Traffic volume prediction device, traffic volume prediction method, and program

Abstract

It is desirable to have technology that can accurately predict traffic flow even when sudden fluctuations in traffic volume occur due to natural congestion. [Solution] A traffic volume prediction device is provided, comprising: a learning data extraction unit that extracts first traffic state data corresponding to the first traffic congestion data as learning data based on the determination that the congestion has reached the target facility in first traffic congestion data which includes a first traffic congestion time which is the time when congestion occurs and a first traffic congestion location which is the location where congestion occurs at the first traffic congestion time; a learning unit that generates a model based on the learning data and first traffic volume data at the target facility; and a prediction unit that predicts second traffic volume data at the target facility based on the model and second traffic state data.

Description

[Technical Field]

[0001] This invention relates to a traffic volume prediction device, a traffic volume prediction method, and a program. [Background technology]

[0002] Generally, traffic volume forecasting methods employ either time-series or spatiotemporal forecasting techniques. Time-series forecasting methods predict traffic volume by analyzing the relationship between time and traffic volume. On the other hand, spatiotemporal forecasting methods predict traffic volume by analyzing the relationship between time, space, and traffic volume. For example, in spatiotemporal forecasting methods, spatiotemporal changes in traffic volume are analyzed using road links. A link is a section of road connecting multiple nodes, which are characteristic points of the road.

[0003] For example, a known method for predicting traffic volume over time is polynomial regression analysis using an autoregressive model. However, in the event of a sudden event (such as an accident or debris falling onto the road) or a large-scale traffic jam, more vehicles may choose detours at junctions located upstream of the congestion point to avoid it. In such cases, traffic volume near the congestion point may fluctuate suddenly.

[0004] For example, a branch road is an interchange, which connects a highway to a local road. Alternatively, a branch road is a junction, which is where multiple highways intersect or connect. Such sudden fluctuations in traffic volume are thought to have little dependence on time series. Therefore, there is a problem in that it is difficult to accurately predict traffic volume over time series using polynomial regression analysis with an autoregressive model.

[0005] To address these problems inherent in time-series traffic volume forecasting methods, a technique is known that employs different traffic volume forecasting methods depending on whether the congestion is caused by a concentration of traffic volume due to increased traffic demand (hereinafter also referred to as "natural congestion") or by an unexpected event. Furthermore, Patent Document 1 discloses a technique for predicting pedestrian flow in the event of a disaster, which is an example of an unexpected event, by considering not only changes in conditions inside the facility but also changes in conditions outside the facility.

[0006] In predicting traffic volume, as in these technologies, a common approach is to not consider sudden traffic volume fluctuations if it is determined that no sudden events have occurred, and to consider sudden traffic volume fluctuations if it is determined that a sudden event has occurred. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2024-8766 [Overview of the project] [Problems that the invention aims to solve]

[0008] However, sudden fluctuations in traffic volume can also be caused by changes in traffic phenomena. In other words, sudden fluctuations in traffic volume do not necessarily occur only when a sudden event occurs. For example, even when natural congestion occurs, if the scale of the congestion exceeds a certain level, vehicles upstream of the congestion point may choose detour routes, which can lead to sudden fluctuations in traffic volume. Therefore, when predicting traffic volume, considering sudden fluctuations in traffic volume only when it is determined that a sudden event has occurred may not improve the accuracy of traffic volume predictions.

[0009] Therefore, it is desirable to have technology that can accurately predict traffic volume even when sudden fluctuations in traffic volume occur due to natural congestion. [Means for solving the problem]

[0010] To solve the above problems, according to one aspect of the present invention, a traffic volume prediction device is provided, comprising: a learning data extraction unit that extracts first traffic state data corresponding to the first traffic congestion data as learning data based on the determination that the congestion has reached the target facility in first traffic congestion data which includes a first congestion time which is the time when congestion occurs and a first congestion location which is the location where congestion occurs at the first congestion time; a learning unit that generates a model based on the learning data and first traffic volume data at the target facility; and a prediction unit that predicts second traffic volume data at the target facility based on the model and second traffic state data.

[0011] The learning data extraction unit may determine whether or not the congestion has reached the target facility in the first congestion data if no predetermined event causing congestion has occurred at the first congestion time.

[0012] The learning data extraction unit may acquire information about the event as the first traffic state data.

[0013] The learning data extraction unit may calculate, based on the first traffic congestion data, at least one of the following as the first traffic condition data: the length of the congestion at the first congestion time, the position of the end of the congestion at the first congestion time, and the average value of the vehicle speed corresponding to the first congestion time.

