) includes a mesh code, date and time information, and weather data. The weather data includes, for example, wind speed, temperature, precipitation, and humidity. That is, each record indicates the weather (wind speed, temperature, precipitation, and humidity) at the date and time (time period) indicated by the date and time information in a mesh identified by a mesh code. For example, the top record in FIG. 3 indicates that in a mesh identified by mesh code "533910001", the wind speed is 7 m / s, the temperature is 5 degrees, the precipitation is 0 mm, and the humidity is 10% between 0:00 and 0:10 on March 1, 2021. Note that the weather information includes not only past records but also future records.

[0021] As shown in FIG. 4, the population information (each record) includes a mesh code, date and time information, and population data. The population data includes the population by gender for each age group. In other words, each record indicates the population by gender for each age group at the date and time (time period) indicated by the date and time information for the mesh identified by the mesh code. Note that the population information includes not only past records, but also future records. The future population is estimated using a known method.

[0022] The map image data is map image data of a mesh. The server 30 stores map image data for each mesh. For example, the server 30 stores a mesh code and map image data of the mesh in association with each other.

[0023] Next, the functional configuration of the demand prediction device 10 will be described with reference to Fig. 5. Fig. 5 is a block diagram showing the functional configuration of the demand prediction device shown in Fig. 1. As shown in Fig. 5, the demand prediction device 10 includes a prediction model construction device 50. The prediction model construction device 50 is a device that constructs a prediction model. The prediction model is a machine learning model for predicting demand for vehicle rental and return in a sharing transportation service. The prediction model construction device 50 functionally includes an acquisition unit 11, a generation unit 12, a classification unit 13, a learning unit 14, and a storage unit 15.

[0024] The demand prediction device 10 further includes an acquisition unit 16, a prediction unit 17, and an output unit 18. That is, the demand prediction device 10 functionally includes an acquisition unit 11, a generation unit 12, a classification unit 13, a learning unit 14, a storage unit 15, the acquisition unit 16, the prediction unit 17, and an output unit 18. The function (operation) of each functional unit will be explained in detail in the explanation of the prediction model construction method and the demand prediction method described later, so the function of each functional unit will be briefly explained here.

[0025] The acquisition unit 11 is a functional unit that acquires various information for constructing a prediction model. The acquisition unit 11 acquires port information, weather information, population information, and map image data from the server 30.

[0026] The generating unit 12 is a functional unit that generates feature quantities for each of a plurality of ports. The generating unit 12 generates the feature quantities based on port information, weather information, population information, and map image data.

[0027] The classification unit 13 is a functional unit that clusters multiple ports based on the feature amount of each port. The classification unit 13 classifies multiple ports that are highly correlated with each other into one cluster. In other words, ports that have similar features are classified into the same cluster.

[0028] The learning unit 14 is a functional unit that trains the prediction model. The learning unit 14 trains the prediction model based on performance data of ports that belong to the same cluster as the target port. The target port is a port that is the target of prediction. The performance data is data related to past demand at the port. The learning unit 14 trains the prediction model, for example, by meta-learning using the above-mentioned performance data. The learning unit 14 trains the prediction model for each area. The learning unit 14 uses an error function (loss function) that follows a Poisson distribution as the error function of the prediction model. The error function is a function that evaluates the difference (error) between correct answer data (performance data) and the prediction result (prediction data) by the prediction model. The prediction data is performance data predicted by the prediction model.

[0029] The storage unit 15 is a functional unit that stores (stores) the values ​​of each parameter of the prediction model.

[0030] The acquisition unit 16 is a functional unit that acquires various information for demand forecasting. For example, the acquisition unit 16 acquires past performance data and statistical data for a target port, and statistical data for a target time for the target port. The target time is the time for which demand is to be predicted. The acquisition unit 16 acquires the values ​​of each parameter of the forecast model from the storage unit 15.

[0031] The prediction unit 17 is a functional unit that predicts demand at a target port. The prediction unit 17 predicts demand at the target port by using a prediction model. The prediction unit 17 inputs past performance data and statistical data at the target port and statistical data at the target port at a target time into the prediction model, and obtains a prediction result (prediction data) of demand at the target port at a target time from the prediction model.

