Information processing apparatus and information processing method

Abstract

To improve the accuracy of demand prediction of a vehicle for each arrangement point by correcting a prediction value on the basis of wide measurement data including data other than a prediction target.SOLUTION: A vehicle dispatch management system 1 includes a data management device 2 that stores a demand measurement value of a vehicle V in one service area, an inter-area data sharing device 7 that stores a demand measurement value of a remaining service area and can transmit the demand measurement value to the data management device 2, a demand prediction device 3 that calculates a demand prediction value obtained by predicting a demand for the vehicle V in a future predetermined time zone in the one service area, and a vehicle dispatch management device 4 that arranges the vehicle V in the one service area. A demand prediction device 3 extracts, from a data management device 2, target time-series data indicating a demand measurement value of an arrangement point of a prediction target in time series and reference time-series data indicating demand measurement values of a plurality of service areas in time series, and corrects a demand prediction value on the basis of similar time-series data which is reference time-series data similar to the target time-series data.SELECTED DRAWING: Figure 1A

Description

[Technical Field]

[0001] The present disclosure relates to a vehicle dispatch management system for vehicles (service vehicles) provided for user mobility services. [Background technology]

[0002] Mobility services are positioned as services that provide smooth transportation of users and cargo by vehicle. In order to increase vehicle utilization rates and improve user (passenger) convenience, mobility services require forecasting vehicle demand for each vehicle deployment location (deployment location) and creating efficient deployment plans.

[0003] For example, Patent Document 1 below discloses a technique for improving prediction accuracy by correcting a predicted value with a correction value calculated based on actual measurement data of various past periods of the prediction target. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent No. 6831280 Summary of the Invention [Problem to be solved by the invention]

[0005] However, the technology disclosed in Patent Document 1 calculates a correction value based on similar data that is similar to measurement data from the most recent past period among past measurement data of only the prediction target. Therefore, if similar data does not exist in the past measurement data of the prediction target, the correction effect is not sufficient and improvement in prediction accuracy cannot be achieved.

[0006] Therefore, an object of the present disclosure is to provide a technology that corrects predicted values ​​based on a wide range of measurement data, including data other than the prediction target, and improves the accuracy of vehicle demand prediction for each deployment location. [Means for solving the problem]

[0007] a data sharing device that stores the demand values ​​measured at each of the deployment points in the remaining districts and is capable of transmitting the demand values ​​to the data management device; a demand prediction device that calculates a demand forecast value that predicts demand for the vehicle for a predetermined time period in the future at each of the deployment points in the one district based on the demand values; and the vehicle dispatch management device that deploys the vehicles to each of the deployment points in the one district based on the demand forecast value, wherein the demand prediction device extracts, from the data management device via a predetermined network, first time series data that indicates, in time series, the demand values ​​measured at the deployment point to be predicted during a most recent past period, and second time series data that is data other than the first time series data and indicates, in time series, the demand values ​​measured at each of the deployment points in the multiple districts during a predetermined past period, and corrects the demand forecast value based on similar time series data that is the second time series data that is similar to the first time series data. [Effects of the Invention]

[0008] According to the present disclosure, it becomes possible to correct predicted values ​​based on a wide range of measurement data, including data other than the target of prediction, and improve the accuracy of vehicle demand prediction for each deployment location. [Brief explanation of the drawings]

[0009] [Figure 1A] FIG. 1 is a diagram illustrating an example of a vehicle dispatch management system according to an embodiment of the present disclosure. [Figure 1B] FIG. 1 is a schematic diagram illustrating locations of vehicles used in a vehicle dispatch service. [Figure 2] 2 is a block diagram illustrating an example of the hardware and functional configuration of a data management device in the vehicle dispatch management system illustrated in FIG. 1. FIG. [Figure 3A]FIG. 10 is a diagram illustrating an example of point divisions in sample data. [Figure 3B] FIG. 10 is a diagram illustrating an example of period divisions in sample data. [Figure 3C] FIG. 10 is a diagram illustrating an example of demand factors in sample data. [Figure 4] 2 is a block diagram illustrating an example of the hardware and functional configuration of a demand prediction device in the vehicle dispatch management system illustrated in FIG. 1. FIG. [Figure 5A] FIG. 1 is a conceptual diagram illustrating an example of actual measurement data used for demand forecasting. [Figure 5B] FIG. 10 is a conceptual diagram illustrating another example of actual measurement data used for demand forecasting. [Figure 5C] FIG. 10 is a conceptual diagram illustrating grouping in demand forecasting. [Figure 6] 2 is a block diagram illustrating an example of hardware and functional configurations of a vehicle dispatch management device in the vehicle dispatch management system illustrated in FIG. 1. FIG. [Figure 7] 1 is a flowchart illustrating an example of a vehicle dispatch management method by a vehicle dispatch management system according to an embodiment of the present disclosure. [Figure 8] 10 is a flowchart illustrating an example of a correction process of a demand forecast value by a vehicle dispatch management system according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0010] Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that the drawings are schematic and may differ from the actual product. Furthermore, the embodiments of the present invention shown below are examples of devices and methods for embodying the technical concept of the present invention, and the technical concept of the present invention does not limit the structure, arrangement, etc. of component parts to those described below. The technical concept of the present invention can be modified in various ways within the technical scope defined by the claims.

[0011] (System Configuration) Fig. 1A is a diagram illustrating an example of a vehicle dispatch management system according to an embodiment of the present disclosure. Fig. 1B is a schematic diagram illustrating a service district, which is an area where a vehicle dispatch service is provided. The vehicle dispatch management system 1 is a system that provides a vehicle dispatch service that dispatches a vehicle V used for user transportation. The vehicle dispatch management system 1 includes a data management device 2, a demand prediction device 3, a vehicle dispatch management device 4, and a vehicle V, which are connected to each other so that they can communicate with each other via a communication network 9. The vehicle dispatch management system 1 also includes a factor data collection device 5, a demand measurement device 6, and an inter-district data sharing device (an example of a data sharing device) 7, which are connected to the data management device 2 so that they can communicate with each other via the communication network 9.

[0012] Each of the service districts A to C is a regional range including one or more deployment points where a vehicle V is deployed for use by a user. For example, service district A is a regional range including deployment points a1 to a8. Similarly, one or more deployment points are set for service districts B and C. For example, each of the service districts A to C can be recognized based on a point of interest (POI) that is frequently used or highly popular with users, such as an airport, a train station, or a large-scale facility that attracts customers. In this disclosure, an example is described in which the vehicle dispatch management system 1 is applied to a mobility service for moving passengers and transporting luggage by vehicle V, but the service application is not limited to this. Furthermore, the deployment point may be a waiting location where vehicle V waits before being dispatched to a predetermined boarding location where a user boards, or after a user disembarks at a predetermined disembarking location, or may be a boarding and disembarking location where multiple vehicles V can stop. The vehicle dispatch service provided by the vehicle dispatch management system 1 covers multiple service areas A to C, but in this example, the vehicle dispatch service in service area A will be described as an example. A data management device 2 may be provided for each service area. In other words, the vehicle dispatch management system 1 may be provided with multiple data management devices 2. Hereinafter, data management devices 2 corresponding to areas other than service area A (service areas B and C) will be referred to as "data management devices for other areas."

[0013] The data management device 2 is a computing device that manages sample data used to predict demand for vehicles V in each service area. In this example, the data management device 2 stores sample data that is used to calculate demand forecast values ​​that predict demand for vehicles V during a predetermined future time period at each deployment point (deployment points a1 to a8) in one service area, service area A. The sample data includes at least measured values ​​(demand measurement values) of past demand for vehicles V in one service area (service area A) and one or more types of demand factor data that affect the demand. In other words, in this embodiment, the data management device 2 stores demand measurement values ​​(an example of demand values) that indicate demand for vehicles V measured at each deployment point where vehicles V are deployed in one service area. The demand for vehicle V may be the number of vehicle dispatch requests (number of requests) from users who wish to ride in the vicinity of each deployment point, or the results of vehicle dispatch to users at each deployment point (number of vehicles dispatched).

[0014] The demand factor data is data on the characteristics of the location (location factor data) and data on the characteristics of each period segment at the location (period segment factor data).Here, the period segment indicates a segmentation of date and time according to predetermined temporal conditions (season, day of the week, time of day, etc.). The sample data is transmitted from other devices to the data management device 2 via the communication network 9. For example, among the sample data, demand factor data is transmitted from the factor data collection device 5, and the measured values ​​of past demand of vehicle V in service area A are transmitted from the demand measurement device 6.

[0015] The data management device 2 stores sample data (other area data) from other service areas (e.g., service areas B and C) in addition to sample data from one service area (service area A) as data used to calculate demand forecast values. This allows the vehicle dispatch management system 1 to make demand forecasts by referring to a wide range of measurement values, including sample data from areas other than the one service area being predicted. The other area data is transmitted from the inter-area data sharing device 7 to the data management device 2 via the communication network 9. Furthermore, in response to a data acquisition request from the demand forecasting device 3, the data management device 2 searches for data and transmits the search results to other devices.

[0016] The demand forecasting device 3 uses at least the sample data stored in the data management device 2 to perform a demand forecast for vehicles V for a predetermined future time period at each deployment point (deployment points a1 to a8) in service area A, which is one service area. Specifically, the demand forecasting device 3 calculates a demand forecast value, which is a value predicting the demand for vehicles V for the predetermined future time period, based on the demand forecast value in the sample data. When performing demand forecasting, the demand forecasting device 3 transmits a request to acquire sample data to the data management device 2 via the communication network 9 to extract sample data (e.g., demand measurement values) to be used for demand forecasting. The demand forecasting device 3 can also correct the calculated demand forecast value based on the sample data and data from other areas. This can improve the accuracy of demand forecasting. The calculation and correction of the demand forecast value will be described in detail later. The demand prediction device 3 also transmits the demand prediction results (demand prediction values) to the vehicle allocation management device 4 via the communication network 9.

[0017] The vehicle allocation management device 4 creates an allocation plan for vehicles V at each allocation point (allocation points a1 to a8) within one service area (service area A in this example) based on the demand forecast value transmitted from the demand forecasting device 3. This allows vehicles V to be efficiently allocated at each allocation point in the service area, improving the utilization rate. The allocation plan may include, for example, the number of vehicles V to be allocated at each allocation point during a predetermined period. The vehicle allocation management device 4 instructs vehicles V via the communication network 9 to move to each allocation point based on the allocation plan.