[0014] The learning data extraction unit may calculate the difference in time required for a vehicle to travel from the target facility to a downstream facility, which is a facility located downstream of the target facility, at the first congestion time and at a time prior to the first congestion time, as the first traffic condition data.

[0015] The learning unit may generate the model by performing an analysis using the learning data as explanatory variables and the first traffic volume data as the dependent variable.

[0016] The learning data extraction unit may determine whether a statistical quantity of vehicle speed corresponding to a first position and a first time is below a threshold, and if it determines that the statistical quantity is below the threshold, it may extract the first position as the first congestion location and the first time as the first congestion time.

[0017] The learning data extraction unit may determine whether the traffic congestion has reached the target facility in the first traffic congestion data by determining whether the target facility is located at the first traffic congestion location.

[0018] The traffic volume prediction device may include a traffic volume calculation unit that calculates a first difference between the amount of traffic flowing into the target facility and the amount of traffic flowing out of the target facility during the first congestion time as the first traffic volume data.

[0019] The learning data extraction unit may determine whether the congestion has reached multiple facilities, including the target facility, in the first congestion data, based on the determination that the congestion has reached the target facility in the first congestion data, or that a predetermined event causing congestion has occurred at the first congestion time. If it determines that the congestion has reached multiple facilities in the first congestion data, it may determine whether there is an upstream time when the target facility is located at the upstream end of the multiple facilities. If it determines that there is an upstream time, it may extract the first traffic state data corresponding to the upstream time and the target facility as the learning data.

[0020] The prediction unit may also include a determination unit that determines whether the congestion has reached the target facility in second congestion data which includes a second congestion time which is the time when congestion occurs and a second congestion location which is the location where congestion occurs at the second congestion time, and extracts second traffic state data corresponding to the second congestion data based on the determination that the congestion has reached the target facility in the second congestion data.

[0021] The prediction unit may include: a traffic volume prediction unit that predicts the amount of traffic flowing out of the target facility at a time after the second congestion time, based on the amount of traffic flowing out of the target facility at the second congestion time, which is the time when congestion occurs; a traffic volume fluctuation prediction unit that predicts a second difference between the amount of traffic flowing into the target facility at the later time and the amount of traffic flowing out of the target facility at the later time, based on the model and the second traffic state data; and a traffic volume correction unit that predicts the second traffic volume data by correcting the amount of traffic flowing out of the target facility at the later time based on the second difference.

[0022] The traffic volume correction unit may correct the outflow traffic volume from the target facility at a later time by subtracting a correction amount corresponding to the second difference from the outflow traffic volume from the target facility at a later time.

[0023] The traffic volume correction unit may increase the correction amount as the second difference obtained by subtracting the outflow traffic volume from the target facility during the second congestion time from the inflow traffic volume to the target facility during the second congestion time is larger.

[0024] Furthermore, in order to solve the above problems, according to another aspect of the present invention, a computer-based traffic volume prediction method is provided, which includes: extracting first traffic state data corresponding to the first traffic congestion data as training data based on the determination that the congestion has reached the target facility in first traffic congestion data which includes a first congestion time which is the time when congestion occurs and a first congestion location which is the location where congestion occurs at the first congestion time; generating a model based on the training data and first traffic volume data at the target facility; and predicting second traffic volume data at the target facility based on the model and second traffic state data.

[0025] Furthermore, in order to solve the above problems, according to another aspect of the present invention, a program is provided that causes a computer to function as: a learning data extraction unit that extracts first traffic state data corresponding to the first traffic congestion data as learning data based on the determination that the congestion has reached the target facility in first traffic congestion data which includes a first congestion time which is the time when congestion occurs and a first congestion location which is the location where the congestion occurs at the first congestion time; a learning unit that generates a model based on the learning data and first traffic volume data at the target facility; and a prediction unit that predicts second traffic volume data at the target facility based on the model and second traffic state data. [Effects of the Invention]

[0026] As described above, the present invention provides a technology that enables accurate prediction of traffic volume even when natural congestion occurs. [Brief explanation of the drawing]