[0032] The output unit 18 is a functional unit that outputs the prediction result. For example, the output unit 18 outputs the prediction result to a display and causes the prediction result to be displayed on the display. The output unit 18 may output (transmit) the prediction result to the server 30 and store the prediction result in the server 30.

[0033] Next, a prediction model construction method performed by the demand prediction device 10 will be described with reference to Figs. 6 to 9. Fig. 6 is a flowchart showing a series of processes of the prediction model construction method performed by the demand prediction device shown in Fig. 1. Fig. 7 is a diagram for explaining the correlation coefficient between ports. Fig. 8 is a flowchart showing the learning process of Fig. 6 in detail. Fig. 9 is a diagram for explaining input and output of the prediction model. The series of processes shown in Fig. 6 is started, for example, when a model construction command is received from a user of the demand prediction device 10. These processes may be started repeatedly at predetermined time intervals.

[0034] As shown in FIG. 6, first, the acquisition unit 11 acquires various information for constructing a prediction model (step S11). In step S11, the acquisition unit 11 acquires port information, weather information, population information, and map image data from the server 30. Note that the acquisition unit 11 does not need to acquire all records for the port information, weather information, and population information. For example, the acquisition unit 11 may acquire the latest records for the port information, weather information, and population information for a predetermined number of days from the time when step S11 is performed. Similarly, the acquisition unit 11 does not need to acquire map image data for all meshes. For example, the acquisition unit 11 may acquire map image data corresponding to the mesh code included in the port information. Then, the acquisition unit 11 outputs the acquired various information to the generation unit 12 and the learning unit 14.

[0035] Next, the generating unit 12 generates feature quantities for each port (step S12). In step S12, when the generating unit 12 receives the port information, weather information, population information, and map image data from the acquiring unit 11, the generating unit 12 generates feature quantities based on the port information, weather information, population information, and map image data. In this embodiment, the generating unit 12 generates a feature quantity vector including multiple elements as feature quantities.

[0036] A method for generating a feature vector of one port will be specifically described. First, the generating unit 12 acquires a mesh code of the port from the port information, and extracts a record including the acquired mesh code from the weather information and the population information. Similarly, the generating unit 12 extracts map image data associated with the mesh code. Then, the generating unit 12 generates a partial vector by, for example, arranging the time, each element of the weather data (wind speed, temperature, precipitation, and humidity), and each element of the population data (population by gender in each age group) in a predetermined order for each time. Furthermore, the generating unit 12 arranges the partial vectors in chronological order (oldest first), and adds image feature values ​​obtained by compressing (vectorizing) the map image data. Through the above process, the feature vector of the port is generated. The generating unit 12 generates feature vectors for other ports in the same manner. Then, the generating unit 12 outputs the feature vector of each port to the classifying unit 13.

[0037] Next, the classification unit 13 clusters the multiple ports (step S13). In step S13, when the classification unit 13 receives the feature vector of each port from the generation unit 12, the classification unit 13 calculates the correlation coefficient between two ports using these feature vectors. More specifically, as shown in formula (1), the classification unit 13 clusters the feature vector v i and the feature vector v of the jth port j Using and, the correlation coefficient r between the two ports is calculated. i,j where i and j are integer values ​​between 1 and Np, inclusive, where Np is the total number of ports.

number

[0038] As shown in FIG. 7, the classification unit 13 calculates a correlation coefficient r i,j Then, the classification unit 13 calculates the correlation coefficient r i,j For example, the classifier 13 clusters the Np ports based on the correlation coefficient r i,j is equal to or greater than a predetermined threshold, the i-th port and the j-th port are classified into the same cluster. For example, when the threshold is set to 0.7, in the example of Fig. 7, a port identified by a port ID "Pa", a port identified by a port ID "Pc", and a port identified by a port ID "Pd" are classified into the same cluster. Then, the classification unit 13 outputs the port IDs of the ports belonging to the same cluster to the learning unit 14 for each cluster.