[0018] The factor data collection device 5 periodically measures the above-mentioned demand factor data (location factor data, period division factor data) at predetermined time intervals and transmits the data to the data management device 2 via the communication network 9. The demand measurement device 6 periodically measures the demand for vehicles V in one service area (service area A) at predetermined unit times (for example, 15 minutes) and transmits demand measurement data indicating the measured demand values ​​to the data management device 2 via the communication network 9. Specifically, the demand measurement device 6 measures the number of requests per unit time at each deployment point in one service area based on the vehicle dispatch request information transmitted from the vehicle dispatch management device 4. As will be described in more detail below, the dispatch request information includes at least the boarding location where the user boards the vehicle V and the date and time (reception date and time) when the dispatch management device 4 receives the dispatch request from the user. The demand measurement device 6 identifies the allocation point where the vehicle V dispatched to the user from the boarding location was waiting, and identifies the unit time during which the dispatch request was made from the reception date and time. This allows the demand measurement device 6 to measure the number of requests (number of dispatch requests) per unit time for each allocation point as a demand measurement value. In this embodiment, each allocation point is associated with a predetermined coverage area, and the allocation point of the vehicle V dispatched to the boarding location can be identified based on the coverage area in which the boarding location is located.

[0019] The inter-area data sharing device 7 is a computing device that manages sample data for areas other than one service area (service area A) stored in the data management device 2 (the remaining areas of the multiple service areas). For example, the inter-area data sharing device 7 periodically collects sample data for each area (service areas B and C) at predetermined time intervals from the data management devices in the other areas that manage the sample data for each area. The inter-area data sharing device 7 transmits the sample data for service area A to the data management device 2 that manages the sample data for each area as other area data. In response to a data acquisition request from the data management device 2, the inter-area data sharing device 7 searches for data and transmits the search results.

[0020] (Configuration of data management device) Next, the data management device 2 in the vehicle dispatch management system 1 will be described with reference to Figure 2. The data management device 2 receives demand factor data, demand measurement data, and other district data from the factor data collection device 5 and the demand measurement device 6, respectively, and stores them as sample data. It also acquires other district data from the inter-district data sharing device 7. Then, in response to a data acquisition request, it transmits this data to the demand forecasting device 3. The data management device 2 is composed of an input device 21, an output device 22, a communication device 23, and a controller 24. The data management device 2 may be, for example, an information processing device such as a personal computer or a server computer.

[0021] The input device 21 is a keyboard or a mouse, and the output device 22 is a display or a printer. The communication device 33 is equipped with a NIC (Network Interface Card) for connecting to a wireless LAN or a wired LAN. The data management device 2 transmits and receives data to and from each device via the communication device 33. The controller 24 is an electronic control unit (ECU) that controls the operation of the data management device 2. The controller 24 includes a processor 24a and a storage device 24b.

[0022] The processor 24a may be, for example, a central processing unit (CPU) or a micro-processing unit (MPU). The storage device 24b may include a memory such as a read-only memory (ROM) or a random access memory (RAM) used as a main storage device, or a non-transitory tangible storage medium such as a memory register or a cache memory. The storage device 24b stores a database serving as a sample data storage module 241. The sample data storage module 241 holds sample data 241a. The sample data 241a is composed of demand factor data (point factor data, period division factor data) transmitted from the factor data collection device 5 and demand measurement data transmitted from the demand measurement device 6. The demand measurement data includes demand measurement values ​​at each deployment point in the service area A and the measurement time of the demand measurement values. The measurement time needs to include at least the start time or end time of the measurement, and may include both. The demand measurement values ​​are measured at regular unit times (for example, 15 minutes) as described above. Therefore, in the sample data storage module 241, demand measurement data including demand measurement values ​​may be held for each unit time.

[0023] Here, the sample data 241a will be described in detail with reference to FIGS. 3A to 3C. As described above, the sample data 241a includes the demand factor data transmitted from the factor data collecting device 5 and the demand measurement data transmitted from the demand measurement device 6. For example, the sample data 241a may be data in which the demand measurement data (demand measurement value, measurement time) is linked with factor classifications (location classifications and period classifications) for linking with the demand factors. The location classification identifies each deployment location within the service area A where the demand measurement data was measured, and is used to associate the location factor data with the demand measurement data. As shown in Fig. 3A, each deployment location (deployment locations a1 to a8) is associated with a location ID (A001 to A008), and by assigning a location ID to the transmitted demand measurement data, the deployment location where the demand measurement data was measured can be identified.

[0024] As mentioned above, period segments are date and time segments based on seasonal conditions (season, day of the week, time period, etc.) and are used to associate period segmentation factor data with the unit time at which demand measurement data was measured. As shown in Figure 3B, period segments are broadly divided into three segmentation types: season, day of the week, and time period. Among the segmentation types, "season" is associated with four segmentation items: "spring, summer, autumn, and winter," each associated with a segmentation ID: "S1 (spring)," "S2 (summer)," "S3 (autumn)," and "S4 (winter)." Furthermore, among the segmentation types, "day of the week" is associated with two segmentation items: "weekday" and "holiday," each associated with a segmentation ID: "Wd (weekday)" and "Hd (holiday)." Furthermore, among the segmentation types, "time period" is divided into eight segmentation items: "early morning, early morning, morning, late afternoon, noon, afternoon, evening, and night." Each time period is divided into three-hour intervals, for example, from midnight to midnight 24 hours later. Period segments are identified by a combination of segment IDs that indicate each item in each segment type. For example, a period segment called "Spring, weekdays, early morning" is identified by the combination of segment IDs "S1," "Wd," and "U2" (S1 / Wd / U2). This combination of segment IDs is called a "period segment ID."

[0025] As shown in Figure 3C, among the demand factor data, the location factor data may be, for example, data indicating population, POI concentration indicating the density of POIs near the location point, and the convenience of public transportation (transport convenience). These are data collected and compiled at a predetermined interval for each location point in one service area (service area A). Note that the location factor data is not limited to these, and may include various factor data indicating location characteristics that affect demand forecasting, such as walkability indicating the difficulty of walking due to the number of slopes, bridges, wide roads, etc. The location factor data is stored in association with a location ID in the sample data storage module 241. In addition, the demand measurement data can be associated with the location factor data by the location ID. This allows the demand prediction device 3 to identify the demand measurement data associated with specific location factor data when predicting the demand for the vehicle V.

[0026] Furthermore, the period classification factor data among the demand factor data may be, for example, climate data such as temperature and weather, data indicating the frequency of public transportation services, and data indicating the degree of congestion of public transportation services. Note that the period classification factor data is not limited to these, and may include various factor data that change depending on seasonal conditions and affect demand forecasting. The period division factor data is data collected and compiled for each period division at each deployment location. Therefore, in the sample data storage module 241, the period division factor data is stored in association with the location ID and the period division ID. Furthermore, the demand measurement data can be associated with the period division factor data by the location ID and the period division factor. When the demand forecasting device 3 forecasts the demand for the vehicle V, it can identify the demand measurement data associated with the period division factor data at a specific deployment location. The sample data storage module 241 accumulates sample data 241a. That is, the sample data 241a is not overwritten and stored, and includes past demand factor data and demand measurement data.

[0027] For example, sample data relating to demand measurement data measured in a 15-minute unit time period starting at "7:00 AM on Friday, March 1, 2024" at deployment point a1 in service area A will have a point ID of "A001," a demand measurement value (e.g., number of requests) of "5," a measurement start time of "2024.03.01.07.00," and a period segment ID of "S1 / Wd / U3," and is associated with point factor data using the point ID, and with period segment factor data using the point ID and period segment ID. Note that the configuration of sample data 241a shown in Figures 3A to 3C is an example, and the configuration of sample data in the present disclosure is not limited to this, and it is sufficient if the configuration allows demand measurement data and demand factor data (location factor data, period division factor data) to be associated with each other.

[0028] 2, the storage device 24b stores programs for implementing control modules such as the specimen data management module 242 and the other area data acquisition module 243. The processor 24a executes these programs to implement the functions of the data management device 2.

[0029] The sample data management module 242 adds and extracts sample data 241a to the sample data storage module 241. Specifically, the sample data management module 242 adds new sample data 241a to the sample data storage module 241 based on the received demand factor data and demand measurement data, and extracts specific data from the sample data 241a based on a sample data acquisition request and transmits the extracted data to the demand forecasting device 3. The other district data acquisition module 243 transmits a data acquisition request to the inter-district data sharing device 7, acquires other district data, and transmits it to the demand forecasting device 3. The other district data acquisition module 243 may acquire other district data from the inter-district data sharing device 7 based on the other district data acquisition request from the demand forecasting device 3, or may acquire other district data at predetermined time intervals.

[0030] (Configuration of demand forecasting device) Next, the demand prediction device 3 in the vehicle dispatch management system 1 will be described using Figure 4. The demand prediction device 3 acquires specific demand measurement data from the sample data 241a from the data management device 2, and calculates a demand prediction value for vehicle V for each deployment point in one service district (service district A) based on the demand measurement data. The calculated demand prediction value is transmitted to the vehicle dispatch management device 4 and the data management device 2. The demand prediction device 3 can also acquire other district data from the data management device 2, and correct the demand prediction value based on the other district data and the demand measurement data. The demand forecasting device 3 is composed of an input device 31, an output device 32, a communication device 33, and a controller 34. The demand forecasting device 3 may be an information processing device such as a personal computer or a server computer. The configurations of the input device 31, output device 32, and communication device 33 are the same as those of the input device 21, output device 22, and communication device 23 in the data management device 2, so detailed explanations will be omitted.

[0031] The controller 34 is an electronic control unit (ECU) that controls the operation of the demand prediction device 3. The controller 34 includes a processor 34a and a storage device 34b. The processor 34a has the same configuration as the processor 24a in the data management device 2, so a detailed description will be omitted. The storage device 34b has basically the same configuration as the storage device 24b in the data management device 2, but the programs stored therein are different. The storage device 34b stores programs for implementing control modules such as a predicted value calculation module 341, an error estimation module 342, and a prediction correction module 343. The processor 34a executes these programs to implement the functions of the demand prediction device 3.

[0032] The predicted value calculation module 341 calculates a demand prediction value for the vehicle V at each deployment point in the service area A based on the demand measurement data and the like stored in the data management device 2. For example, the predicted value calculation module 341 may calculate a demand prediction value for the next unit time (7:15 to 7:30) at the start timing (e.g., 7:00) of a new unit time when the demand measurement value is measured.

[0033] The forecast value calculation module 341 calculates a demand forecast value based on, for example, the sample data 241a stored in the storage device 24b of the data management device 2. Specifically, the forecast value calculation module 341 may calculate the latest demand forecast value based on past demand measurement data and demand factor data (location factor data, period segment factor data) of the placement location of the prediction target (e.g., placement location a1) among the sample data 241a. For example, the forecast value calculation module 341 may extract, from the data management device 2, the sample data 241a corresponding to the placement location of the prediction target and the period segment to which the unit time of the prediction target belongs, and calculate the demand forecast value based on the extracted sample data 241a. The forecast value calculation module 341 may include a demand forecast model that has been subjected to machine learning according to a predetermined machine learning algorithm, with the acquired sample data 241a (demand measurement data, demand factor data) as explanatory variables and the demand forecast value (e.g., the number of requests per unit time) as a target variable.