[0027] [Figure 1] This figure shows an example of the functional configuration of a traffic volume prediction device 1 according to an embodiment of the present invention. [Figure 2] This figure shows a detailed configuration example of the learning model generation unit 140. [Figure 3] This figure shows a detailed configuration example of the correction processing unit 152. [Figure 4] This figure shows an example of master data stored by the master data storage unit 126. [Figure 5] This figure shows an example of free-flow data stored by the free-flow data storage unit 123. [Figure 6] This figure shows the traffic volume entering and leaving Facility IC1 to date. [Figure 7] This figure shows an example of probe data stored by the probe data storage unit 122. [Figure 8] This diagram schematically shows the mesh data used for training. [Figure 9] This figure shows the future outflow traffic volume from facility IC1, as predicted by the traffic volume prediction unit 151. [Figure 10] This is a diagram showing a magnified portion of the mesh data. [Figure 11] This is a diagram to explain the extraction of traffic congestion data. [Figure 12] This is a diagram showing traffic congestion data D1. [Figure 13] This flowchart shows an example of the overall operation of the traffic volume prediction device 1 according to an embodiment of the present invention. [Figure 14] This flowchart shows the details of the learning model generation process. [Figure 15] This is a flowchart (first half) showing the details of the training data extraction process. [Figure 16] This is a flowchart (second half) showing the details of the training data extraction process. [Figure 17] This flowchart shows the details of the correction process. [Figure 18] This is a flowchart (first half) showing the details of the process for determining whether correction is necessary. [Figure 19] This is the flowchart (second half) showing the details of the process for determining whether correction is necessary. [Figure 20] This graph (Day 1) shows the relationship between actual traffic volume, predicted traffic volume (uncorrected), and predicted traffic volume (corrected) and time. [Figure 21] This graph (Day 2) shows the relationship between actual traffic volume, predicted traffic volume (uncorrected), and predicted traffic volume (corrected) and time. [Figure 22] This graph (Day 3) shows the relationship between actual traffic volume, predicted traffic volume (uncorrected), and predicted traffic volume (corrected) and time. [Figure 23] This graph (Day 4) shows the relationship between actual traffic volume, predicted traffic volume (uncorrected), and predicted traffic volume (corrected) and time. [Figure 24] This figure shows the hardware configuration of an information processing device 900 as an example of a traffic volume prediction device 1 according to an embodiment of the present invention. [Modes for carrying out the invention]

[0028] Preferred embodiments of the present invention will be described in detail below with reference to the attached drawings. In this specification and drawings, components having substantially the same functional configuration are denoted by the same reference numerals, and redundant descriptions will be omitted.

[0029] (1. Details of the Embodiment) The following describes in detail the embodiments of the present invention.

[0030] (1-1. Configuration of the traffic volume prediction device) First, an example of the configuration of the traffic volume prediction device 1 according to an embodiment of the present invention will be described. Figure 1 is a diagram showing an example of the functional configuration of the traffic volume prediction device 1 according to an embodiment of the present invention. The traffic volume prediction device 1 is a device that predicts traffic flow. Traffic flow is data related to traffic and may be a concept that includes, for example, traffic volume, traffic density, and vehicle speed. The traffic volume prediction device 1 may be an example of an information processing device.

[0031] Referring to Figure 1, vehicles M1 to M3 are shown as examples of vehicles traveling on a road. Furthermore, referring to Figure 1, vehicles M1 and M2 are traveling in the far lane (vehicle M1 is following vehicle M2), and vehicle M3 is traveling in the near lane, in the opposite direction to vehicles M1 and M2 in the far lane. Thus, the embodiment of the present invention mainly assumes a road composed of multiple lanes, but the road may also consist of a single lane. Each of vehicles M1 to M3 is equipped with an on-board device. Note that the term "lane" can also be used as "lane."

[0032] The traffic volume prediction device 1 includes an event information input unit 160. The traffic volume prediction device 1 also includes an event information storage unit 120, a driving history data storage unit 121, a probe data storage unit 122, a free flow data storage unit 123, a traffic volume data storage unit 124, a master data storage unit 126, a learning model storage unit 127, a prediction result storage unit 128, and a mesh data storage unit 129.

[0033] Furthermore, the traffic volume prediction device 1 includes a statistical processing unit 131, a traffic volume calculation unit 133, a mesh data generation unit 135, a learning model generation unit 140, and a prediction unit 150. The prediction unit 150 includes a traffic volume prediction unit 151 and a correction processing unit 152.

[0034] Figure 2 shows a detailed configuration example of the learning model generation unit 140. As shown in Figure 2, the learning model generation unit 140 comprises a learning data extraction unit 141 and a traffic volume fluctuation learning unit 142. These blocks of the learning model generation unit 140 mainly operate during the learning phase. The detailed functions of these blocks of the learning model generation unit 140 will be described later.