[0039] Next, the learning unit 14 performs learning (step S14). In step S14, the learning unit 14 trains a prediction model for each area, and constructs a prediction model for each area. Hereinafter, a method for constructing a prediction model for an area to which one target port belongs will be described with reference to FIG. 8. In this embodiment, a Transformer-based CVAE (Conditioned Variational Autoencoder) is used as the prediction model.

[0040] As shown in FIG. 8, first, the learning unit 14 prepares performance data and statistical data of ports belonging to the same cluster as the target port as learning data (step S21). In step S21, the learning unit 14 prepares N pieces of learning data for each port. The learning data includes an area ID, a port ID, performance data at time t, and statistical data at time t. The time t is incremented by 1 every time a predetermined time elapses. The area ID is information that uniquely indicates the area to which the port identified by the port ID belongs. The performance data is performance data of the port identified by the port ID. The statistical data is statistical data of the port identified by the port ID. The N pieces of learning data are learning data at N consecutive times.

[0041] The performance data includes the number of vehicles loaned, the number of vehicles returned, the number of vehicles available, and the increase / decrease. The number of vehicles loaned is the number of vehicles loaned from the port between time (t-1) and time t. The number of vehicles returned is the number of vehicles returned to the port between time (t-1) and time t. The number of vehicles available is the number of vehicles available at the port at time t. The increase / decrease is the increase / decrease in the number of available vehicles, and is obtained by subtracting the number of available vehicles at time (t-1) from the number of available vehicles at time t. The performance data is generated based on port information, for example.

[0042] The statistical data includes flag data, weather data, and population data. The flag data includes, for example, a holiday flag indicating whether it is a weekday or a holiday, a day of the week flag indicating the day of the week, and a time flag indicating the time t. The weather data includes the wind speed, temperature, humidity, rainfall, etc. at time t in the mesh to which the port identified by the port ID belongs. The population data includes the population of the mesh to which the port identified by the port ID belongs. The statistical data is generated based on, for example, weather information and population information.

[0043] The learning unit 14 may generate performance data and statistical data, or may obtain performance data and statistical data that have been generated in advance.

[0044] Next, the learning unit 14 calculates the area parameter φ k (Step S22), where k indicates the area number. k is an integer between 1 and K. K is the total number of areas. In step S22, the learning unit 14 optimizes the area parameter φ by using N pieces of learning data for each of the ports that belong to the same cluster as the target port. k Specifically, the learning unit 14 first divides the N pieces of learning data for each port into m pieces of learning data at new times and the remaining learning data, and extracts statistical data contained in the m pieces of learning data.

[0045] For the sake of convenience, m pieces of learning data are set as a data set DSf of learning data from time (t+1) to time (t+m), the remaining learning data are set as a data set DSp of n+1 pieces of learning data from time (tn) to time t, and the statistical data are set as a data set DSst of statistical data from time (t+1) to time (t+m). In other words, N=n+m+1. The data set DSp is set as the result data X t-n ~X t and statistical data S t-n ~S t The data set DSf includes the performance data X t+1 ~X t+m and statistical data S t+1 ~S t+m The data set DSst includes the statistical data S t+1 ~S t+m Includes.

[0046] 9, the learning unit 14 selects one port from the target port and the ports belonging to the same cluster as the target port, inputs the data set DSp, data set DSf, and data set DSst of the port to the prediction model PM, and obtains the data set DSt of prediction data from time (t+1) to time (t+m) from the prediction model PM. The data set DSt is the prediction data X' t+1 ~X' t+mThe learning unit 14 then calculates the area parameter φ so as to reduce the difference (error) between the actual data at each time included in the data set DSf and the predicted data at each time included in the data set DSt. k At this time, the learning unit 14 adjusts the value of the area parameter φ k In this embodiment, the learning unit 14 uses an error function according to the Poisson distribution as the error function of the prediction model, and adjusts the value of the area parameter φ so that the value of the error function becomes small. k In addition, the initial value of the global parameter θ and the area parameter φ are adjusted in the prediction model before learning. k Then, the learning unit 14 repeats the above process until the target port and all ports that belong to the same cluster as the target port are selected.