[0034] The predicted value calculation module 341 transmits predicted value data including the calculated demand forecast value to the vehicle dispatch management device 4 via the communication device 33. The predicted value data includes the demand forecast value, the deployment point of the target deployment, and the start time of the target unit time. This allows the vehicle dispatch management device 4 to create a deployment plan for the vehicle V at each deployment point for the next unit time based on the latest demand forecast value. Furthermore, the predicted value calculation module 341 transmits the predicted value data to the data management device 2. The sample data management module 242 of the data management device 2 stores the received predicted value data in the sample data storage module 241 for each unit time. This allows the data management device 2 to store the demand forecast value and the demand measurement data in association with each other. In other words, the storage device 24b of the data management device 2 stores error time-series data indicating the error between the demand measurement value actually measured and the demand forecast value at each deployment point in one district (service district A). Note that the predicted value data may be stored in a database separate from the sample data storage module 241.

[0035] As described above, in this embodiment, the demand forecast value is calculated based on the sample data 241a, but various factors can make it difficult to forecast demand based on demand factor data or past demand measurement values ​​at the target deployment location, resulting in a deviation between the calculated demand forecast value and the demand measurement value. In this case, vehicle allocation efficiency will be insufficient, reducing user convenience.

[0036] Therefore, in this embodiment, the prediction value calculation module 341 may use error time series data (an example of first error data) that shows prediction error data for the most recent past period at the placement point of the prediction target in a time series format to determine whether demand prediction is difficult (whether deviation from the demand measurement value is likely to occur) regarding the latest demand prediction value. That is, the forecast value calculation module 341 may extract the error time series data from the data management device 2, and determine whether or not the latest calculated demand forecast value needs to be corrected based on the error time series data.

[0037] Here, the most recent past period refers to a period including the unit time "t-1" (the immediately preceding unit time) when the current unit time is "t0." If one unit time (for example, 15 minutes) is considered to be "1 unit," the most recent past period may be a past period of n units including unit time "t-1," that is, a continuous period from unit time "t-1" to unit time "tn" (15 minutes x n). In other words, the error time series data is data that indicates, in time series, the prediction error for multiple unit times (n units) including the immediately preceding unit time.

[0038] For example, if the error time-series data satisfies both of the following prediction conditions (i) and (ii), the predicted value calculation module 341 determines that correction of the demand forecast value is unnecessary and does not correct the calculated latest demand forecast value. In other words, it determines that there is little risk of a deviation between the latest demand forecast value and the future demand measurement value (future forecast error) and that it is appropriate to apply the demand forecast value to the creation of a deployment plan in the vehicle dispatch management device 4, and transmits the latest demand forecast value to the vehicle dispatch management device 4. (i) Trends in forecast errors In the error time series data, the forecast error is on a decreasing trend (the value of the forecast error decreases as time progresses). (ii) Latest forecast error situation The difference between the demand forecast value and the demand measurement value in the immediately preceding unit time (latest forecast error) is equal to or smaller than a predetermined threshold value.

[0039] Furthermore, the forecast value calculation module 341 may determine a correction to the latest demand forecast value if the error time-series data does not satisfy either of the forecast conditions exemplified in (i) or (ii) below. If the error time-series data does not satisfy either of the forecast conditions (i) or (ii), it is estimated that a future forecast error is likely to occur. Therefore, the forecast value calculation module 341 determines that the demand forecast value is not appropriate for application to the creation of a deployment plan in the vehicle dispatch management device 4. In other words, it determines that a correction of the demand forecast value is necessary, outputs the calculated demand forecast value to the error estimation module 342, and requests the correction of the demand forecast value.

[0040] The error estimation module 342 estimates future forecast errors in the latest demand forecast values ​​based on the error time-series data and corrects the latest demand forecast values ​​based on the future forecast errors. Specifically, when it is determined that the demand forecast values ​​should be corrected (when the demand forecast values ​​are input from the forecast value calculation module 341), the error estimation module 342 generates error-converted data by logarithmically transforming the forecast errors (values ​​of the forecast errors for n units) in the error time-series data. The error estimation module 342 determines whether the time progression of at least one of the error time-series data and the error-converted data is linear. For example, the error estimation module 342 may determine that the time progression of the data is linear when the amount of change in the value of the forecast error over time is constant or increases stepwise in the error time-series data or the error-converted data.

[0041] If the error estimation module 342 determines that the time progression of the data is linear, it estimates future prediction errors in the latest demand forecast values ​​using a model that mathematically represents the time progression of the error time-series data, and corrects the latest demand forecast values ​​based on the estimated prediction errors. Because linear data is easy to predict, correcting the demand forecast values ​​with the estimated value of the prediction errors can improve the accuracy of the demand forecast for vehicle V. For example, the error estimation module 342 adds the estimated value to the latest demand forecast value to make the correction. For example, if the estimated value to be added is a positive number, the latest demand forecast value will be a value that is more than the calculated latest demand forecast value by the estimated value, and if it is a negative number, the latest demand forecast value will be a value that is less by the estimated value. The error estimation module 342 transmits the corrected demand forecast value to the vehicle allocation management device 4. This makes it possible to create an efficient allocation plan for the vehicles V based on the corrected demand forecast value.

[0042] On the other hand, if the error estimation module 342 determines that the time progression of the above data is not linear, it determines that it is difficult to correct the demand forecast value due to the forecast error and that it is necessary to correct the demand forecast value using a wider range of data, and outputs the latest demand forecast value calculated by the forecast value calculation module 341 to the forecast correction module 343 and requests that the demand forecast value be corrected.

[0043] The forecast correction module 343 corrects the demand forecast value based on past demand measurement values ​​(demand measurement data) at each deployment point in a plurality of districts (service districts A to C in this example) where the vehicle dispatch service is provided. The forecast correction module 343 corrects the demand forecast value based on target time series data (an example of first time series data) that indicates, in time series, the demand measurement values ​​for the most recent period in the past at the deployment point to be forecasted (e.g., deployment point a1), and reference time series data that indicates, in time series, the demand measurement values ​​measured for a predetermined period in the past at each deployment point in the multiple service districts A to C. Here, the reference time series data is multiple time series data other than the target time series data.

[0044] FIG. 5A is a conceptual diagram illustrating an example of target time series data, and FIG. 5B is a conceptual diagram illustrating an example of reference time series data. 5A, the most recent past period in which the demand measurement values ​​in the target time series data were measured may be a past period of n units (a continuous period from unit time "t-1" to unit time "tn") including the unit time "t-1" (the immediately preceding unit time) just before the present (t0), similar to the error time series data. In other words, the most recent past period in the target time series data is the same period as the most recent past period in the error time series data, and the error time series data may be data that indicates, in time series, the prediction error in the most recent past period in which the target time series data was measured. In this case, the time period to be predicted is, for example, a future unit time "t", for example, the unit time next to the current unit time.

[0045] As shown in FIG. 5B, the past reference period in which the demand measurement values ​​in the reference time series data were measured may be a reference past period (a continuous period from unit time "T" to unit time "T-(n-1)"), which is a past period of n units including a predetermined past unit time "T" that serves as a reference standard. In other words, the reference time series data may be data that shows the demand measurement values ​​measured within a predetermined past period in a time series for each unit time. In this example, the most recent past period and the reference past period may be periods of the same length (number of unit times (number of units)). There are multiple reference time series data for one deployment point. In other words, the number of reference time series data (x) for one deployment point is the number of n units (x = m / n) obtained by dividing all units (m units) accumulated up to the latest unit in the previous unit time into n units for one deployment point. For example, if 1000 units (unit time) of demand measurement values ​​are accumulated for one deployment point (m = 1000), and if 4 units of demand measurement values ​​are used as one reference time series data (n = 4), the number of reference time series data will be 250 (= 1000 / 4).

[0046] The forecast correction module 343 extracts the target time series data and multiple reference time series data from the data management device 2 via a predetermined network, and corrects the latest demand forecast value based on similar time series data, which is reference time series data similar to the target time series data. Of the reference time series data, data on other service areas (service areas B and C in this example) is extracted from the inter-area data sharing device 7 via the data management device 2. As a result, the vehicle dispatch management system 1 according to this embodiment collects a wide range of demand measurement data and corrects the demand forecast value based on similar time series data searched from many reference time series data including those from locations other than the deployment location to be predicted, thereby improving the accuracy of the demand forecast for vehicle V at each deployment location.

[0047] The data extracted by the prediction correction module 343 from the data management device 2 as the target time series data and the reference time series data is time series data of a length indicated by a predetermined reference value. As described above, the reference value may be n units of unit time (= unit time × n). If the time series in the target time series data or the reference time series data spans multiple days, the forecast correction module 343 may divide the time series data by date and perform a series of processes to correct the demand forecast value using the time series data with the longest length.

[0048] For example, if the unit time is 15 minutes and the target time series data consists of four units around midnight (11:45 PM to 12:45 AM), the forecast correction module 343 divides the target time series data into a unit before midnight (one unit from 11:45 PM to 12:00 AM) and a unit after midnight (three units from 12:00 AM to 12:45 AM). Since the length of the unit after midnight (three units) is the longest, the forecast correction module 343 uses the three units of time series data after midnight as the target time series data for correcting the demand forecast value. The same applies to the reference time series data. This prevents data with different time periods from being mixed in the time series data, improving the accuracy of the following correction process.

[0049] The prediction correction module 343 normalizes the extracted target time series data and reference time series data, and then sequentially determines the similarity between the target time series data and the reference time series data. Here, normalization may be, for example, a min-max method that scales data from a minimum value of "0" to a maximum value of "1," or any other known method. The prediction / correction module 343 may calculate the similarity between the target time series data and the reference time series data, and determine that the target time series data and the reference time series data are similar if the similarity is above a predetermined threshold. The calculation of the similarity may be performed using a known method (e.g., distance-based similarity determination). The prediction / correction module 343 may also include a similarity model that has been subjected to machine learning according to a predetermined machine learning algorithm.

