Congestion prediction device

Abstract

The present disclosure aims to provide a method of generating a learned model capable of predicting road conditions over a long period, a non-transitory storage medium, and a congestion prediction device. The method is implemented by a processor to perform the following processing: acquiring person number information indicating the number of persons who have departed from a facility within each of a plurality of predetermined periods, weather information for each of the predetermined periods, and vehicle information related to vehicles that have traveled in the vicinity of the facility, using the vehicle information to determine a congestion condition indicating the presence or absence of congestion on a road in the vicinity of the facility within the predetermined periods, and generating a learned model for predicting congestion on a road by machine learning using the person number information, the weather information, and the congestion condition to which the person number information and the weather information have a corresponding relationship as teaching data.

Description

Technical Field

[0001] This disclosure relates to a method for generating a learned model for predicting road congestion, a non-transitory storage medium, and a congestion prediction device. Background Technology

[0002] For example, Japanese Patent Application Publication No. 2014-191578 discloses a road traffic server that collects vehicle location information and individual information from vehicles, and predicts current and future traffic volume for vehicle movement paths based on the collected location and individual information. The road traffic server predicts current and future traffic volume several hours in advance based on information obtained from multiple vehicles and terminals owned by users riding in the vehicles, and provides alerts to users.

[0003] Traffic congestion is a significant social problem, necessitating effective congestion prediction methods. Current technologies rely on information gathered from multiple vehicles currently traveling on the road to predict dynamically changing congestion conditions.

[0004] However, it is designed to predict congestion conditions up to the nearest future (e.g., hours later) that are dynamically changing, and not necessarily for longer periods such as days later. Summary of the Invention

[0005] This disclosure provides a method for generating a learned model capable of predicting road conditions over a long period, a non-transitory storage medium, and a congestion prediction device.

[0006] In the first approach, the method obtains personnel information representing the number of users who departed from the facility during each pre-defined period, including users riding in vehicles, weather information for each pre-defined period, and vehicle information related to vehicles passing through the vicinity of the facility. The vehicle information is then used to determine congestion conditions, which indicate the presence or absence of congestion on roads surrounding the facility during the pre-defined period. A learned model for predicting road congestion is generated by using machine learning, which employs the personnel information, the weather information, and the congestion conditions that are correlated with the personnel information and the weather information, as teaching data.

[0007] In the first approach, information on the number of users departing from the facility, weather information, and vehicle information on roads surrounding the facility are obtained. In the machine learning approach, congestion information on roads surrounding the facility is assessed based on the vehicle information, and the learning model performs machine learning on the relationship between the obtained user information and the assessed congestion information. Here, "facility" refers to companies and commercial facilities, and "users departing from the facility" refers to users leaving work or returning home. This generates a learned model capable of predicting long-term road conditions.

[0008] The second approach is that, in the first approach, the number of people information includes the number of users involved in a period prior to the period in which the congestion situation is determined.

[0009] According to the second approach, the time from departure from the facility to arrival at the road has a significant impact on congestion, thus enabling the generation of a learned model that can predict congestion by taking into account factors that have a greater impact on congestion.

[0010] The third method is that, in the first or second method, the vehicle information includes location information related to the vehicle's position and speed information related to the vehicle's speed, and the location information and the speed information are used to determine the congestion situation.

[0011] According to the third approach, a learned model can be generated that can judge congestion conditions without being affected by severe weather or time periods such as night.

[0012] The fourth method is to use the location information and the speed information in the third method to derive the required time to pass through a pre-set interval, and the congestion is judged as congestion when the intervals with a required time of more than a threshold exist continuously in a pre-set number of intervals.

[0013] According to the fourth approach, the benchmark for representing congestion is made clear, thereby generating a learned model that can more accurately and effectively judge the congestion situation.

[0014] The fifth approach involves using a random forest model in any of the first to fourth approaches.

[0015] According to the fifth approach, compared with other existing machine learning models, a learned model that can predict congestion conditions with greater accuracy can be generated.

[0016] The sixth method's non-transitory storage medium stores a program that causes the processor to perform the following process: using a learned model generated by any one of the first to fifth methods, and predicting congestion conditions on roads surrounding the facility by inputting inferred number information representing the number of users inferred to have departed from the facility for each pre-defined period, including users riding in vehicles, and the weather information.