[0035] Figure 3 shows a detailed configuration example of the correction processing unit 152. As shown in Figure 3, the correction processing unit 152 comprises a determination unit 1521, a traffic volume fluctuation prediction unit 1522, and a traffic volume correction unit 1523. These blocks of the correction processing unit 152 mainly operate during the prediction stage. The detailed functions of these blocks of the correction processing unit 152 will be described later.

[0036] A probe antenna 112 is connected to the driving history data storage unit 121, and a free-flow antenna 114 is connected to the free-flow data storage unit 123.

[0037] The statistical processing unit 131, the traffic volume calculation unit 133, the mesh data generation unit 135, the learning model generation unit 140, and the prediction unit 150 all include a computing device such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit), and their functions can be realized by a program stored in ROM (Read Only Memory) being loaded into RAM by the computing device and executed. In this case, a computer-readable recording medium on which the program is stored may also be provided.

[0038] Alternatively, the statistical processing unit 131, the traffic volume calculation unit 133, the mesh data generation unit 135, the learning model generation unit 140, and the prediction unit 150 may be composed of dedicated hardware or a combination of multiple hardware components. The data necessary for calculations by the computing unit is appropriately stored in a storage unit (not shown).

[0039] The event information storage unit 120, the driving history data storage unit 121, the probe data storage unit 122, the free flow data storage unit 123, the traffic volume data storage unit 124, the master data storage unit 126, the learning model storage unit 127, the prediction result storage unit 128, and the mesh data storage unit 129 are implemented by storage units (not shown). Such storage units may consist of memory such as RAM (Random Access Memory), a hard disk drive, or flash memory.

[0040] (Master data storage unit 126) The master data storage unit 126 stores various types of master data in advance. Master data is data that is input to the traffic volume prediction device 1 by the user and managed by the traffic volume prediction device 1. Here, an example of master data will be explained with reference to Figure 4.

[0041] Figure 4 shows an example of master data stored by the master data storage unit 126. As shown in Figure 4, the master data storage unit 126 stores the driving history time width 611, driving history section width 612, simulation execution time interval 621, simulation execution cell width 622, and congestion flow determination speed 641 as examples of master data. Details of this master data will be explained later.

[0042] (Event information input section 160) The event information input unit 160 receives event information from the user, including the time and location of a predetermined sudden event (hereinafter also referred to as "event") that causes traffic congestion on the road. The type of event is not limited. For example, an event may be any sudden event that has the potential to cause traffic congestion at a point on the road, and may include a traffic accident, traffic restrictions, the presence of an obstacle on the road, or planned construction work.

[0043] For example, the event information input unit 160 may be configured with an input device such as a mouse or keyboard. However, the event information input unit 160 may be configured with other input devices. For example, the event information input unit 160 may be configured with a touch panel or buttons.

[0044] For example, let's assume that the user inputs the starting point of the road, "Point A," and the distance from "Point A," "24km," as the event location. Let's also assume that the simulation execution cell width of 622 is "200m." A cell corresponds to each block when the road is logically divided into multiple blocks at equal intervals, and the simulation execution cell width of 622 is the width of one cell. In this case, the 120th cell from the starting point of the road, "Point A" (24km / 200m =) is designated as the event location.

[0045] Event information received from the user by the event information input unit 160 is output to the event information storage unit 120.

[0046] The event information input unit 160 may acquire event information from a predetermined storage location (for example, a web page containing event information). Alternatively, the event information input unit 160 may estimate whether an event has occurred, time by time and cell by cell, based on the traffic volume data stored in the traffic volume data storage unit 124 and the probe data stored in the probe data storage unit 122.

[0047] (Event information storage unit 120) The event information storage unit 120 stores the event information output from the event information input unit 160.

[0048] (Free-flow antenna 114) The free-flow antenna 114 functions as an example of a vehicle detection unit that detects vehicles traveling at various locations on the road. In other words, the embodiments of the present invention primarily assume that the vehicle detection unit includes the free-flow antenna 114. This allows the ETC free-flow antenna of an already constructed ETC (Electronic Toll Collection) system to be used as the vehicle detection unit, eliminating the need to provide a new vehicle detection unit. However, other vehicle detection units (e.g., vehicle detectors, infrared sensors, or ultrasonic sensors) may be used instead of the free-flow antenna 114.