[0050] The forecast correction module 343 determines one piece of reference time series data whose similarity to the target time series data is above a predetermined threshold as similar time series data, and corrects the latest demand forecast value based on the similar time series data. More specifically, the forecast correction module 343 extracts, as a similar measurement value, from the data management device 2 (or the inter-district data sharing device 7), a demand measurement value measured at the next unit time (for example, unit time "T+1") following the predetermined period at which the similar time series data was measured at the deployment point where the similar time series data was measured. The forecast correction module 343 calculates a scale ratio between the target time series data and the similar time series data (the scale ratio between the data before normalization), and sets a value obtained by correcting the scale of the similar measurement value based on the scale ratio as a demand forecast value. In other words, the forecast correction module 343 sets the scale-corrected similar measurement value as a demand forecast value instead of the demand forecast value calculated by the forecast value calculation module 341 as a correction of the demand forecast value, and transmits the scale-corrected similar measurement value to the vehicle dispatch management device 4.

[0051] Similar measurement values ​​are inferred to be similar to the time period (unit time "t") to be predicted, which is a future time period of the target time series data, because the time series data of demand measurement values ​​up to the previous time period (unit time "T") (similar time series data) is similar to the time series data of demand measurement values ​​for the most recent past period at the deployment point to be predicted (target time series data). For this reason, by using similar measurement values ​​as demand forecast values, it is possible to more reliably improve forecast accuracy.

[0052] If there is no reference time series data whose similarity to the target time series data is above a predetermined threshold, the prediction / correction module 343 may treat the reference time series data with the highest similarity among the multiple reference time series data as the similar time series data. In this case, the similar time series data may be, for example, the data with the highest similarity among the reference time series data. The prediction / correction module 343 may store the similarity in a predetermined storage area of ​​the storage device 34b each time it calculates it. For example, the similarity may be stored together with data that can identify the reference time series data. The prediction / correction module 343 may assign an identification number to each piece of data when extracting the reference time series data, or may identify the reference time series data by the location and period (such as a timestamp of the unit time) at which the reference time series data was measured.

[0053] The reference time series data used to correct demand forecast values ​​includes data from other service areas, so the data set (multiple reference time series data) to search for similar time series data is large. For this reason, it is important to improve the efficiency of searching for similar time series data by calculating the similarity between the target time series data and the reference time series data. The search for similar time series data will be described in detail below.

[0054] For example, the prediction correction module 343 may classify the deployment points of a plurality of service areas (service areas A to C) into groups with similar characteristics, thereby making the search for similar time-series data more efficient. The prediction correction module 343 may acquire location factor data for each location based on, for example, location classification (see FIG. 3A), and may group and classify each location into multiple groups based on the similarity of the location factor data (see FIG. 3C). The grouping may be performed using a known method such as cluster analysis (e.g., hierarchical clustering). The prediction correction module 343 may include a grouping model that performs clustering processing according to a predetermined machine learning algorithm.

[0055] The forecast correction module 343 may also derive, for example, a time transition pattern (daily average demand transition pattern) of the average daily demand measurement value at each deployment point in multiple service areas (service areas A to C), and classify the deployment points into multiple groups based on the similarity of the daily average demand transition patterns. The daily average demand transition pattern may indicate, for example, a change in the transition of the average amount of change in the demand measurement value per unit time in a day. The forecast correction module 343 may include a transition determination model that executes a process of determining the similarity of daily average demand transition patterns according to a predetermined machine learning algorithm. Location points with similar location factor data or daily average demand transition patterns are likely to measure similar demand (demand measurement data). Therefore, by deriving location points with similar characteristics through grouping, it is possible to improve the efficiency of searching for similar time series data.

[0056] Here, the plurality of groups into which each deployment point is classified by grouping based on the above-mentioned point characteristics (point factor data or daily average demand transition pattern) are referred to as "point groups." The prediction correction module 343 may select a group among multiple location groups to which the placement location to be predicted belongs as a target group, and may preferentially determine whether the reference time series data measured at each placement location belonging to the target group is similar to the target time series data (whether the similarity is equal to or greater than a predetermined threshold). The placement points that belong to the same location group (target group) as the placement point of the prediction target have a higher similarity in characteristics (location factor data) with the placement point of the prediction target than the placement points that belong to other groups. Therefore, the reference time series data measured at the placement points in the target group is more likely to be similar to the target time series data than the reference time series data measured at the placement points that belong to other groups. Therefore, by giving priority to determining the similarity of the reference time series data measured at the placement points in the target group, it is possible to reliably improve the efficiency of the search for similar time series data.

[0057] Furthermore, if there is no similar time series data among the reference time series data measured at each location point in the target group, the prediction correction module 343 may determine whether the reference time series data measured at each location point in other location groups is similar to the target time series data. Specifically, the prediction correction module 343 may determine the similarity of the location factor data between multiple location groups (location group similarity), and prioritize whether the reference time series data measured at each location point belonging to other location groups with relatively high location group similarity is similar to the target time series data. This allows for efficient searching of similar time series data by prioritizing location groups that are likely to be similar to the target time series data, even if similar reference time series data (similar time series data) does not exist within the target group.

[0058] The prediction correction module 343 may also group and classify each placement point into multiple groups based on the similarity of the location factor data and the period segment factor data. Specifically, the prediction correction module 343 may acquire location factor data and period segment factor data (see FIG. 3C) for each placement point based on the location segment and period segment (see FIG. 3B), and group each placement point based on the similarity of the location factor data and the period segment factor data. At this time, each placement point that has been grouped (classified into multiple location groups) based on the location factor data may be regrouped based on the period segment factor data. Here, the multiple groups into which each placement point is classified by the regrouping are referred to as "location-period groups."

[0059] The forecast correction module 343 may derive, for example, a time transition pattern (category average demand transition pattern) of the average demand measurement value for each period segment at each deployment point in multiple service districts (service districts A to C), and may group and classify each period segment at each deployment point into multiple groups based on the similarity of the category average demand transition patterns. The category average demand transition pattern may indicate, for example, a change in the transition of the average amount of change in the demand measurement value per unit time in each period segment. Between deployment points (target groups in a location / period group) with similar period division factor data or division average demand transition patterns, there is a high possibility that the demand measurement values ​​(demand measurement data) will be even more similar than between deployment points (target groups in a location group) with similar only location characteristics (location factor data or daily average demand transition patterns).For this reason, by deriving deployment points with similar characteristics through regrouping by location / period group, it is possible to further improve the efficiency of searching for similar time series data.

[0060] The prediction correction module 343 may treat the location / period group to which the placement location and time period (unit time) of the prediction target belongs as the target group, and may preferentially determine whether the reference time series data measured at each placement location and period segment belonging to the target group is similar to the target time series data. The location points and period segments that make up the same location point and unit time group (target group) as the location point and unit time of the prediction target have a higher similarity in terms of the characteristics (location point factor data and period segment factor data) of the location point of the prediction target than the location points and period segments that belong to other groups. Therefore, the reference time series data measured at the location points and period segments within the target group are more likely to be similar to the target time series data than the reference time series data measured at the location points and period segments that belong to other location and period groups. Therefore, by prioritizing the similarity determination of the reference time series data measured at the location points within the target group, it is possible to reliably improve the efficiency of the search for similar time series data.

[0061] Furthermore, if there is no similar time series data among the reference time series data measured at each location and period segment within the target group, the prediction correction module 343 may determine whether the reference time series data measured at each location in other location groups is similar to the target time series data. Specifically, the prediction correction module 343 may determine the similarity of the location factor data and period segment factor data between multiple location and period groups (location and period group similarity), and may preferentially determine whether the reference time series data measured at each location in other location groups with relatively high location and period group similarity is similar to the target time series data. This ensures efficient search for similar time series data by prioritizing location and period groups that are likely to be similar to the target time series data, even if similar reference time series data (similar time series data) does not exist within the target group.

[0062] FIG. 5C is a conceptual diagram illustrating grouping by the prediction correction module 343. In this example, for example, the prediction correction module 343 classifies each placement location into five location / period groups, Gr1 to Gr5, based on the similarity of the location factor data and the period segment factor data (groups Gr3 to Gr5 are not shown in FIG. 5C). Each group includes multiple combinations of placement location and period segment. These combinations are referred to as members constituting a location / period group. For example, group Gr1 includes members M11 to M15 as five combinations (members) of placement location and period segment. FIG. 5C shows the location segment indicating the placement location and the period segment ID indicating the period segment for each member. In the location-period group, even the same location location is classified into each group as a different member for each period segment (for example, member M11 and member M21, etc.). When grouping only the placement points, each placement point belongs to only one group, and the same placement point does not overlap with multiple groups.

[0063] In this example, member M11 (location category: A001 (location location a1), period category ID: S1 / Wd / U3 (spring, weekday, morning)) is the member that indicates the location location and time period of the prediction target. Therefore, group Gr1 is the target group. For example, within group Gr1, which is the target group, the prediction correction module 343 preferentially determines whether the reference time series data that corresponds to the location location and period category indicated by members M12 to M15 is similar to the target time series data. If similar time series data is not found in group Gr1 (if there is no reference time series data with a similarity equal to or greater than a predetermined threshold), the prediction correction module 343 determines whether the reference time series data corresponding to the location location and period segment indicated by each member (combination of location location and period segment) of other location-period groups is similar to the target time series data. In other words, the prediction correction module 343 searches for similar time series data from reference time series data corresponding to members of other location-period groups that have a relatively high location-period group similarity with group Gr1 (target group).

[0064] In this example, among the other location-period groups (groups Gr2 to Gr5), group Gr2 is the group with the highest location-period group similarity to group Gr1 (target group). In this case, the prediction correction module 343 searches for similar time series data within group Gr2, prioritizing it over other groups. For example, the prediction correction module 343 sequentially determines the similarity between the target time series data and the reference time series data corresponding to member M21 (location category: A001 (placement location a1), period category ID: S2 / Wd / U6 (summer, weekday, afternoon)). If similar time series data is not found in group Gr2, the prediction correction module 343 may search for similar time series data in other location / period groups and gradually expand the search range. For example, the prediction correction module 343 may search for similar time series data in other location / period groups in descending order of location / period group similarity.

[0065] When determining the similarity between groups (location group similarity, location-period group similarity), the prediction correction module 343 may use the distance (similarity) of demand factors (location factor data or period segment factor data) for each location location or for each location location and period segment derived during the grouping process. For example, other groups having members who are close to members in the target group (location location, or a combination of location location and period segment) may be determined to be groups with high similarity to the target group (high group similarity). The forecast correction module 343 may also predetermine a representative value of the demand factor data and determine the similarity between the representative value of the target group and the representative values ​​of other groups. That is, when determining the similarity between location groups, the similarity may be determined between the representative values ​​of the location factor data of the target group and the location factor data of other groups. When determining the similarity between location-period groups, the similarity may be determined between the representative values ​​of the location factor data and period segment factor data of the target group and the location factor data and period segment factor data of other groups.