[0017] Based on the non-temporary storage medium of the sixth method, long-term road conditions can be predicted.

[0018] The congestion prediction device of the seventh method includes a memory, which serves as a non-temporary storage medium for the sixth method, and a processor. The processor infers the estimated number of people during the target period for which the congestion situation is predicted, and inputs the estimated number of people and the weather information into the learned model to predict the congestion situation during the target period.

[0019] According to the congestion prediction device of the seventh method, the user's load can be reduced in the process of predicting congestion conditions.

[0020] The eighth method of congestion prediction device, in the seventh method of congestion prediction device, uses the number of past users who have accessed the facility to infer the estimated number of users during the target period.

[0021] According to the congestion prediction device of the eighth method, it is possible to infer the number of visitors to the facility on the days involved in the target period based on past visitor numbers, and to infer the number of people departing from the facility during the target period based on the inferred visitor numbers. In other words, it is possible to predict congestion information by inferring future visitor numbers.

[0022] The ninth method of congestion prediction device provides a prompt regarding the congestion status during each predicted object period, as is done in the seventh or eighth method of congestion prediction device.

[0023] According to the congestion prediction device of the ninth method, users can be aware of the predicted congestion situation.

[0024] According to this disclosure, it is possible to predict road conditions over a long period of time. Attached Figure Description

[0025] Exemplary embodiments of the present invention will be described in detail with reference to the following accompanying drawings, wherein: Figure 1 A diagram illustrating the simplified structure of the congestion prediction system according to this embodiment; Figure 2 A block diagram illustrating the hardware structure of the vehicle involved in this embodiment; Figure 3 This is a block diagram illustrating the functional structure of the vehicle-mounted unit involved in this embodiment; Figure 4 A block diagram illustrating the hardware structure of the central server involved in this embodiment; Figure 5 This is a block diagram illustrating the functional structure of the central server involved in this embodiment; Figure 6 A graph showing the average time required for vehicles to pass through each section and each time period, used in the description of the inferences about congestion conditions involved in this embodiment. Figure 7 A diagram illustrating the structure of the learning data involved in this embodiment; Figure 8 A diagram illustrating the structure of the setting data involved in this embodiment; Figure 9 A graph illustrating an example of the proportion of the number of people leaving get off work for each date and each time period, used in the description of the inference of the number of people leaving get off work in this embodiment. Figure 10 This is a schematic diagram showing a screen displaying the prediction results involved in this embodiment; Figure 11 A data flow diagram illustrating the process of generating data in the process of generating a learned model as described in this embodiment; Figure 12 A data flow diagram illustrating the data flow in the congestion prediction process described in this embodiment; Figure 13 A flowchart illustrating the process of generating the learned model executed in the central server of this embodiment; and Figure 14 This is a flowchart illustrating the process of predicting congestion conditions performed in the central server of this embodiment. Detailed Implementation

[0026] The following congestion prediction system will be described, comprising: a vehicle equipped with the vehicle-mounted unit of this disclosure; a central server serving as a congestion prediction device; and an information providing server that displays weather information and attendance results information. The congestion prediction system uses the number of employees working at the facility, including those commuting by vehicle, and the number of employees leaving the facility for each pre-set period (time period) to predict the congestion conditions of roads surrounding the facility for each pre-set period (time period).

[0027] Here, the number of people leaving the facility in this embodiment is an example of "information on the number of people who have left the facility." Furthermore, "attendance" in this embodiment is an example of "visit." Additionally, this embodiment describes the facility as an office building owned by a company. However, it is not limited to this. Facilities can also be, for example, commercial facilities with multiple shops, and leisure facilities such as amusement parks. In this case, the congestion prediction system in this embodiment predicts the congestion situation of roads surrounding the commercial and leisure facilities based on the number of people arriving at the commercial and leisure facilities and the number of people leaving during each pre-set period (time period). Furthermore, in the following text, the pre-set period for predicting road conditions will be referred to as a "time period." Furthermore, "roads surrounding the facility" in this embodiment refers to roads that are directly or indirectly connected to the facility and are located within a pre-set distance from the facility.

[0028] (Overall structure) like Figure 1 As shown, the congestion prediction system 10 of this disclosure is configured to include a vehicle 12, a central server 30, and an information providing server 40. Furthermore, a vehicle-mounted unit 20, serving as an in-vehicle device, is mounted on the vehicle 12 and is interconnected with the central server 30 via a network N.