[0049] More specifically, the free-flow antenna 114 detects vehicles in real time by receiving vehicle identification information (vehicle ID) from the vehicle through communication with an on-board unit installed in the vehicle. In the embodiments of the present invention, the case where an ETC on-board unit is used as an example of an on-board unit is mainly assumed. It is assumed that the free-flow antenna 114 is compatible with multiple versions of ETC on-board units, while the probe antenna 112, which will be described later, is assumed to be compatible with only a specific version of ETC on-board unit. In other words, the free-flow antenna 114 can detect more vehicles than the probe antenna 112.

[0050] The "locations" on the road where a vehicle is detected by the free-flow antenna 114 are not particularly limited, as long as they are locations where a vehicle can be detected by the vehicle detection unit. Hereinafter, the locations on the road where a vehicle can be detected by the free-flow antenna 114 may simply be referred to as "vehicle detection locations."

[0051] The free-flow antenna 114 detects a vehicle in real time by receiving a vehicle ID through communication with an in-vehicle device mounted on the vehicle, and outputs the vehicle detection result (hereinafter also referred to as "free-flow data") to the free-flow data storage unit 123 in real time. The vehicle detection result output from the free-flow antenna 114 to the free-flow data storage unit 123 in real time can then be used in real time by the traffic volume calculation unit 133.

[0052] Here, "real time" can refer to any short period of time between the time a vehicle reaches the vehicle detection location and before the traffic conditions on the road change. If a vehicle is detected at this time, and the vehicle detection result is used by the traffic volume calculation unit 133 at this time, some action can be taken according to the traffic conditions at the time the vehicle was detected, before the traffic conditions on the road change.

[0053] (Free-flow data storage unit 123) Figure 5 shows an example of free-flow data stored by the free-flow data storage unit 123. The free-flow data corresponds to the detection result of a vehicle traveling at the vehicle detection location. As shown in Figure 5, the free-flow data is associated with a "vehicle ID" and a "passing time". The "vehicle ID" is the vehicle identification information received from the vehicle by the free-flow antenna 114 through communication between the free-flow antenna 114 and an in-vehicle device mounted on the vehicle. The "passing time" is the time when the free-flow antenna 114 received the vehicle ID from the vehicle, and may correspond to the time when the vehicle passed the vehicle detection location.

[0054] (Traffic volume calculation unit 133) The traffic volume calculation unit 133 acquires free flow data from the free flow data storage unit 123. Then, based on the free flow data (vehicle ID and passing time), the traffic volume calculation unit 133 calculates the number of vehicles that passed the vehicle detection location per unit time (in other words, the number of vehicle IDs detected per unit time by the free flow antenna 114) as the traffic volume at the vehicle detection location. The traffic volume calculation by the traffic volume calculation unit 133 can be repeated every unit time.

[0055] It should be noted that not all vehicles passing through a vehicle detection location are equipped with an on-board device capable of communicating with the free-flow antenna 114. Therefore, not all vehicles passing through a vehicle detection location are detected by the free-flow antenna 114. Accordingly, if the ratio of vehicles equipped with an on-board device capable of communicating with the free-flow antenna 114 to the total number of vehicles (including vehicles not equipped with such a device) is set in advance as the "on-board device installation ratio," and the number of vehicles detected by the free-flow antenna 114 is defined as the "number of detected vehicles," then it is desirable for the traffic volume calculation unit 133 to estimate the traffic volume at the vehicle detection location with greater accuracy using the following formula (1).

[0056] (Estimated traffic volume) = (Number of detected vehicles) ÷ (Percentage of vehicles equipped with onboard devices) ... (1)

[0057] However, such estimation may be omitted if in-vehicle devices capable of communicating with the free-flow antenna 114 are widely available. As described above, the traffic volume calculation unit 133 can obtain traffic volume data in which the estimated traffic volume, the measurement time which is the end of the time period corresponding to that traffic volume, and the vehicle detection position corresponding to that traffic volume are associated.

[0058] As will be explained later, in the embodiments of the present invention, the volume of traffic flowing out of a facility (hereinafter also referred to as "facility outflow traffic") is predicted. Here, the facility is a place where traffic flows in and out. Specifically, when the prediction target is the traffic flow of an expressway, the facility may be an interchange or junction, etc. Alternatively, when the prediction target is the traffic flow of a general road, the facility may be a commercial facility or a road junction, etc.