[0066] Furthermore, the forecast correction module 343 may store, for each deployment point in one service area (service area A), the deployment point (achieved deployment point) where proven reference time series data that was used as similar time series data when correcting past demand forecast values ​​was measured, in a predetermined storage area of ​​the storage device 34b. For example, if the reference time series data used as similar time series data for allocation point a1 in service area A is measured at allocation point b1 in service area B, the prediction correction module 343 may store this "allocation point b1 (location category ID: B001)" as an actual allocation point in a predetermined storage area in association with the allocation point a1 of the prediction target. The prediction correction module 343 may store (accumulate) the actual allocation point each time it corrects the demand forecast value for each allocation point. When similar time series data (reference time series data) measured at the same actual allocation point is repeatedly used to correct one allocation point of the prediction target, the prediction correction module 343 may simply add up the number of times the same allocation point is used as an actual allocation point.

[0067] When correcting the demand forecast value, the forecast correction module 343 may extract from the data management device 2 a group of actual time series data, including reference time series data measured at one or more actual deployment locations, and determine whether the reference time series data included in the group of actual time series data is similar to the target time series data, giving priority to the search for similar time series data by grouping as described above. That is, the forecast correction module 343 may search for similar time series data in the group of actual time series data based on the actual deployment locations, rather than searching for similar time series data by grouping as described above. For example, the forecast correction module 343 may first search for similar time series data in the group of actual time series data, and if no similar time series data is found among the reference time series data constituting the group of actual time series data, perform the search for similar time series data by grouping as described above. Furthermore, the forecast correction module 343 may calculate the similarity in the group of actual time series data, for example, starting with reference time series data measured at actual deployment locations with the highest usage frequency. This allows the search for similar time series data to be prioritized, giving priority to reference time series data estimated to have a high similarity to the target time series data. For example, if "allocation point b1 (location category ID: B001)" is newly stored as an actual allocation point corresponding to allocation point a1 as described above, the next time the demand forecast value of allocation point a1 is corrected, data measured at allocation point b1 will be newly included in the actual time series data group. By searching for similar time series data using the actual time series data group consisting of reference time series data measured at actual allocation points used for correction in the past, it is possible to identify similar time series data more efficiently and quickly.

[0068] The prediction correction module 343 may store, as the actual period segment, the period segment of the reference time series data with the record used as the similar time series data, in addition to the actual placement point. That is, for each placement point in one service area A, the prediction correction module 343 may store, as the actual measurement information, the placement point (actual placement point) where the reference time series data with the record used as the similar time series data was measured and the period segment (actual period segment) to which the measured time period belongs, in a predetermined storage area. For example, if the reference time series data with the record used as the similar time series data for placement point a1 in service area A is data measured at placement point b1 in service area B in the period segment "spring, weekday, afternoon," the prediction correction module 343 may store (accumulate) the actual placement point (placement point b1 (location segment ID: B001)) and the actual period segment (spring, weekday, afternoon (period segment ID: S1 / Wd / U6)) in association with the placement point a1 to be predicted as the actual measurement information. The actual measurement information may be stored in association with the placement point of the prediction target (e.g., placement point a1) and the period segment of the prediction target (e.g., spring, weekday, morning). When similar time series data (reference time series data) based on the same actual measurement information is repeatedly used to correct one placement point of the prediction target, the prediction correction module 343 may add up the number of times the actual measurement information is used.

[0069] When correcting the demand forecast value, the forecast correction module 343 may extract a group of actual time series data including reference time series data from the data management device based on one or more pieces of actual measurement information, and determine whether the reference time series data included in the group of actual time series data is similar to the target time series data, giving top priority to the reference time series data included in the group of actual time series data. In other words, the forecast correction module 343 may search for similar time series data from the group of actual time series data based on the actual measurement information (actual placement location and actual period division), giving priority to the search for similar time series data by the above-mentioned grouping. By using a group of actual time series data consisting of reference time series data measured at actual placement locations and time periods used for correction in the past as the search target for similar time series data, it is possible to identify similar time series data more efficiently and quickly. The prediction correction module 343 may calculate the similarity in the group of actual time series data, for example, in descending order of reference time series data corresponding to actual measurement information used frequently. This allows the search for similar time series data to prioritize reference time series data that is estimated to have a high similarity to the target time series data.

[0070] Each module in the demand prediction device 3 has been described above. In order to efficiently search for similar time series data, the forecast correction module 343 preferably first searches the actual time series data group. If there is no reference time series data corresponding to similar time series data in the actual time series data group, similar time series data may be searched for by grouping as described above, or reference time series data at each deployment point for each service area may be searched for sequentially. For example, for each deployment point (deployment points a1 to a8) in service area A, reference time series data may be extracted from the previous unit time by going back n steps from the previous unit time to cover all past demand measurement values, and similar time series data may be searched for sequentially. If there is no similar time series data in the reference time series data covering the past period of each deployment point in service area A, reference time series data may be extracted sequentially for each deployment point in other service areas B and C in the same way to search for similar time series data. In the demand forecasting device 3, for each deployment point in one service area (service area A), processing is executed in the predicted value calculation module 341, the error estimation module 342, and the prediction correction module 343, whereby calculation and correction of the demand forecast value is performed for each deployment point (deployment points a1 to a8), and the calculated and corrected demand forecast value is transmitted to the vehicle dispatch management device 4.

[0071] (Configuration of vehicle dispatch management device) Next, the vehicle allocation management device 4 in the vehicle allocation management system 1 will be described with reference to Fig. 6. The vehicle allocation management device 4 acquires demand forecast results (demand forecast values) for each allocation point from the demand forecasting device 3, and creates an allocation plan for vehicle V for each allocation point in one service area (service area A) based on the demand forecast results.

[0072] Furthermore, the vehicle dispatch management device 4 creates a dispatch plan for vehicle V based on a dispatch request from a user wishing to board vehicle V, and dispatches vehicle V from each dispatch point to the boarding location desired by the user. Furthermore, the vehicle dispatch management device 4 may transmit to the demand measurement device 6 dispatch request information that associates the dispatch request with the dispatch point where vehicle V dispatched to the user was located. This allows the demand measurement device 6 to tally up the number of requests (number of dispatch requests) per unit time for each dispatch point as a demand measurement value. Note that this is not limited to this, and the demand measurement device 6 may extract dispatch request information and dispatch plan information stored in a storage device 44b (described later) of the vehicle dispatch management device 4, and measure the number of requests per unit time for each dispatch point based on these.

[0073] The vehicle dispatch management device 4 is composed of an input device 41, an output device 42, a communication device 43, and a controller 44. The vehicle dispatch management device 4 may be, for example, an information processing device such as a personal computer or a server computer. The configurations of the input device 41, the output device 42, and the communication device 43 are the same as those of the input device 21, the output device 22, and the communication device 23 in the data management device 2, so detailed explanations will be omitted.

[0074] The controller 44 is an electronic control unit (ECU) that controls the operation of the vehicle dispatch management device 4. The controller 44 includes a processor 44a and a storage device 44b. The processor 44a has the same configuration as the processor 24a in the data management device 2, and therefore a detailed description thereof will be omitted. The storage device 44b has basically the same configuration as the storage device 24b in the data management device 2, but stores different programs. The storage device 44b stores programs for implementing control modules such as the vehicle allocation management module 441. The processor 44a executes these programs to implement the functions of the vehicle allocation management device 4.

[0075] When the demand forecast value is transmitted from the demand forecasting device 3 as the demand forecast result for each deployment point, the vehicle allocation management module 441 creates a deployment plan for each deployment point based on the demand forecast value. For example, the vehicle allocation management module 441 determines a vehicle V to be "relocated" by moving its allocation point based on the current location information of multiple vehicles V and the demand forecast value. The current location information of the vehicle V is stored in a predetermined storage area of ​​the storage device 44b as "vehicle information" together with a "vehicle ID" that identifies the vehicle V. The vehicle allocation management module 441 derives the surplus or shortage of vehicles V currently allocated at each allocation point based on the demand forecast value and vehicle information, and moves the vehicles V from an allocation point where there is a surplus of vehicles V (surplus allocation point) to an allocation point where there is a shortage of vehicles V (shortage allocation point). The vehicle allocation management module 441 extracts vehicles V that are allocated at surplus allocation points from the group of vehicles V, and transmits instructions (movement instruction information) to the extracted vehicles V to move to the shortage allocation point. The movement instruction information includes at least the location information of the allocation point to which the vehicles are to be moved.

[0076] The vehicle V is equipped with a communication device (not shown) and an output device (not shown) that can output information from the vehicle dispatch management device 4, and is configured to be able to receive and output instructions from the vehicle dispatch management device 4. For example, the vehicle V displays movement instruction information on the output device (e.g., a display device). Furthermore, upon receiving the movement instruction information, the vehicle V uses a navigation function to search for a movement route to the destination location and navigates the driver. Alternatively, in the case of a fully autonomous vehicle, the vehicle V that receives the movement instruction information moves to the destination location by autonomous driving. In this way, the vehicle dispatch management device 4 deploys the vehicles V at each deployment point within one area (service area A) based on the demand forecast value calculated by the demand forecasting device 3 as described above. This allows the vehicles V to be deployed so as to increase the utilization rate of the vehicles V at each deployment point and improve user convenience.

[0077] The vehicle dispatch management module 441 also creates a dispatch plan for vehicle V based on a dispatch request received from a user's terminal device (not shown) via the communication device 43, and dispatches vehicle V from each deployment point to the user's desired boarding location. The vehicle dispatch management system 1 provides, for example, an on-demand mobility service (vehicle dispatch service) using multiple vehicles V, targeting multiple service areas A to C. For this reason, the vehicle dispatch request includes information on the boarding location (desired boarding location) where the user wishes to board vehicle V, the disembarking location (desired disembarking location) where the user wishes to disembark from vehicle V, and the number of people using vehicle V, but does not include information on the date and time (desired boarding date and time) at which the user wishes to board vehicle V. The vehicle dispatch plan may include, for example, the location of the vehicle V dispatched to the user, a driving schedule, and a driving route. The driving schedule may include the estimated time of arrival at the boarding location where the user boards and the estimated time of arrival at the disembarking location where the user disembarks. The driving route indicates the driving route from the boarding location to the disembarking location.

[0078] For example, when the vehicle allocation management module 441 creates a vehicle allocation plan, it transmits the vehicle allocation plan to a predetermined vehicle V that is allocated at a location closest to the boarding location, and moves the vehicle V to the boarding location by the desired boarding date and time. In this way, the vehicle V is allocated to the user. Furthermore, for example, when the vehicle allocation management module 441 creates a vehicle allocation plan and allocates a vehicle V to a user, it associates the vehicle allocation plan with the date and time when the vehicle allocation request was received, and transmits this as vehicle allocation request information to the demand measurement device 6. This enables the demand measurement device 6 to measure the demand measurement value (number of requests) per unit time at each allocation point. Note that the vehicle allocation request information may include the contents of the vehicle allocation request.