[0029] The vehicle-mounted unit 20 is a device for collecting vehicle-related information, including location information indicating the position of the vehicle 12 and speed information indicating the speed at which the vehicle 12 is traveling, and sending the aforementioned information to the central server 30.

[0030] The central server 30 is located, for example, at the manufacturer of vehicle 12 or a dealership of that manufacturer's vehicles. The central server 30 obtains vehicle information from the on-board unit 20 and uses the vehicle information of vehicle 12 traveling on roads surrounding the facility to determine the congestion status, indicating whether there is congestion on those roads.

[0031] In addition, the central server 30 obtains weather information and attendance data from the information providing server 40, as described later. The weather information includes weather conditions and precipitation, and the attendance data includes the number of employees present at the company or other facilities and the number of employees leaving for each time period. The central server 30 uses the obtained weather and attendance information to predict traffic congestion conditions on roads surrounding the facility for each time period.

[0032] Information providing server 40 is a server that provides weather information and attendance record information to central server 30. Furthermore, the weather information involved in this embodiment includes past meteorological observations and future weather forecasts, such as those for days when congestion is predicted. In addition, as attendance record information, the attendance record information includes the number of people who attended in the past and the number of people who left in the past.

[0033] (vehicle) like Figure 2 As shown, the vehicle 12 involved in this embodiment is configured to include an on-board unit 20 and multiple ECUs (Electronic Control Units) 22.

[0034] The vehicle-mounted unit 20 is configured to include a CPU (Central Processing Unit) 20A, a ROM (Read Only Memory) 20B, a RAM (Random Access Memory) 20C, an in-vehicle communication I / F (Interface) 20D, a wireless communication I / F 20E, and an input / output I / F 20F. The CPU 20A, ROM 20B, RAM 20C, in-vehicle communication I / F 20D, wireless communication I / F 20E, and input / output I / F 20F are interconnected via an internal bus 20G in a manner that enables them to communicate with each other.

[0035] CPU 20A is the central processing unit, which executes various programs or controls various parts. That is, CPU 20A reads programs from ROM 20B and uses RAM 20C as its working area to execute programs.

[0036] ROM 20B stores various programs and data. In this embodiment, ROM 20B stores a collection program 100. With the execution of the collection program 100, the vehicle unit 20 performs a process of collecting various data from vehicle-mounted devices and sensors (not shown) and sending them to the central server 30. RAM 20C temporarily stores programs or data as a working area.

[0037] The in-vehicle communication I / F20D is an interface used to connect to each ECU 22. This interface uses a communication standard conforming to the CAN communication protocol. The in-vehicle communication I / F20D is connected to the external bus 20H.

[0038] The wireless communication module I / F20E is used to communicate with the central server 30. This wireless communication module uses communication standards such as 5G, LTE, and Wi-Fi (registered trademark). The wireless communication module I / F20E is connected to network N.

[0039] The input / output I / F20F is an interface for communicating with the microphone 24, speaker 25, monitor 26, camera 27, GPS device 28, and input unit 29 mounted on the vehicle 12. Alternatively, the microphone 24, speaker 25, monitor 26, camera 27, GPS device 28, and input unit 29 can also be directly connected to the internal bus 20G.

[0040] ECU 22 is an electronic control module used to control onboard equipment and sensors (not shown) mounted on vehicle 12. Examples of ECU 22 include ADAS (Advanced Driver Assistance System) ECU, steering ECU, and engine ECU. In this embodiment, ECU 22 detects the speed of vehicle 12 using a vehicle speed sensor (not shown).

[0041] Microphone 24 is a device that is installed on the dashboard, center console, front pillar, and front wall of the vehicle compartment to pick up the voices of the occupants of the vehicle 12.

[0042] The speaker 25 is a device that is installed in the dashboard, center console, front pillar, or front wall of the vehicle body and is used to output voice.

[0043] Monitor 26 is a liquid crystal monitor installed on the dashboard, front wall of the passenger compartment, etc. of vehicle 12, and used to display various kinds of information. In addition, monitor 26 in this embodiment displays images and text representing exchange objects received from the central server.