[0047] Next, the learning unit 14 optimizes the overall parameter θ (step S23). In step S23, the learning unit 14 optimizes the area parameter φ k The value of is set in the prediction model PM. Then, the learning unit 14 inputs the data sets DSp, DSf, and DSst of each port into the prediction model PM, and obtains the data set DSt from the prediction model PM. Then, the learning unit 14 adjusts the value of the overall parameter θ so that the difference (error) between the actual data at each time included in the data set DSf and the predicted data at each time included in the data set DSt becomes small. As in step S22, the learning unit 14 uses an error function according to the Poisson distribution as the error function of the prediction model, and adjusts the value of the overall parameter θ so that the value of the error function becomes small.

[0048] Next, the learning unit 14 determines whether to continue learning (step S24). In step S24, the learning unit 14 determines to continue learning, for example, when the number of times steps S22 and S23 have been processed has not reached a preset number, and determines to end learning, when the number of times steps S22 and S23 have been processed has reached a preset number. The learning unit 14 determines to continue learning, when the amount of change in the overall parameter θ in step S23 is greater than a predetermined threshold, and determines to end learning, when the amount of change in the overall parameter θ in step S23 is smaller than a predetermined threshold. The amount of change in the overall parameter θ is the difference (absolute value of the difference) between the value of the parameter θ before adjustment and the value of the parameter θ after adjustment. Instead of the amount of change in the overall parameter θ, the area parameter φ k A variation of may be used.

[0049] If it is determined in step S24 that learning is to be continued (step S24; YES), the learning unit 14 performs the processes of steps S22 to S24 again. Note that in step S22 from the second time onwards, the learning unit 14 sets the value of the overall parameter θ adjusted in the previous step S23 to the prediction model PM. On the other hand, if it is determined in step S24 that learning is to be ended (step S24; NO), the learning unit 14 ends the learning process of step S14.

[0050] Next, the learning unit 14 outputs each parameter (its value) to the storage unit 15 and stores it in the storage unit 15 (step S15).

[0051] This completes the series of processes in the prediction model construction method. As described above, the area parameter φ k and the global parameter θ are alternately learned. In other words, a prediction model is learned for each area, and a prediction model for each area is constructed. Note that step S23 may be performed before step S22.

[0052] Next, a demand forecasting method performed by the demand forecasting device 10 will be described with reference to Figs. 10 and 11. Fig. 10 is a flowchart showing a series of processes of the demand forecasting method performed by the demand forecasting device shown in Fig. 1. Fig. 11 is a conceptual diagram of demand forecasting using a forecasting model. The series of processes shown in Fig. 10 is started, for example, when a forecast command is received from a user of the demand forecasting device 10. These processes may be started each time time t is incremented by 1. In the following description, the time at which prediction is performed (current time) is assumed to be a time slightly after time t.

[0053] As shown in FIG. 10, first, the acquiring unit 16 acquires performance data and statistical data for a target port (step S31). In step S16, the acquiring unit 16 acquires past performance data and statistical data, and statistical data for a target time for the target port. In this embodiment, the acquiring unit 16 acquires past performance data (performance data X t-n ~X t ) and statistical data (Statistical Data S t-n ~S t ) and statistical data from time (t+1) to time (t+m) (statistical data S t+1 ~S t+m In other words, each time from time (t+1) to time (t+m) is the target time.

[0054] Specifically, the acquisition unit 16 extracts records from the port information stored in the server 30, including a port ID for identifying a target port, from a time (tn) to a time t, and obtains performance data X t-n ~X t The acquisition unit 16 also extracts records from time (tn) to time (t+m) including a mesh code indicating the mesh to which the target port belongs from each of the weather information and population information stored in the server 30. Furthermore, the acquisition unit 16 acquires flag data from time (t+1) to time (t+m) from the server 30. Then, the acquisition unit 16 generates statistical data S for each time based on the extracted records and flag data. t-n ~St+m The acquiring unit 16 may acquire the performance data and statistical data that have been generated in advance.