[0059] Furthermore, as will be explained later, the prediction of facility outflow traffic volume uses the difference between the traffic volume flowing into the facility to date (hereinafter also referred to as "facility inflow traffic volume") and the traffic volume flowing out of the facility (hereinafter also referred to as "facility difference traffic volume"). Therefore, the traffic volume calculation unit 133 obtains the facility inflow traffic volume and facility outflow traffic volume to date from the traffic volume data, and calculates the facility difference traffic volume based on the facility inflow traffic volume and facility outflow traffic volume to date.

[0060] In the following explanation, we will primarily assume that the traffic volume calculation unit 133 obtains the traffic volume entering and leaving the facility up to the present from traffic volume data. However, the traffic volume calculation unit 133 may obtain the traffic volume entering and leaving the facility up to the present in any way. For example, it is conceivable that cameras are installed near the facility. In such a case, the traffic volume calculation unit 133 may obtain the traffic volume entering and leaving the facility based on the number of vehicles recognized from the images obtained by the cameras.

[0061] Furthermore, in the following explanation, we may specifically consider examples where the traffic volume flowing out of X facilities (where X is an integer greater than or equal to 1) is predicted. We may also assume that each of the X facilities is an interchange, and these X facilities may be referred to as facility IC1, IC2, ..., and ICX. Additionally, any facility may be referred to as facility ICj (where j is any integer satisfying 1 ≤ j ≤ X).

[0062] In the following explanation, we primarily assume that facilities IC1, IC2, ..., ICX are located at equal intervals (every 20 cells) from upstream to downstream. However, the intervals between facilities IC1, IC2, ..., ICX do not have to be equal. Also, the intervals between facilities IC1, IC2, ..., ICX do not have to be 20 cells. If j is an integer greater than or equal to 2, facility ICj-1 is located one position upstream of facility ICj. Similarly, if j is an integer less than or equal to X-1, facility ICj+1 is located one position downstream of facility ICj.

[0063] Figure 6 shows the traffic volume entering and leaving facility IC1 up to the present. Referring to Figure 6, cells 1 to n are shown from the upstream side to the downstream side of the road. The upstream end of cell 1 is the starting point of the road, and the downstream end of cell n is the ending point of the road. i is any integer satisfying 1 ≤ i ≤ n. Δt is the time interval corresponding to the cell (simulation execution time interval 621). h is an integer greater than or equal to 1.

[0064] Referring to Figure 6, the current traffic volume entering facility IC1 is the traffic volume Q of facility IC1 corresponding to time t-hΔt. IC1_in From (t-hΔt), the inflow traffic volume Q of facility IC1 corresponding to time t (current time) IC1_in (t) is shown. The traffic volume calculation unit 133 calculates the facility inflow traffic volume Q IC1_in (t-hΔt)~Q IC1_in Get (t).

[0065] Furthermore, referring to Figure 6, the current outflow traffic volume at facility IC1 is shown as the inflow traffic volume Q at facility IC1 corresponding to time t-hΔt. IC1_out From (t-hΔt), the outflow traffic volume Q of facility IC1 corresponding to time t (current time) IC1_out (t) is shown. The traffic volume calculation unit 133 calculates the facility outflow traffic volume Q IC1_out (t-hΔt)~Q IC1_out Get (t).

[0066] The traffic volume calculation unit 133 calculates the facility differential traffic volume R IC1_in (t) based on the facility inflow traffic volume Q IC1_out (t) and the facility outflow traffic volume Q IC1 (t). More specifically, the traffic volume calculation unit 133 subtracts the facility outflow traffic volume Q IC1 (t) _in from the facility inflow traffic volume Q IC1_out (t) to calculate Q IC1_in (t) - Q IC1_out (t) as the facility differential traffic volume R IC1 (t). Similarly for other times, the traffic volume calculation unit 133 calculates the facility differential traffic volume R IC1 (t - Δt) to R IC1 (t - hΔt). Furthermore, similarly for other facilities, the traffic volume calculation unit 133 calculates the facility differential traffic volume R IC2 (t) to R IC2 (t - hΔt), ···, R ICX (t) to R ICX (t - hΔt).

[0067] The traffic volume calculation unit 133 outputs the facility outflow traffic volume Q IC1_out (t) to Q IC1_out (t - hΔt), ···, Q ICX_out (t) to Q ICX_out (t - hΔt) of the facility IC1 to ICX to the traffic volume data storage unit 124. Furthermore, the traffic volume calculation unit 133 outputs the facility differential traffic volume R IC1 (t) to R IC1 (t - hΔt), ···, R ICX (t) to R ICX (t - hΔt) of the facility IC1 to ICX to the traffic volume data storage unit 124.