[0044] Camera 27 is a camera device that is mounted on the upper part of the front window or adjacent to the interior rearview mirror and is used to capture images of the occupants in the vehicle 12.

[0045] GPS device 28 is a device for determining the current position of vehicle 12. GPS device 28 includes an antenna (not shown) for receiving signals from GPS satellites. Alternatively, GPS device 28 can also be connected to vehicle-mounted unit 20 via a vehicle navigation system (not shown).

[0046] The input section 29 includes a touch panel and buttons for input by passengers of the vehicle 12.

[0047] like Figure 3 As shown, in the vehicle-mounted unit 20 of this embodiment, the CPU 20A functions as a collection unit 200 and an output unit 210 by executing the collection program 100.

[0048] The collection unit 200 has the function of collecting speed information detected by sensors from the ECU 22 and location information obtained by the GPS device 28 as vehicle information.

[0049] The output unit 210 has the function of outputting the collected vehicle information to the central server 30.

[0050] (Central Server) like Figure 4 As shown, the central server 30 is configured to include a CPU 30A, ROM 30B, RAM 30C, memory 30D, and communication I / F 30E. The CPU 30A, ROM 30B, RAM 30C, memory 30D, and communication I / F 30E are connected together via an internal bus 30F in a manner enabling mutual communication. The functions of the CPU 30A, ROM 30B, RAM 30C, and communication I / F 30E are the same as those of the CPU 20A, ROM 20B, RAM 20C, and wireless communication I / F 20E of the vehicle-mounted unit 20 described above. Furthermore, the communication I / F 30E can also implement wired communication.

[0051] The storage device 30D, serving as memory, is constructed from either an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs and data. In this embodiment, the storage device 30D stores a prediction program 110, a vehicle information database 120, actual attendance result information 130, weather information 140, and a learned model 150. Additionally, the ROM 30B, also serving as memory, can store the prediction program 110, vehicle information database 120, actual attendance result information 130, weather information 140, and the learned model 150.

[0052] The prediction program 110 is a program used to control the central server 30. Along with the execution of the prediction program 110, the central server 30 obtains vehicle information from the vehicle-mounted unit 20 and attendance data (as headcount information) and weather information from the information providing server 40. The central server 30 uses the obtained vehicle information, attendance data, and weather information to generate the learned model 150, which will be described later, and to predict congestion conditions.

[0053] Vehicle information DB120 stores vehicle information obtained from multiple vehicles 12. Attendance actual result information DB130 stores attendance actual result information obtained from information providing server 40. Weather information DB140 stores weather information, including observation results and forecasts, obtained from information providing server 40.

[0054] The learned model 150 is a learned model that has undergone machine learning to process vehicle information, actual attendance results, and weather information. The learned model 150 uses the number of people attending work on the day of the prediction, the number of people leaving get off work in each time period, and weather information to predict congestion conditions for each time period. Here, the learned model 150 uses a random forest to predict congestion conditions. The random forest generates multiple groups of decision trees with low correlation, and derives the prediction result by merging and averaging their predictions. Control parameters for the random forest include the number of explanatory variables to be selected, the number of branches in the decision trees, and the thresholds set at the queries within the decision trees. Furthermore, the learned model 150 described in this embodiment uses a random forest. However, it is not limited to this. The learned model 150 can also be a neural network or a support vector machine.

[0055] Next, refer to Figure 5 This section explains the functions of the central server. As an example, it provides a detailed description. Figure 5As shown, in the central server 30 of this embodiment, the CPU 30A, which is a processor, performs functions as an acquisition unit 300, a judgment unit 310, a setting unit 320, a learning unit 330, a storage unit 340, an inference unit 350, a prediction unit 360, and a prompting unit 370 by executing the prediction program 110.

[0056] The acquisition unit 300 has the function of acquiring vehicle information sent from the vehicle-mounted unit 20 of the vehicle 12, as well as weather information and attendance results information sent from the information providing server 40.

[0057] The judgment unit 310 uses the acquired vehicle information to determine the presence or absence of congestion on roads surrounding the facility. Specifically, the judgment unit 310 divides the area into sections from intersections surrounding the facility at predetermined distances (e.g., 50m), and uses the vehicle information to calculate the required time for a vehicle 12 to pass through each section. For each time period, the judgment unit 310 calculates the average required time (hereinafter referred to as "average required time") for vehicles 12 passing through each section, and detects the number of consecutive sections where the average required time exceeds a threshold. If, within each time period, the judgment unit 310 detects a number of consecutive sections exceeding a predetermined threshold (meaning the average required time exceeds a predetermined distance), it determines that congestion has occurred.