[0055] Next, the acquisition unit 16 acquires the area parameter φ of the area to which the target port belongs. k In step S32, the acquisition unit 16 acquires the area parameter φ of the area to which the target port belongs. k The value of the overall parameter θ and the value of the overall parameter θ are read from the storage unit 15. Then, the acquisition unit 16 reads the performance data X t-n ~X t and statistical data S t-n ~S t+m Together with the area parameter φ k and the value of the overall parameter θ to the prediction unit 17.

[0056] Next, the prediction unit 17 predicts the demand at the target port (step S33). In step S33, the prediction unit 17 predicts the demand at the target port by using the prediction model learned by the learning unit 14. To be more specific, as shown in FIG. 11, the prediction unit 17 predicts the demand at the target port by using the performance data X t-n ~X t and statistical data S t-n ~S t+m and the area parameter φ k When the value of the area parameter φ and the value of the overall parameter θ are received from the acquisition unit 16, k and the value of the overall parameter θ are set in the prediction model PM.

[0057] Then, the prediction unit 17 uses the performance data X t-n ~X t and statistical data S t-n ~S t Let DSp be the data set, and let S be the statistical data t-n ~S t+m is input to the prediction model PM as the data set DSst. By this process, the prediction unit 17 inputs the data set DSt (prediction data X' t+1 ~X' t+m ) is obtained from the prediction model PM. Prediction data X't+1 ~X' t+m is the prediction result of the performance data from time (t+1) to time (t+m). Then, the prediction unit 17 outputs the prediction result to the output unit .

[0058] Next, when the output unit 18 receives the prediction result from the prediction unit 17, it outputs the prediction result (step S34). In step S34, the output unit 18 outputs the prediction result to a display, for example, and causes the prediction result to be displayed on the display. The output unit 18 may output (transmit) the prediction result to the server 30, and store the prediction result in the server 30.

[0059] This completes the series of processes in the demand forecasting method. Note that step S32 may be performed before step S31, or may be performed in parallel with step S31.

[0060] In the demand forecasting device 10 (prediction model construction device 50) described above, multiple ports are clustered based on the feature quantities of the multiple ports, and a prediction model is learned based on the performance data of ports that belong to the same cluster as the target port. According to this configuration, ports with similar features can be classified into the same cluster, so even if the shared transportation service is used infrequently in the target port, a sufficient amount of performance data can be obtained by using the performance data of ports that belong to the same cluster as the target port. Therefore, since the prediction model can be trained with a sufficient amount of performance data, it is possible to build a prediction model that can improve the accuracy of demand forecast for vehicles in the shared transportation service. Then, the demand in the target port is predicted using the prediction model, so it is possible to improve the accuracy of demand forecast.

[0061] The learning unit 14 learns the prediction model by meta-learning using the performance data. kTherefore, even if a sufficient number of performance data are not available for an area (port), the area parameter φ is learned by using performance data of ports having similar characteristics. k and the overall parameter θ are learned, the learning accuracy of the prediction model can be improved. According to the prediction model obtained by this learning method, even events that occur only rarely can be effectively predicted.

[0062] The conditions for providing a shared transportation service may differ from area to area. For example, when the surrounding environment of a port is similar, such as a residential area, a commercial facility, and an office district, the service is used in the same way regardless of the area. However, the usage pattern of the service, business hours, and the presence or absence of restrictions on the number of returnable vehicles may differ from area to area. Since one cluster may include ports in different areas, if a prediction model is trained using performance data of ports belonging to the same cluster, a prediction result that contradicts the provision conditions may be obtained. To address this problem, the learning unit 14 trains a prediction model for each area. According to this configuration, a prediction model suitable for each area can be constructed taking into account the characteristics of the above-mentioned area. As a result, it is possible to further improve the accuracy of demand forecasting.

[0063] The learning unit 14 uses an error function that follows a Poisson distribution as the error function of the prediction model. The number of users of shared transportation services such as bike sharing can be considered to follow a Poisson distribution. Therefore, by using an error function that follows a Poisson distribution, the learning accuracy of the prediction model can be improved. In other words, by using an error function that follows the same distribution as the actual data, the prediction model becomes resistant to outliers and the learning accuracy of the prediction model is improved.