[0068] (Traffic volume data storage unit 124) The traffic volume data storage unit 124 stores the facility outflow traffic volume Q IC1_out (t) to Q IC1_out (t - hΔt), ···, Q ICX_out (t) to Q ICX_out(t-hΔt) is stored. Furthermore, the traffic volume data storage unit 124 stores the facility difference traffic volume R of facilities IC1 to ICX, which is output from the traffic volume calculation unit 133. IC1 (t)~R IC1 (t-hΔt), ..., R ICX (t)~R ICX (t-hΔt) is stored. Note that facility outflow traffic volume and facility difference traffic volume may be included in the traffic volume data.

[0069] (Probe antenna 112) The probe antenna 112 functions as an example of a driving history data acquisition unit that acquires vehicle driving history data. That is, in the embodiments of the present invention, it is mainly assumed that the driving history data acquisition unit includes the probe antenna 112. As a result, the ETC probe antenna of an already constructed ETC system can be used as the driving history data acquisition unit, so there is no need to set up a new driving history data acquisition unit. However, other driving history data acquisition units (for example, a mobile phone base station) may be used instead of the probe antenna 112.

[0070] More specifically, when the probe antenna 112 acquires driving history data from the vehicle through communication with an in-vehicle device mounted on the vehicle, it outputs the acquired driving history data to the driving history data storage unit 121.

[0071] (Driving history data storage unit 121) The driving history data storage unit 121 stores the driving history data output from the probe antenna 112. The driving history data stored in the driving history data storage unit 121 can be used by the statistical processing unit 131.

[0072] The driving history data includes the speed of each vehicle that traveled for each section (between kilometer posts) on the road defined by a driving history section width of 612 (e.g., 100m), and for each driving history time width of 611 (e.g., 1 minute), as well as the collection time, which is the time when the speed was collected by the probe antenna 112.

[0073] (Statistical Processing Unit 131) The statistical processing unit 131 acquires driving history data from the driving history data storage unit 121, performs statistical processing on the driving history data, and outputs the statistically processed driving history data as probe data to the probe data storage unit 122. The probe data output from the statistical processing unit 131 to the probe data storage unit 122 can then be mainly used by the mesh data generation unit 135, the learning model generation unit 140, and the prediction unit 150.

[0074] For example, the statistical processing unit 131 applies a predetermined statistical process (e.g., averaging) to the speed of one or more vehicles that have traveled for each section (between kilometer posts) on the road and for each travel history time interval 611. This provides statistical data of vehicle speed for each travel history time interval 611 in each section of the road. An example of averaging is taking the harmonic mean.

[0075] (Probe data storage unit 122) Figure 7 shows an example of probe data stored by the probe data storage unit 122. As shown in Figure 7, the probe data is associated with "time of passage", "kilometer post", "speed", and "collection time".

[0076] "Passing time" is the end time of each 611-second interval in the travel history timeline. For example, "Passing time: 2019 / 01 / 01 / 10:24" is the end time of the period from 10:23 AM to 10:24 AM on January 1, 2019, which is a 611-second interval in the travel history timeline. "Kilometer post" is a distance marker from the starting point of the road.

[0077] "Speed" is a statistical value (e.g., average speed) of the speed of one or more vehicles traveling through each section (between kilometer posts) on the road for every 611 time intervals (between passing times). For example, "Speed: 80 km / h" is a statistical value of the speed of one or more vehicles traveling from "Kilometer Post: 100" to "Kilometer Post: 100.1" between 10:23 AM on January 1, 2019 and 10:24 AM on January 1, 2019.

[0078] "Collection time" is the time when the "speed," which is a statistical measure of the vehicle's speed, was collected by the probe antenna 112.

[0079] (Mesh data generation unit 135) The mesh data generation unit 135 acquires probe data from the probe data storage unit 122. As described above, the probe data is composed of data associated with passage time, kilometer post, speed, and collection time. On the other hand, in the following learning process, data composed of multiple data (hereinafter also called "mesh") associated with time, cell, and speed (hereinafter also called "mesh data") is used. Therefore, the mesh data generation unit 135 generates mesh data based on the probe data.

[0080] Here, the time that constitutes the mesh may correspond to the first time. Also, each cell that constitutes the mesh may correspond to the first location. The width of the section on the road (between kilometer posts) is defined by the travel history section width of 612. On the other hand, the width of the cell is defined by the simulation execution cell width of 622.