[0079] (Operation of the vehicle dispatch management system) 7 is a flowchart showing an example of a vehicle dispatch management method in the vehicle dispatch management system 1 according to an embodiment of the present disclosure. This method is realized by each device constituting the vehicle dispatch management system 1 executing a vehicle dispatch management program under the control of a processor, thereby cooperating with hardware resources.

[0080] 7, the demand forecasting device 3 (forecasted value calculation module 341) acquires sample data 241a corresponding to the factor classification (location classification (location ID) and period classification (period classification ID)) of the forecast target (S701). Specifically, the forecasted value calculation module 341 extracts sample data 241a (demand measurement data, demand factor data) corresponding to the location classification indicating the placement location of the forecast target and the period classification to which the unit time of the forecast target belongs from the data management device 2 (S701). Next, the demand forecasting device 3 (the forecasted value calculation module 341) calculates a demand forecast value based on the extracted sample data 241a (S702). Next, the demand forecasting device 3 (forecast value calculation module 341) evaluates the forecast error for the most recent past period at the placement point of the forecast target (S703). Specifically, the forecast value calculation module 341 extracts the above-mentioned error time series data from the data management device 2, and evaluates whether the error time series data satisfies the above-mentioned forecast conditions (i) and (ii).

[0081] If the demand forecasting device 3 (forecast value calculation module 341) determines based on the evaluation result that the above-mentioned forecast error time series data (the above-mentioned forecast error) satisfies both of the forecast conditions (i) and (ii), i.e., the forecast error in the error time series data is on a decreasing trend and the latest forecast error data (the error between the demand measurement value and the demand forecast value for the immediately preceding unit time) is below a predetermined threshold (Yes in S704), it transmits the calculated demand forecast value to the vehicle dispatch management device 4 (S705).

[0082] On the other hand, if the demand forecasting device 3 (forecast value calculation module 341) determines that the error time series data (the above-mentioned forecast error) does not satisfy both of the forecast conditions (i) and (ii) (No in S704), it outputs the demand forecast value to the error estimation module 342, and the error estimation module 342 determines whether the time progression of the forecast error (error time series data) is linear (S706). Specifically, the error estimation module 342 generates error-transformed data by logarithmically transforming the error time-series data, and if it determines that the time transition of at least one of the error time-series data and the error-transformed data is linear (Yes in S706), it estimates a future forecast error in the latest demand forecast value based on the time transition of the forecast error (error time-series data) (S707). Specifically, the error estimation module 342 estimates a future forecast error in the latest demand forecast value calculated using a model that mathematically represents the time transition of the error time-series data.

[0083] Next, the demand prediction device 3 (error estimation module 342) transmits the latest demand prediction value corrected by the estimated value of the prediction error to the vehicle dispatch management device 4 (S708). Specifically, the error estimation module 342 adds the estimated value to the latest demand prediction value and transmits the result as the demand prediction value to the vehicle dispatch management device 4. On the other hand, if the demand forecasting device 3 (error estimation module 342) determines that the time progression of both the error time-series data and the error converted data is not linear (nonlinear) (No in S706), it outputs the latest demand forecast value to the forecast correction module 343 and requests correction of the demand forecast value. When requested to correct the demand forecast value, the forecast correction module 343 performs correction processing of the demand forecast value based on reference time-series data similar to the target time-series data measured at the deployment point of the prediction target, i.e., similar time-series data (similarity correction processing) (S709). Next, the demand forecasting device 3 (forecast correction module 343) transmits the demand forecast value corrected by the all-area data correction processing to the vehicle dispatch management device 4 (S710).

[0084] The vehicle allocation management device 4 (vehicle allocation management module 441) creates an allocation plan based on the demand forecast value transmitted from the demand forecast device 3, and allocates the vehicle V at the target allocation point (S711). Furthermore, by performing the processing of steps S701 to S711 for each deployment point in one service area (for example, service area A), it is possible to calculate demand forecast values ​​for each deployment point (deployment points a1 to a8) in one service area A and create a deployment plan.

[0085] 8 is a flowchart illustrating an example of a similarity correction process among the corrections of demand forecast values ​​in the vehicle dispatch management system 1 according to an embodiment of the present disclosure. The figure shows details of the process of S709 shown in FIG. The demand forecasting device 3 (forecast correction module 343) extracts, as target time series data, target time series data indicating demand measurement values ​​for the most recent past period at the placement point (for example, placement point a1) to be forecasted in time series form from the data management device 2 (S801). The forecast correction module 343 normalizes the extracted target time series data as described above. Next, the demand forecasting device 3 (forecast correction module 343) extracts a group of actual time series data from the data management device 2 based on the performance measurement information (performance placement location and performance period division) (S802). For example, as described above, the forecast correction module 343 extracts, as a group of actual time series data, one or more reference time series data that have been used as similar time series data during past corrections of demand forecast values ​​at the placement location and time period (unit time) to be forecast. The forecast correction module 343 normalizes the reference time series data in the extracted group of actual time series data in the same way as the similar time series data.

[0086] Next, the demand forecasting device 3 (forecast correction module 343) calculates the similarity between the reference time series data constituting the actual time series data group and the target time series data (S803), and if it determines that there is reference time series data in the actual time series data group whose similarity with the target time series data (target data) is equal to or greater than a predetermined threshold (Yes in S804), it determines the reference time series data whose similarity is equal to or greater than the threshold as similar time series data (S811). On the other hand, if the demand forecasting device 3 (forecast correction module 343) determines that the actual time series data group does not contain reference time series data whose similarity with the target time series data is equal to or greater than a predetermined threshold (No in S804), it groups and classifies each of the deployment points in the multiple service districts A to C into multiple groups (S805). The forecast correction module 343, for example, classifies each deployment point into multiple location groups based on the similarity of the location factor data, and further classifies (regroups) the combinations of each deployment point and period segment into multiple location-period groups based on the similarity of the demand factor data (location factor data and period segment factor data).

[0087] Next, the demand forecasting device 3 (forecast correction module 343) identifies the location / period group to which the placement location and time period (unit time) of the forecast target belong as the target group (S806), and extracts (and normalizes) reference time series data corresponding to other members (combinations of placement location / period division) other than the forecast target in the target group from the data management device 2 (S807). The demand forecasting device 3 (forecast correction module 343) calculates the similarity between the reference time series data (extracted data) corresponding to other members of the target group and the target time series data (target data), and if it determines that there is extracted data whose similarity is greater than or equal to a predetermined threshold (Yes in S808), it determines the reference time series data (extracted data) whose similarity is greater than or equal to the threshold as similar time series data (S811).

[0088] On the other hand, when the demand forecasting device 3 (forecast correction module 343) determines that there is no reference time series data (extracted data) corresponding to other members of the target group whose similarity with the target time series data (target data) is equal to or greater than a predetermined threshold (No in S808), it extracts (and normalizes) reference time series data corresponding to members of other location / period groups that have a high similarity (location / period group similarity) with the target group from the data management device 2 (S809). Next, the demand forecasting device 3 (forecast correction module 343) calculates the similarity between the reference time series data (extracted data) corresponding to members of other location / period groups and the target time series data (target data), and if it determines that there is extracted data whose similarity is equal to or greater than a predetermined threshold (Yes in S810), it determines the reference time series data (extracted data) whose similarity is equal to or greater than the threshold as similar time series data (S811). If there are multiple other location / period groups, the process continues to determine the similarity of the reference time series data corresponding to the members of each group to the target time series data (S810) until reference time series data with a similarity equal to or greater than the threshold is derived. At this time, similar time series data is searched for, for example, by prioritizing other location / period groups with higher location / period group similarity (starting from other location / period groups with higher period group similarity). This allows for efficient search for similar time series data targeting other location / period groups.

[0089] On the other hand, when the demand forecasting device 3 (forecast correction module 343) determines that there is no reference time series data (extracted data) corresponding to members of other location / period groups whose similarity with the target time series data is equal to or greater than a predetermined threshold (No in S810), it determines the reference time series data with the highest similarity among the reference time series data whose similarity with the target time series data has been determined so far as the similar time series data (S812). For example, each time the prediction correction module 343 calculates the similarity of reference time series data to target time series data, it stores the similarity and data that can identify the reference time series data in a specified memory area of ​​the storage device 34b, and if the similarity of the reference time series data corresponding to each member of all groups (location / period groups) is less than a specified threshold, it identifies the reference time series data with the highest similarity.

[0090] When the demand forecasting device 3 (forecast correction module 343) determines the similar time series data, it calculates the scale ratio between the target time series data (target data) and the similar time series data (S813), extracts from the data management device 2 the similar measurement value, which is the demand measurement value for the next unit time (next time slot) following the period in which the similar time series data was measured, and sets the value obtained by correcting the scale of the similar measurement value by the scale ratio as the latest demand forecast value (S814). In other words, the demand forecast value calculated by the forecast value calculation module 341 is corrected (updated) based on the similar measurement value. In this example, in step S805, each deployment point in the multiple service areas A to C is classified into multiple location / period groups, but this is not limited to this, and each deployment point may be classified into multiple location groups to search for similar error data.

[0091] As described above, in the vehicle dispatch management system 1, the demand forecasting device 3 extracts target time series data, which indicates in time series the demand measurement values ​​measured in the most recent past period at the deployment point to be forecasted, and reference time series data, which is data other than the target time series data and indicates in time series the demand measurement values ​​measured in a predetermined past period at each deployment point in multiple service areas (service areas A to C), from the data management device 2 via the communication network 9 (S801, S802, S807, S809), and corrects the demand forecast value based on similar time series data, which is reference time series data similar to the target time series data (S804: Yes, S808: Yes, S810: Yes, S811, S813, S814). As a result, the vehicle dispatch management system 1 can correct the demand forecast value based on a wide range of measurement data for all of the multiple service areas A to C, and improve the accuracy of the demand forecast for the vehicle V at each deployment point.

[0092] (Variation) In the above embodiment, in the correction by the demand prediction device 3 (prediction correction module 343), the demand measurement value (similar measurement value) of the next unit time following the period in which similar time series data was measured (for example, "T+1" shown in Figure 5B) was used to correct the demand prediction value, but the present disclosure is not limited to this. For example, the demand forecasting device 3 may calculate a similar predicted value that predicts the demand for vehicle V in a predetermined future time period (the unit time of the prediction target) at the deployment point (similar deployment point) where the similar time series data was measured. If the similar time series data satisfies a predetermined condition, the demand forecasting device 3 may use a scale-corrected value of the similar predicted value based on the scale ratio between the target time series data and the similar time series data as the demand forecasting value at the deployment point of the prediction target. By using the demand forecasting value of the similar deployment point where similar time series data that satisfies the predetermined condition (described below) measured, i.e., data that is close to the current situation at the deployment point and unit time of the prediction target, for the correction, the accuracy of the demand forecast for vehicle V at the deployment point of the prediction target can be further improved. Here, the unit time of the prediction target may be the unit time next to the current unit time (the future unit time "t" shown in FIG. 5A), as described above.