[0064] The prediction unit 17 obtains a demand prediction result by inputting the past performance data and statistical data of the target port and the statistical data of the target port at the target time into the prediction model. With this configuration, not only past information but also the statistical data of the target port at the target time is taken into account, so that it is possible to further improve the accuracy of the demand prediction.

[0065] Although the embodiments of the present disclosure have been described above, the present disclosure is not limited to the above embodiments.

[0066] The demand prediction device 10 may be configured by one device that is physically or logically coupled, or may be configured by multiple devices that are physically or logically separated from each other. For example, the demand prediction device 10 may be realized by multiple computers distributed over a network, such as in cloud computing. As described above, the configuration of the demand prediction device 10 may include any configuration that can realize the functions of the demand prediction device 10. Similarly, the configuration of the prediction model construction device 50 may include any configuration that can realize the functions of the prediction model construction device 50.

[0067] The demand prediction device 10 does not need to include the prediction model construction device 50. In this case, the acquisition unit 16 acquires various information for demand prediction from, for example, an external prediction model construction device 50.

[0068] The predictive model construction device 50 does not need to include the storage unit 15. In this case, the values ​​of each parameter of the predictive model are stored in an external device such as the server 30.

[0069] The learning unit 14 may construct a common prediction model for all areas, in which case the learning process can be simplified.

[0070] The learning unit 14 may learn the prediction model using a learning method other than meta-learning. As the error function of the prediction model, an error function other than the error function according to the Poisson distribution may be used.

[0071] The block diagrams used to explain the above embodiments show functional blocks. These functional blocks (components) are realized by any combination of at least one of hardware and software. The method of realizing each functional block is not particularly limited. That is, each functional block may be realized by using one device that is physically or logically combined, or may be realized by using two or more devices that are physically or logically separated and directly or indirectly connected (for example, by wire, wirelessly, etc.). The functional block may be realized by combining the one device or the multiple devices with software.

[0072] Functions include, but are not limited to, judgment, determination, judgement, calculation, computation, processing, derivation, investigation, search, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, consideration, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating, mapping, and assignment. For example, a functional block (component) that performs the function of transmission is called a transmitting unit or a transmitter. In either case, as described above, there is no particular limitation on the method of realization.

[0073] For example, the demand prediction device 10 according to an embodiment of the present disclosure may function as a computer that performs the processing of the present disclosure. Fig. 12 is a diagram illustrating an example of a hardware configuration of the demand prediction device 10 according to an embodiment of the present disclosure. The above-mentioned demand prediction device 10 may be physically configured as a computer device including a processor 1001, a memory 1002, a storage 1003, a communication device 1004, an input device 1005, an output device 1006, a bus 1007, and the like.

[0074] In the following description, the term "apparatus" may be interpreted as a circuit, a device, a unit, etc. The hardware configuration of the demand prediction apparatus 10 may be configured to include one or more of the apparatuses shown in the drawings, or may be configured to exclude some of the apparatuses.

[0075] Each function in the demand forecasting device 10 is realized by loading a specific software (program) onto hardware such as a processor 1001 and a memory 1002, causing the processor 1001 to perform calculations, control communications via a communication device 1004, and control at least one of the reading and writing of data in the memory 1002 and the storage 1003.

[0076] The processor 1001, for example, operates an operating system to control the entire computer. The processor 1001 may be configured with a central processing unit (CPU) including an interface with peripheral devices, a control device, an arithmetic unit, and a register. For example, each function of the demand prediction device 10 described above may be realized by the processor 1001.

[0077] The processor 1001 reads out a program (program code), a software module, data, etc. from at least one of the storage 1003 and the communication device 1004 into the memory 1002, and executes various processes according to the read out programs. As the program, a program that causes a computer to execute at least a part of the operations described in the above-mentioned embodiment is used. For example, each function of the demand prediction device 10 may be realized by a control program stored in the memory 1002 and operated in the processor 1001. Although the above-mentioned various processes have been described as being executed by one processor 1001, they may be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The program may be transmitted from a network via a telecommunication line.