[0058] As an example Figure 6 The graph illustrates each time period and the average time required within a predefined interval. For example, as... Figure 6 As shown, the judgment unit 310 derives the average required time for each interval according to each time period. The judgment unit 310 detects intervals with an average required time of 4.0 seconds or more. Here, in Figure 6 In the diagram, the shaded areas represent the regions detected by the judgment unit.

[0059] If, within each time period, the judgment unit 310 detects eight or more consecutive intervals (400m), it determines that congestion has occurred within that time period. For example, if... Figure 6 As shown, since the judgment unit 310 detects that the average time required is more than 4.0 seconds in the A to I intervals within the time period e (17:00 to 17:15), and the detected intervals are nine consecutive intervals, it is determined that congestion occurred within the time period e.

[0060] Additionally, as an example, for Figure 6The base point (0m location) for setting the interval shown is, for example, an intersection. However, it is not limited to this. Any location with factors that could cause congestion can be set as the base point (0m location). For example, it could be a location where the number of lanes or the road width is reduced, a location under construction, a location where an accident has occurred, or a location where acceleration is reduced, such as an uphill slope.

[0061] Figure 5 The setting unit 320 shown uses the congestion situation, weather information, and actual attendance results determined by the judgment unit 310 as an example for setting. Figure 7 The learning data shown.

[0062] like Figure 7 As shown, the learning data is set to include weather information, actual attendance results, and judgment results. The weather information includes date, time period, calendar, weather conditions, and precipitation. The actual attendance results include employee attendance rate, number of employees leaving get off work, number of employees leaving work one period ago, number of employees leaving get off work two periods ago, and number of employees leaving get off work three periods ago. The judgment results include congestion conditions. Here, date refers to the day the data was obtained from the object, time period refers to the time period the data was obtained from the object, and calendar information represents the characteristics of the object's date, such as ten-day periods, weekdays, and holidays. Furthermore, weather information refers to the weather conditions on the object's date, and precipitation refers to the amount of rainfall on the object's date. Additionally, attendance rate is the ratio of the number of employees who attended work on the object's date to the total number of employees in the facility, and number of employees leaving get off work is the number of employees who left the facility in each time period. Furthermore, in the case of time period d, the number of people leaving get off work one period ago is, for example, the number of people leaving get off work in time period c; the number of people leaving work two periods ago is, for example, the number of people leaving get off work two periods ago in time period b; and the number of people leaving get off work three periods ago is, for example, the number of people leaving work three periods ago in time period a. In other words, the setting unit 320 sets the number of people leaving get off work for a period as "the number of people leaving work one period ago" one period later, "the number of people leaving get off work two periods ago" two periods later, and "the number of people leaving get off work three periods ago" three periods later, respectively, as learning data. Thus, the learning data includes the number of people leaving work in previous time periods compared to the time period in which congestion prediction is implemented, thereby making it possible to predict congestion conditions considering the time required from when employees leave work until they reach the road to the target of prediction.

[0063] Furthermore, the setting unit 320 uses weather information and the inference results derived by the inference unit 35 (described later) to set the setting data for implementing congestion prediction. As an example... Figure 8As shown, the setting data is configured to include weather information and inference results. The weather information includes date, time period, calendar, weather conditions, and precipitation, while the inference results include the inference unit 350, which will describe this later, the inferred employee attendance rate, the inferred number of employees leaving get off work, the number of employees leaving get off work one period ago, the number of employees leaving get off work two periods ago, and the number of employees leaving get off work three periods ago. Here, the inferred number of employees leaving get off work in this embodiment is an example of "inferred employee information."

[0064] Figure 5 The learning unit 330 shown uses Figure 7 The learning data shown is used to perform machine learning for predicting congestion and generate a learned model 150. Specifically, the learning unit 330 uses weather information and actual attendance results included in the learning data as input data, and uses the input data included in the learning data and the congestion conditions that have been correlated with the input data as teaching data to perform machine learning.