[0093] Specifically, in this modification, the demand forecasting device 3 extracts, from the data management device 2, similar error data that indicates, in time series, the forecast errors for the most recent past period in which the similar time series data in the reference time series data was measured, and sets the similar forecast value as the demand forecast value for the target deployment point if the similar error data satisfies both of the above-mentioned forecast conditions (i) and (ii) (first condition), or if the time progression of at least one of the similar error data and data obtained by logarithmically transforming the similar error data (similar error transformed data described below) is linear (second condition). As a result, the similar forecast value that satisfies the predetermined conditions (the above-mentioned first and second conditions) and is unlikely to cause a future forecast error can be used to correct the demand forecast value, thereby further reliably improving the accuracy of the demand forecast for the vehicle V.

[0094] More specifically, in this modification, the prediction correction module 343 determines similar time series data from reference time series data measured during the same period (the most recent period in the past) as the period during which the target time series data was measured, and extracts similar error data indicating, in time series, the prediction error data for the most recent period in the past during which the similar time series data was measured from the data management device 2 via the communication network 9. Note that the inter-district data sharing device 7 stores other-district prediction error data indicating the error between the demand measurement value and the demand forecast value at each deployment point in other service districts (service districts B and C), and the inter-district data sharing device 7 can transmit the other-district prediction error data to the data management device 2. Therefore, even if the similar deployment point where the similar time series data was measured is the other service districts B and C, the prediction correction module 343 can extract similar error data corresponding to the measurement period of the similar time series data from the data management device 2. The prediction correction module 343 outputs the extracted similarity error data to the prediction value calculation module 341, and requests it to calculate a similarity prediction value.

[0095] When the similar error data is output from the prediction correction module 343, the prediction value calculation module 341 calculates a demand forecast value for the similarly deployed point in the same way as when calculating a demand forecast value for the deployment point of the prediction target. For example, if the similarly deployed point belongs to one service district (service district A), the prediction value calculation module 341 extracts sample data 241a (demand measurement data, demand factor data) corresponding to the location category of the similarly deployed point and the period category to which the unit time of the prediction target belongs from the data management device 2. Furthermore, if the similarly deployed point belongs to either of the other service districts B or C, the prediction value calculation module 341 extracts other district data (demand measurement data, demand factor data) corresponding to the location category of the similarly deployed point and the period category to which the unit time of the prediction target belongs from the data management device 2. The predicted value calculation module 341 calculates the demand forecast value of the similarly located point in the unit time of the forecast target as a similar predicted value based on the extracted data (sample data 241a or data from other areas), and if the similar error data satisfies both of the above-mentioned prediction conditions (i) and (ii) (the forecast error is on a decreasing trend and the latest forecast error is equal to or less than a predetermined threshold), outputs the calculated similar predicted value to the forecast correction module 343. Note that if the latest (next unit time) demand forecast value of the similarly located point already exists, the demand forecast value may be extracted from the data management device 2.

[0096] On the other hand, when the similar error data does not satisfy both of the above-mentioned prediction conditions (i) and (ii), the predicted value calculation module 341 outputs the similar error data and the similar predicted value to the error estimation module 342. When the similar error data and the similar predicted value are output, the error estimation module 342 estimates the detailed error of the similar predicted value in the same way as the estimation of the future prediction error described above. Specifically, the error estimation module 342 generates similar error transformed data by logarithmically transforming the prediction errors in the error time-series data, and when the time progression of at least one of the similar error data and the similar error transformed data is linear, estimates the error of the similar predicted value using a model that mathematically represents the time progression of the data, and outputs the similar predicted value corrected with the estimated value to the prediction correction module 343. The forecast correction module 343 scale-corrects the similar forecast value output from the forecast value calculation module 341 or the error estimation module 342 based on the scale ratio between the target time series data and the similar time series data, and sets the scale-corrected similar forecast value as the demand forecast value at the placement point of the forecast target.

[0097] (Effects of the embodiment) (1) The vehicle dispatch management system 1 of this embodiment includes a data management device 2 that stores demand measurement values ​​indicating the demand for vehicle V measured at each deployment point where vehicle V is deployed in one service area (service area A), an inter-area data sharing device 7 that stores demand measurement values ​​measured at each deployment point in the remaining service areas (service areas B and C) and is capable of transmitting the demand measurement values ​​to the data management device 2, a demand prediction device 3 that calculates a demand forecast value that predicts the demand for vehicle V at each deployment point in one service area for a specified future time period based on the demand measurement values, and a vehicle dispatch management device 4 that deploys vehicle V at each deployment point in one service area based on the demand forecast value. The demand forecasting device 3 (forecast correction module 343) extracts, from the data management device 2 via the communication network 9, time series data indicating demand measurement values ​​measured in the most recent past period at the deployment point to be predicted, and reference time series data which is data other than the target time series data and indicates demand values ​​measured in a predetermined past period at each deployment point of multiple service areas A to C, and corrects the demand forecast value based on the similar time series data which is reference time series data similar to the target time series data. According to this configuration, it is possible to search for and determine similar time series data to be used for correcting the demand forecast value from a wide range of data (reference time series data) including not only the deployment point of the forecast target but also past demand measurement values ​​at each deployment point of multiple service districts A to C. Therefore, it is possible to search for data with a higher degree of similarity than when the deployment point of the forecast target is the search target for similar time series data, and as a result, it is possible to improve the accuracy of the demand forecast for vehicles at each deployment point. (2) The demand forecasting device 3 (forecast correction module 343) may extract, from the data management device, the demand value measured in the next unit time following the specified period at the location where the similar time series data was measured, as a similar measurement value, and may perform a correction to set the demand forecast value as a value obtained by correcting the scale of the similar measurement value based on the scale ratio between the target time series data and the similar time series data. According to this configuration, by correcting similar measurement values ​​using the above-mentioned scale ratio, demand measurement values ​​measured at other placement points can be adapted as demand forecast values ​​for the placement point to be predicted, thereby further improving the accuracy of demand forecasting for vehicle V.

[0098] (3) The data management device 2 stores a forecast error indicating the error between the demand measurement value and the demand forecast value at each location point in a service area, and the demand forecasting device 3 (forecast value calculation module 341) extracts error time series data indicating the forecast error in a time series for the most recent past period in which the target time series data was measured from the data management device 2, and determines whether or not the demand forecast value needs to be corrected based on the error time series data. This allows similar predicted values ​​that are less likely to cause prediction errors in the future to be used to correct the demand forecast value, thereby further improving the accuracy of the demand forecast for the vehicle V. In addition, the processing load on the demand forecasting device 3 can be reduced by suppressing unnecessary correction processes from being executed. (4) The demand forecasting device 3 (forecast value calculation module 341) may decide not to correct the demand forecast value if the forecast error in the error time series data is on a decreasing trend and the latest forecast error in the error time series data is below a predetermined threshold. This allows a similar forecast value that is less likely to cause a future forecast error to be reliably determined and used to correct the demand forecast value.

[0099] (5) When the demand forecasting device 3 (error estimation module 342, forecast correction module 343) determines that a correction of a demand forecast value is to be performed, it determines whether the time progression of at least one of the error time series data and the data obtained by logarithmically transforming the forecast error in the error time series data (error transformation data) is linear, and if the time progression is linear, it corrects the demand forecast value based on the error of the demand forecast value estimated using a model that mathematically represents the time progression of the error time series data, and if the time progression is nonlinear, it corrects the demand forecast value based on the above-mentioned similar time series data. This improves the accuracy of the demand forecast for vehicle V by enabling corrections based on the prediction error when the accuracy of the error estimation is determined to be high, and also reduces the processing load of the demand forecasting device 3 by preventing excessive execution of similar correction processing using a wide range of data. (6) The demand forecasting device 3 (forecast value calculation module 341, forecast correction module 343) calculates a similar forecast value that predicts the demand for vehicle V at a similar deployment point where similar time series data was measured during a specified future time period, and if the similar time series data satisfies specified conditions, may set the scale-corrected value of the similar forecast value based on the scale ratio between the target time series data and the similar time series data as the demand forecast value for the deployment point to be predicted. This allows the demand forecast values ​​of similar deployment locations, i.e., data that is close to the current situation at the deployment location and unit time being predicted, to be used for correction, thereby further improving the accuracy of the demand forecast for vehicle V at the deployment location being predicted. (7) The inter-area data sharing device 7 is capable of transmitting to the data management device 2 a forecast error indicating the error between the demand measurement value and the demand forecast value at each placement point of the remaining service areas B and C; the predetermined period in the past during which the reference time series data was measured is the same period as the most recent period in the past during which the target time series data was measured; The demand forecasting device 3 (forecast value calculation module 341, error estimation module 342, forecast correction module 343) may extract similar error data, which indicates in time series the forecast error in the most recent past period in which the similar time series data of the second time series data was measured, from the data management device via a predetermined network, and may set the similar forecast value as the demand forecast value of the placement point to be forecasted if the forecast error in the similar error data is on a decreasing trend and the latest forecast error is below a predetermined threshold, or if the time progression of the data is linear in at least one of the similar error data and data obtained by logarithmically transforming the similar error data (analogous error transformation data). This allows similar predicted values ​​that are less likely to cause prediction errors in the future to be used to correct the demand prediction value, thereby further improving the accuracy of the demand prediction for vehicle V.

[0100] (8) The demand forecasting device 3 (forecast correction module 343) may classify each deployment point in a plurality of service areas (service areas A to C) into a plurality of groups based on the similarity of the location factors that are characteristics of each deployment point among the demand factor data that affect the demand for vehicle V at each deployment point, or the time trend pattern of the average demand value per day at each deployment point (daily average demand trend pattern), and may select the group to which the deployment point to be predicted belongs as a target group among the plurality of groups, and may preferentially determine whether the reference time series data measured at each deployment point belonging to the target group is similar to the target time series data. This allows the reference time series data measured at a location whose characteristics (characteristics indicated by the location factor data) are similar to those of the location of the prediction target to be used as a search target for similar time series data, thereby ensuring efficient search for similar time series data. (9) If there is no data similar to the target time series data among the reference time series data measured at the placement points within the target group, the demand forecasting device 3 (forecast correction module 343) may determine the similarity of the location factor data between the multiple groups, and may preferentially determine whether the reference time series data measured at the placement points belonging to other groups that have a relatively high similarity to the target group are similar to the target time series data. This ensures efficient searching for similar time series data by prioritizing searching for other location groups that are likely to be similar to the target time series data, even if similar time series data does not exist within the target group.