[0078] The memory 1002 is a computer-readable recording medium, and may be configured by, for example, at least one of a Read Only Memory (ROM), an Erasable Programmable ROM (EPROM), an Electrically Erasable Programmable ROM (EEPROM), and a Random Access Memory (RAM). The memory 1002 may be called a register, a cache, a main memory (primary storage device), or the like. The memory 1002 can store a program (program code), a software module, or the like that is executable to implement a prediction model construction method and a demand forecasting method according to an embodiment of the present disclosure.

[0079] Storage 1003 is a computer-readable recording medium, and may be composed of at least one of an optical disk such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disk, a digital versatile disk, a Blu-ray (registered trademark) disk), a smart card, a flash memory (e.g., a card, a stick, a key drive), a floppy (registered trademark) disk, and a magnetic strip. Storage 1003 may be called an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, a server, or other suitable medium including at least one of memory 1002 and storage 1003.

[0080] The communication device 1004 is hardware (transmission / reception device) for performing communication between computers via at least one of a wired network and a wireless network, and is also called, for example, a network device, a network controller, a network card, or a communication module. The communication device 1004 may be configured to include a high-frequency switch, a duplexer, a filter, and a frequency synthesizer to realize, for example, at least one of Frequency Division Duplex (FDD) and Time Division Duplex (TDD). For example, the above-mentioned acquisition unit 11, acquisition unit 16, output unit 18, and the like may be realized by the communication device 1004.

[0081] The input device 1005 is an input device (e.g., a keyboard, a mouse, a microphone, a switch, a button, a sensor, etc.) that receives input from the outside. The output device 1006 is an output device (e.g., a display, a speaker, an LED lamp, etc.) that performs output to the outside. The input device 1005 and the output device 1006 may be integrated into one device (e.g., a touch panel).

[0082] Each device such as the processor 1001 and the memory 1002 is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or may be configured using different buses between each device.

[0083] The demand prediction device 10 may be configured to include hardware such as a microprocessor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a programmable logic device (PLD), or a field programmable gate array (FPGA), and some or all of the functional blocks may be realized by the hardware. For example, the processor 1001 may be implemented using at least one of these pieces of hardware. The prediction model construction device 50 may also be configured with the same hardware as the demand prediction device 10.

[0084] Notification of information is not limited to the aspects / embodiments described in this disclosure and may be performed using other methods.

[0085] In the process steps, sequences, flow charts, etc. of each aspect / embodiment described in this disclosure, the order may be changed unless inconsistent. For example, the methods described in this disclosure present elements of various steps using an example order, and are not limited to the particular order presented.

[0086] The information may be output from a higher layer to a lower layer, or may be output from a lower layer to a higher layer. The information may be input / output via a plurality of network nodes.

[0087] The input and output information may be stored in a specific location (e.g., memory) or may be managed using a management table. The input and output information may be overwritten, updated, or added to. The output information may be deleted. The input information may be transmitted to another device.

[0088] The determination may be based on a value represented by a single bit (0 or 1), a Boolean value (true or false), or a numerical comparison (e.g., comparison with a predetermined value).

[0089] Each aspect / embodiment described in this disclosure may be used alone, in combination, or switched according to execution. Notification of predetermined information (e.g., notification of "X") is not limited to explicit notification, and may be performed implicitly (e.g., by not notifying the predetermined information).

[0090] Although the present disclosure has been described in detail above, it is clear to those skilled in the art that the present disclosure is not limited to the embodiments described in the present disclosure. The present disclosure can be implemented as modified and altered forms without departing from the spirit and scope of the present disclosure as defined by the claims. Therefore, the description of the present disclosure is for illustrative purposes only and does not have any limiting meaning on the present disclosure.

[0091] Software shall be construed broadly to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, threads of execution, procedures, functions, etc., whether referred to as software, firmware, middleware, microcode, hardware description language, or otherwise.

[0092] Software, instructions, information, and the like may be transmitted and received over a transmission medium. For example, if the software is transmitted from a website, server, or other remote source using wired technologies (such as coaxial cable, fiber optic cable, twisted pair, Digital Subscriber Line (DSL)), and / or wireless technologies (such as infrared, microwave), then these wired and / or wireless technologies are included within the definition of transmission media.