[0065] The storage unit 340 stores the learned model 150 generated by the learning unit 330. In addition, the storage unit 340 stores the vehicle information, attendance result information and weather information obtained by the acquisition unit 300 into the vehicle information DB120, attendance result information DB130 and weather information DB140 respectively.

[0066] The inference unit 350 uses past attendance results information stored in the attendance results information DB130 to perform statistics and infer the number of employees who will be present at the facility on the predicted date. Specifically, the inference unit 350 uses past attendance results information and, based on the average of the moving average of the number of employees who attended on the same day of the week over the past four weeks (e.g., the number of employees who attended on Wednesdays), derives the number of employees present on the predicted date (hereinafter referred to as the "inferred number"). The inference unit 350 uses the derived inferred number to infer the proportion of the inferred number to all employees (hereinafter referred to as the "inferred attendance rate").

[0067] Furthermore, as an example, such as Figure 9 As shown, the inference unit 350 uses past attendance data to perform statistics and derives the ratio of the number of employees who left work to the number of employees who attended get off work, for each date and time period. Using the derived ratio of employees who left work and the inferred number of employees, the inference unit 350 infers the number of employees who left work in each time period on the predicted date (hereinafter referred to as the "inferred number of employees who left work").

[0068] Prediction Department 360 uses Figure 8 The set data shown is used to predict the congestion situation for each time period on the target day. Specifically, the prediction unit 360 predicts the congestion situation for each time period on the target day by inputting the set data into the learned model 150.

[0069] The prompting unit 370 provides a prompt regarding the prediction result obtained by the prediction unit 360. Specifically, the prompting unit 370 provides a prompt regarding the prediction result obtained by the prediction unit 360. Figure 10 The prediction prompt screen 400 shown provides a prompt. As an example, Figure 10 The prediction prompt screen 400 shown includes a target road prompt area 410, a prediction result prompt area 420, and an observation result prompt area 430. The target road prompt area 410 uses a map or similar tool to indicate the roads for which congestion is predicted. The prediction result prompt area 420 displays the prediction results for each time period predicted by the prediction unit 360. The observation result prompt area 430 displays the congestion assessment results for each time period on the same day of the week (e.g., Wednesday) one week ago, two weeks ago, and three weeks ago.

[0070] Before explaining the function of the congestion prediction system 10, refer to Figure 11 as well as Figure 12 This section explains the flow of data in the learning phase of generating the learned model 150 in the central server 30, which serves as a congestion prediction device, and in the prediction phase of predicting congestion. Figure 11 This is a data flow diagram representing an example of the flow of data generated in the learning phase of the learned model 150 in the central server 30.

[0071] First, refer to Figure 11 This is used to explain the flow of data during the learning phase. For example... Figure 11 As shown, the acquisition unit 300 acquires vehicle information, attendance result information, and weather information from vehicle information DB120, attendance result information DB130, and weather information DB140. The acquisition unit 300 inputs the vehicle information into the judgment unit 310, and inputs the attendance result information and weather information into the setting unit 320.

[0072] like Figure 6 As shown, the judgment unit 310 uses vehicle information to detect the average required time for each preset interval and each time period, and uses the judgment result 500 to judge the congestion situation. The judgment unit 310 inputs the judged congestion situation into the setting unit 320.

[0073] The setting unit 320 uses the input attendance results, weather information, and traffic congestion information to set the learning data 510 and inputs it into the learning unit 330. Here, the setting unit 320 uses the number of people leaving get off work included in the attendance results information and sets the number of people leaving get off work one period ago, two periods ago, and three periods ago as the learning data 510. In addition, the setting unit 320 sets the learning data 510 by dividing it into training data and test data.

[0074] The learning unit 330 uses the input learning data 510 to perform machine learning and generates a learned model 150, which is then input into the storage unit 340. Here, the learning unit 330 uses training data from the learning data 510 to generate the learned model 150 and uses test data to evaluate the generated learned model 150.

[0075] The storage unit 340 stores the input learned model 150 into the storage unit 30D.

[0076] Next, refer to Figure 12 This will explain the data flow during the forecasting phase. For example... Figure 12 As shown, the acquisition unit 300 acquires the attendance actual result information and weather information from the attendance actual result information DB130 and the weather information DB140. The acquisition unit 300 inputs the attendance actual result information into the inference unit 350 and inputs the weather information into the setting unit 320.