[0101] (10) Among the demand factor data, the characteristics of each deployment point for each period segment divided by specified temporal conditions are used as period segment factors, and the demand forecasting device 3 (forecast correction module 343) regroups each deployment point classified into multiple location groups based on the similarity of the period segment factor data or the time transition pattern of the average demand measurement value in the period segment (segment average demand transition pattern) to classify them into multiple location / period groups, and uses the location / period group to which the deployment point and time period to be forecasted belong as a target group, and may preferentially determine whether reference time series data measured at each deployment point belonging to the target group is similar to the target time series data. This allows for more efficient searching for similar time series data by prioritizing target groups in grouping (regrouping) of locations and periods where the demand measurement values ​​(demand measurement data) are more likely to be similar, thereby making it possible to more reliably search for similar time series data. (11) When there is no reference time series data similar to the target time series data among the reference time series data measured at the placement locations and period segments within the target group, the demand forecasting device 3 (forecast correction module 343) may determine the similarity of the location factor data and period segment factor data between the above-mentioned multiple location / period groups, and may preferentially determine whether the reference time series data measured at the placement locations and period segments belonging to other location / period groups that have a relatively high similarity to the target group are similar to the target time series data. As a result, when grouping by location / period group, even if similar time series data does not exist within the target group, the location / period group that is likely to be similar to the target time series data is given priority in searching for similar time series data, thereby making it possible to more reliably and efficiently search for similar time series data.

[0102] (12) The demand forecasting device 3 (forecast correction module 343) may store, for each deployment point in one service area (service area A), the deployment point at which proven reference time series data used as similar time series data was measured as the achieved deployment point in a predetermined memory area, and when correcting the demand forecast value, may extract from the data management device 2 a group of achieved time series data including the reference time series data measured at one or more proven deployment points, and may give top priority to the reference time series data included in the group of achieved time series data to determine whether it is similar to the target time series data. This allows the search for similar time series data to be more efficient and the possibility of determining similar time series data at an early stage to be improved by prioritizing the search for similar time series data from a group of actual time series data including reference time series data that has a proven track record as similar time series data before grouping. (13) The demand forecasting device 3 (forecast correction module 343) may store, for each location in one service area A, the location where the proven reference time series data used as similar time series data was measured and the period category to which the measured time zone belongs as actual measurement information in a predetermined storage area, and when correcting the demand forecast value, may extract from the data management device 2 a group of actual time series data including the reference time series data based on one or more pieces of actual measurement information, and may give top priority to the reference time series data included in the group of actual time series data to determine whether it is similar to the target time series data. This allows the reference time series data included in the performance time series data group to be further narrowed down by period division, thereby making it possible to more reliably and efficiently search for similar time series data. (14) The demand forecasting device 3 (forecast correction module 343) extracts target time series data and reference time series data as time series data of a length indicated by a predetermined reference value, and if the time series in the target time series data or the reference time series data spans multiple days, it may divide the time series data by date and perform a series of processes to correct the demand forecast value using the time series data with the longest length. This prevents data with different time periods from being mixed in the time series data, and improves the accuracy of the following correction process. [Explanation of symbols]

[0103] 1. Vehicle dispatch management system 2 Data management device 21 Input Devices 22 Output Devices 23 Communication equipment 24 Controller 24a processor 24b Storage device 241 Sample Data Storage Module 241a Specimen data 242 Specimen Data Management Module 243 Other Area Data Acquisition Module 3. Demand forecasting device 31 Input Devices 32 Output Devices 33 Communication equipment 34 Controller 34a processor 34b Storage device 341 Prediction Value Calculation Module 342 Error Estimation Module 343 Prediction Correction Module 4. Vehicle dispatch management device 41 Input Devices 42 Output Devices 43 Communication equipment 44 Controller 44a processor 44b Storage device 441 Vehicle Dispatch Management Module Five-factor data collection device 6. Demand measurement device 7. Inter-district data sharing device 9. Communication Networks V vehicle

Claims

1. A vehicle allocation management system that manages the allocation of a plurality of vehicles to each of a plurality of districts, a data management device that stores a demand value indicating the demand for the vehicle measured at each location where the vehicle is located in one of the areas; a data sharing device that stores the demand values ​​measured at each deployment point in the remaining districts and is capable of transmitting the demand values ​​to the data management device; a demand prediction device that calculates a demand prediction value that predicts the demand for the vehicle in a future predetermined time period at each deployment point in the one district based on the demand value; a vehicle allocation management device that allocates the vehicles to each allocation point within the one district based on the demand forecast value, The demand prediction device extracting, from the data management device via a predetermined network, first time series data indicating, in time series, the demand values ​​measured in the most recent past period at the deployment points to be predicted, and second time series data, which is data other than the first time series data and indicates, in time series, the demand values ​​measured in a predetermined past period at each deployment point in the plurality of districts; The vehicle dispatch management system corrects the demand forecast value based on similar time series data, which is the second time series data similar to the first time series data.

2. the second time-series data is data indicating the demand values ​​measured within the predetermined period in time series for each unit time, The demand prediction device extracting, from the data management device, a demand value measured in a next unit time following the predetermined period at the placement point where the similar time series data was measured, as a similar measurement value; a correction is performed to set the demand forecast value to a value obtained by correcting the scale of the similar measurement value based on a scale ratio between the time series data and the similar time series data; The vehicle dispatch management system according to claim 1.

3. The data management device storing a prediction error indicating an error between the demand value and the demand prediction value at each deployment point in the one district; the demand forecasting device extracts, from the data management device, first error data indicating, in time series, a forecast error for a most recent period in the past in which the temporal series data was measured, and determines whether or not correction of the demand forecast value is necessary based on the first error data; The vehicle dispatch management system according to claim 1 or 2.

4. The demand prediction device determining not to correct the demand forecast value when the forecast error in the first error data is on a decreasing trend and the latest forecast error in the first error data is equal to or less than a predetermined threshold value; The vehicle dispatch management system according to claim 3 .

5. The demand prediction device when it is determined that the demand forecast value should be corrected, determining whether a time transition of at least one of the first error data and data obtained by logarithmically transforming the forecast error in the first error data is linear; When the time transition is linear, correcting the demand forecast value based on an error of the demand forecast value estimated using a model that mathematically represents the time transition of the first error data; When the time transition is nonlinear, the demand forecast value is corrected based on the similar time series data. The vehicle dispatch management system according to claim 3 .

6. The demand prediction device calculating a similar predicted value that predicts the demand in a future predetermined time period at the similar deployment point, which is the deployment point at which the similar time series data was measured; When the similar time series data satisfies a predetermined condition, a scale-corrected value of the similar prediction value based on a scale ratio between the time series data and the similar time series data is set as the demand prediction value of the deployment point to be predicted. The vehicle dispatch management system according to claim 1.

7. the data sharing device is capable of transmitting a prediction error indicating an error between the demand value and the demand prediction value at each deployment point in the remaining district to the data management device; the predetermined period in the past during which the second time series data was measured is the same period as the most recent period in the past during which the primary time series data was measured, The demand prediction device extracting, from the data management device via a predetermined network, similar error data that indicates, in time series, prediction errors in the most recent past period in which the similar time series data was measured, from the second time series data; When the prediction error in the similar error data is on a decreasing trend and the latest prediction error is equal to or less than a predetermined threshold, or when the time transition of at least one of the similar error data and data obtained by logarithmically transforming the similar error data is linear, the similar prediction value is set as the demand prediction value of the placement point to be predicted. The vehicle dispatch management system according to claim 6.

8. The demand prediction device classifying the deployment points in the plurality of districts into a plurality of groups based on a location factor that is a characteristic of each deployment point among demand factors that affect the demand for the vehicle at each deployment point, or a similarity in a time transition pattern of the average daily demand value at each deployment point; 2. The vehicle dispatch management system according to claim 1, wherein, among the plurality of groups, a group to which a deployment point to be predicted belongs is set as a target group, and the second time series data measured at each deployment point belonging to the target group is preferentially determined to be similar to the first time series data.

9. The demand prediction device If there is no data similar to the temporal series data in the second time series data measured at the placement point in the target group, determine the similarity of the location factors between the plurality of groups; 9. The vehicle dispatch management system according to claim 8, wherein the second time series data measured at the deployment location belonging to another group having a relatively high similarity to the target group is preferentially determined to be similar to the first time series data.

10. Among the demand factors, the characteristics of each of the allocation points for each period division divided by predetermined time conditions are set as period division factors, The demand prediction device regrouping the placement points classified into the plurality of groups based on the period classification factor or the similarity of the time transition pattern of the average demand value in the period classification into a plurality of location / period groups; The location / period group to which the placement location and time period of the prediction target belong is defined as the target group, and the second time series data measured at each placement location belonging to the target group is preferentially determined to be similar to the first time series data. The vehicle dispatch management system according to claim 8.

11. The demand prediction device If there is no second time series data similar to the first time series data among the second time series data measured at the placement locations and period divisions within the target group, determine the similarity of the location factors and the period division factors between the plurality of location / period groups; The vehicle dispatch management system according to claim 10, wherein the second time series data measured at deployment locations and period segments belonging to other groups having a relatively high similarity to the target group are preferentially determined to be similar to the first time series data.

12. The demand prediction device For each deployment point in the one district, the deployment point at which the second time series data with a proven track record used as the similar time series data was measured is stored in a predetermined storage area as a proven deployment point; 9. The vehicle dispatch management system according to claim 1, wherein, when correcting the demand forecast value, a group of actual time series data including the second time series data measured at one or more of the actual deployment points is extracted from the data management device, and the second time series data included in the group of actual time series data is given top priority to determine whether it is similar to the first time series data.

13. The demand prediction device For each location in the one district, the location where the second time series data with a proven track record used as the similar time series data was measured and the period segment to which the measured time period belongs are stored as track record measurement information in a predetermined storage area; 11. The vehicle dispatch management system according to claim 10, wherein, when correcting the demand forecast value, a group of actual time series data including the second time series data is extracted from the data management device based on one or more pieces of actual measurement information, and the second time series data included in the group of actual time series data is given top priority to determine whether it is similar to the first time series data.

14. The demand prediction device extracting the first time series data and the second time series data as time series data having a length indicated by a predetermined reference value; 2. The vehicle dispatch management system according to claim 1, wherein, when a time series in the first time series data or the second time series data spans multiple days, the time series data is divided by date, and a series of processes for correcting the demand forecast value is performed using the time series data with the longest length.

Images