[0093] The information, signals, etc. described in this disclosure may be represented using any of a variety of different technologies. For example, the data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltages, currents, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0094] In addition, terms explained in this disclosure and terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings.

[0095] As used in this disclosure, the terms "system" and "network" are used interchangeably.

[0096] The information, parameters, etc. described in this disclosure may be expressed using absolute values, may be expressed using relative values ​​from a predetermined value, or may be expressed using corresponding other information.

[0097] The names used for the above parameters are not limiting in any way, and the formulas etc. using these parameters may differ from the formulas etc. explicitly disclosed in this disclosure.

[0098] The terms "determining" and "determining" as used in this disclosure may encompass a wide variety of actions. "Determining" and "determining" may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, search, inquiry (e.g., searching in a table, database, or other data structure), ascertaining, and the like. "Determining" and "determining" may include receiving (e.g., receiving information), transmitting (e.g., sending information), input, output, accessing (e.g., accessing data in a memory), and the like. "Judgment" and "decision" may include the deeming of resolving, selecting, choosing, establishing, comparing, etc. as having been "judged" or "decided." In other words, "judgment" and "decision" may include the deeming of some action as having been "judged" or "decided." "Judgment" may be interpreted as "assuming," "expecting," or "considering," etc.

[0099] The terms "connected," "coupled," or any variation thereof, refer to any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are "connected" or "coupled" to each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, "connected" may be read as "access." As used in this disclosure, two elements may be considered to be "connected" or "coupled" to each other using at least one of one or more wires, cables, and printed electrical connections, as well as electromagnetic energy having wavelengths in the radio frequency range, microwave range, and optical (both visible and invisible) range, as some non-limiting and non-exhaustive examples.

[0100] As used in this disclosure, the phrase "based on" does not mean "based only on," unless expressly stated otherwise. In other words, the phrase "based on" means both "based only on" and "based at least on."

[0101] Any reference to an element using designations such as "first" and "second" used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient way to distinguish between two or more elements. Thus, a reference to a first and a second element does not imply either that only two elements may be employed or that the first element must precede the second element in some way.

[0102] The "parts" in the configuration of each of the above devices may be replaced with "circuits" or "devices", etc.

[0103] When the terms "include," "including," and variations thereof are used in this disclosure, these terms are intended to be inclusive, similar to the term "comprising." Further, when used in this disclosure, the term "or" is not intended to be an exclusive or.

[0104] In this disclosure, where articles are added by translation, such as "a," "an," and "the" in English, the disclosure may include that the nouns following these articles are in the plural form.

[0105] In the present disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "coupled" may also be interpreted in the same way as "different." [Explanation of symbols]

[0106] 10...demand forecasting device, 11...acquisition unit, 12...generation unit, 13...classification unit, 14...learning unit, 15...memory unit, 16...acquisition unit, 17...prediction unit, 18...output unit, 50...prediction model construction device, PM...prediction model.

Claims

1. A prediction model construction device that constructs a prediction model for predicting demand for vehicle rental and return in a sharing transportation service, A generation unit that generates feature quantities for each of a plurality of ports; A classification unit that clusters a plurality of ports based on the feature amount; a learning unit that learns the prediction model based on performance data on the demand of a port that belongs to the same cluster as a target port that is a port to be predicted; A predictive model construction device comprising:

2. The predictive model construction device according to claim 1 , wherein the learning unit learns the predictive model by meta-learning using the performance data.

3. each of the plurality of ports belongs to one of a plurality of areas; The predictive model construction device according to claim 2 , wherein the learning unit trains the predictive model for each area.

4. 4. The predictive model construction device according to claim 1, wherein the learning unit uses an error function according to a Poisson distribution as an error function of the predictive model.

5. A demand prediction device that predicts demand for vehicle rental and return in a sharing transportation service, A predictive model construction device according to any one of claims 1 to 4, a prediction unit that predicts the demand at the target port by using the prediction model; A demand forecasting device comprising:

6. 6. The demand forecasting device according to claim 5, wherein the prediction unit obtains a forecast result of the demand by inputting past performance data and statistical data for the target port and statistical data for a target time for predicting the demand at the target port into the forecast model.

Images