[0077] The inference unit 350 uses the input actual attendance results information to perform statistics and infers the inferred attendance rate for the predicted day and the inferred number of people leaving get off work for each time period. As the inference result 530, the inference unit 350 inputs the inferred attendance rate and the inferred number of people leaving get off work into the setting unit 320.

[0078] The setting unit 320 uses the input estimated attendance rate, estimated number of employees leaving get off work, and weather information to set the setting data 540 and input it into the prediction unit 360. Here, the setting unit 320 uses the estimated number of employees leaving get off work and sets the number of employees leaving get off work one period ago, two periods ago, and three periods ago as the setting data 540.

[0079] The prediction unit 360 uses the learned model 150 stored in the storage 30D to predict the congestion situation based on the input setting data 540, and inputs the prediction result 550 into the prompt unit 370.

[0080] The prompting unit 370 uses the input prediction result 550 to display a prompt on the prediction prompt screen 400. Here, the prompting unit 370 obtains vehicle information and displays congestion information for each time period of the same day of the week along with the prediction result 550. Furthermore, when the prompting unit 370 receives an instruction from the user to display a prompt based on the prediction result 550, it sends the prediction prompt screen 400 to the user's terminal as the prediction result 550 and displays the prompt.

[0081] (Control process) use Figure 13 as well as Figure 14 The flowchart below illustrates the process executed by the congestion prediction system 10 of this embodiment. Each process in the central server 30 is executed by the CPU 30A of the central server 30, which functions as an acquisition unit 300, a judgment unit 310, a setting unit 320, a learning unit 330, a storage unit 340, an inference unit 350, a prediction unit 360, and a prompting unit 370. Figure 13 The generation process shown is performed, for example, when an instruction is input to perform the process of generating the learned model 150.

[0082] In step S101, CPU30A obtains vehicle information, actual attendance results, and weather information.

[0083] In step S102, CPU30A uses vehicle information to determine the congestion situation.

[0084] In step S103, CPU30A uses the actual attendance results, weather information, and traffic congestion status to set the learning data.

[0085] In step S104, CPU30A uses the training data from the learning data to perform machine learning.

[0086] In step S105, as a result of the machine learning process, CPU30A generates a learned model 150.

[0087] In step S106, CPU30A uses test data from the learning data to evaluate the generated learned model 150. Here, as an evaluation result, the matching rate between the congestion prediction achieved by the learned model 150 and the congestion conditions included in the test data is derived.

[0088] In step S107, CPU 30A determines whether the matching rate derived as the evaluation result exceeds a threshold. If the matching rate exceeds the threshold (Yes in step S107), CPU 30A proceeds to step S108. On the other hand, if the matching rate does not exceed the threshold (the matching rate is below the threshold) (No in step S107), CPU 30A proceeds to step S104 and performs machine learning again.

[0089] In step S108, CPU 30A stores the generated learned model 150 into storage 30D.

[0090] Next, refer to Figure 14 The process of congestion prediction executed in the central server 30 of this embodiment will be described.

[0091] In step S201, CPU30A obtains the actual attendance results and weather information.

[0092] In step S202, CPU30A uses the obtained actual attendance results to infer the attendance rate and the number of people leaving get off work, as the inference results.

[0093] In step S203, CPU30A uses the inference results and weather information to set the set data.

[0094] In step S204, CPU30A uses set data to predict the congestion situation as the prediction result.

[0095] In step S205, CPU 30A determines whether congestion conditions for all time periods have been predicted on the target date. If congestion conditions for all time periods have been predicted (yes in step S205), CPU 30A proceeds to step S206. On the other hand, if congestion conditions for all time periods have not been predicted (there are time periods for which congestion conditions have not been predicted) (no in step S205), CPU 30A proceeds to step S204 and predicts congestion conditions for the time periods for which no predictions were made.

[0096] In step S206, as a prediction result, the CPU30A displays the predicted congestion situation on the prediction prompt screen 400.

[0097] In step S207, CPU 30A determines whether to terminate the prediction process. If the process is terminated (Yes in step S207), CPU 30A terminates the prediction process. On the other hand, if the process is not terminated (No in step S207), CPU 30A proceeds to step S201 and obtains the actual attendance results and weather information.

[0098] In this embodiment, the central server 30 determines the congestion situation of roads surrounding the facility based on vehicle information obtained from vehicle 12, and generates a learned model 150 that incorporates machine learning based on weather information, actual attendance results, and congestion conditions. The central server 30 infers the inferred attendance rate for the predicted day and the inferred number of people leaving work for each time period based on the actual attendance results, and uses the learned model 150 to predict the congestion situation for each time period on the predicted day.

[0099] Based on this embodiment, it is possible to predict road conditions over a long period of time.

[0100] Furthermore, in the above embodiment, the method described is that the prompting unit 370 obtains vehicle information and displays the judgment result of the congestion situation determination, along with congestion information for each time period of the same day of the week in the past, together with the prediction result. However, it is not limited to this. For example, the prompting unit 370 may obtain the judgment result determined by the judgment unit 310 and display it together with the prediction result.

[0101] Furthermore, in the above embodiments, the various processes executed by the CPU20A and CPU30A reading the software (program) can also be executed by various processors other than the CPU. Examples of processors in this case include FPGAs (Field-Programmable Gate Arrays) and PLDs (Programmable Logic Devices) whose circuit structure can be changed after manufacturing, as well as ASICs (Application-Specific Integrated Circuits) and other dedicated circuits that have circuit structures specifically designed for executing specific processes. Furthermore, the various processes described above can be executed by one of these various processors, or by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, and a combination of a CPU and an FPGA). More specifically, the hardware configuration of these various processors is a circuit composed of circuit elements such as semiconductor elements.

[0102] Furthermore, in the above embodiments, the method described is that each program is pre-stored (installed) in a computer-readable non-transitory storage medium. For example, the prediction program 110 in the central server 30 is pre-stored in the ROM 30B. However, this is not a limitation; each program may also be provided by storing it in a non-transitory storage medium such as a CD-ROM (Compact Disc Read Only Memory), DVD-ROM (Digital Versatile Disc Read Only Memory), or USB (Universal Serial Bus) memory. Additionally, the program may be configured to be downloaded from an external device via a network.

[0103] The processing flow described in the above embodiments is an example, and unnecessary steps can be deleted, new steps can be added, or the processing order can be replaced without departing from the main idea.

Claims

1. A congestion prediction device, comprising: As a non-temporary storage medium, memory stores programs that enable the processor to execute processes. processor, The processor performs the following processing: The system acquires personnel information representing the number of users who departed from the facility within each preset period, including users riding in vehicles; weather information for each preset period; and vehicle information related to vehicles passing through the vicinity of the facility. The vehicle information is used to determine congestion conditions, which indicate the presence or absence of congestion on roads surrounding the facility within a pre-set period. Machine learning is used to generate a learned model for predicting road congestion by employing teaching data, which includes the population information, the weather information, and the congestion conditions that are correlated with the population information and the weather information. The congestion situation on roads surrounding the facility is predicted by inputting estimated number information representing the number of users presumed to have departed from the facility for each pre-defined period, including users riding in vehicles, and the weather information. The estimated number of people during the period in which the congestion prediction is implemented is inferred. The inferred number of people and the weather information are input into the learned model to predict the congestion situation during the specified period. The prediction prompt screen, which displays the congestion situation during the predicted target period, displays the roads for which the congestion situation is predicted using a map, the prediction results for each preset period within the predicted target period, and the judgment results of the congestion situation for each preset period on the same day of the week in the past target period. The number of people includes the number of departures one period ago, the number of departures two periods ago, and the number of departures three periods ago—the number of users involved in the period preceding the period in which the congestion situation is determined.

2. The congestion prediction device as described in claim 1, wherein, The vehicle information includes location information related to the vehicle's position and speed information related to the vehicle's speed. The processor uses the location information and speed information to determine the congestion situation.

3. The congestion prediction device as described in claim 2, wherein, The processor, using the position information and the speed information, calculates the required time to traverse the pre-set interval. If a congestion condition is determined to be congested when the intervals exceeding the required time threshold exist continuously in a predetermined number of intervals.

4. The congestion prediction device as described in claim 1, wherein, The learned model is one that uses random forest.

5. The congestion prediction device as described in claim 1, wherein, The processor infers the estimated number of users during the target period using the number of past users who have accessed the facility.

6. The congestion prediction device as claimed in claim 1 or claim 5, wherein, The processor provides alerts regarding the congestion status during each inferred object period.

Images