[0081] The width of a cell may be the same as the width of an interval, or it may be different from the width of an interval. For example, if the width of an interval is 100m, the width of a cell may be 500m. In cases where the width of a cell is different from the width of an interval, as in this example, the mesh data generation unit 135 may identify multiple intervals belonging to the cell, perform statistical processing on the velocity corresponding to each of these intervals, and obtain the velocity after statistical processing as the velocity corresponding to the cell.

[0082] The time intervals corresponding to sections on the road are defined in the driving history time interval 611. On the other hand, the time intervals corresponding to cells are defined in the simulation execution time interval 621.

[0083] The simulation execution time interval 621 (the time interval corresponding to the cell) may be the same as the driving history time width 611 (the time interval corresponding to the section), or it may be different from the driving history time width 611 (the time interval corresponding to the section). For example, if the driving history time width 611 (the time interval corresponding to the section) is 1 minute, the simulation execution time interval 621 (the time interval corresponding to the cell) may be 5 minutes. In cases like this example, where the simulation execution time interval 621 (the time interval corresponding to the cell) is different from the driving history time width 611 (the time interval corresponding to the section), the mesh data generation unit 135 may identify the time corresponding to the section that belongs to the time corresponding to the cell, perform statistical processing on the speed corresponding to the identified time, and obtain the speed after statistical processing as the speed corresponding to the cell and time (the end time of the time corresponding to the cell).

[0084] Figure 8 schematically shows the mesh data used for training. Referring to Figure 8, the velocity V1(t) ~ V n (t) (km / hour) is shown. Below, we will mainly explain the case where time t is the current time, but time t is not limited to the current time. The velocity V1(t-hΔt)~V corresponding to time t-hΔt n From (t-hΔt), the velocity V1(t)~V corresponds to time t (current time). n Up to (t), it is mainly used by the learning model generation unit 140.

[0085] The mesh data generation unit 135 generates a speed V1(t-hΔt)~V corresponding to time t-hΔt. n From (t-hΔt), the velocity V1(t)~V corresponds to time t (current time). nThe data up to (t) is output as mesh data to the mesh data storage unit 129. For example, the mesh data generation unit 135 updates the time t (current time) at time intervals Δt, and the speed V1(t)~V corresponding to the updated time t (current time) n (t) may be output to the mesh data storage unit 129.

[0086] (Mesh data storage unit 129) The mesh data storage unit 129 stores the mesh data output from the mesh data generation unit 135.

[0087] (Traffic volume forecasting unit 151) The traffic volume prediction unit 151 receives the facility outflow traffic volume Q for facilities IC1 to ICX from the traffic volume data storage unit 124. IC1_out (t) ~ Q IC1_out (t-hΔt), ..., Q ICX_out (t) ~ Q ICX_out The (t-hΔt) is obtained. Then, the traffic volume prediction unit 151 calculates the current facility outflow traffic volume Q at facility IC1. IC1_out (t) ~ Q IC1+1_out Based on (t-hΔt), the future outflow traffic volume at facility IC1 is predicted. Similarly, the traffic volume prediction unit 151 predicts the future outflow traffic volume at facilities IC2 to ICX.

[0088] Furthermore, any prediction method can be used to predict future traffic flow out of facilities, such as time-series prediction methods like the Autoregressive Integrated Moving Average (ARIMA) model or the Seasonal Autoregressive Integrated Moving Average (SARIMA) model, or spatiotemporal prediction methods like graph neural networks (GNNs).

[0089] Figure 9 shows the future outflow traffic volume of facility IC1 predicted by the traffic volume prediction unit 151. In the example shown in Figure 9, M is an integer greater than or equal to 1. Referring to Figure 9, the future inflow traffic volume of facility IC1 is the outflow traffic volume of facility IC1 corresponding to time t+Δt Q. IC1_out From (t+Δt), the outflow traffic volume Q of facility IC1 corresponding to time t+MΔt is calculated. IC1_out The expression up to (t+MΔt) has been shown.

[0090] The traffic volume prediction unit 151 calculates the predicted facility outflow traffic volume Q for facilities IC1 to ICX. IC1_out (t+Δt)~Q IC1_out (t+MΔt), ..., Q ICX_out (t+Δt)~Q ICX_out (t+MΔt) is output to the prediction result storage unit 128.

[0091] (Prediction result storage unit 128) The prediction result storage unit 128 stores the facility outflow traffic volume Q for facilities IC1 to ICX, which is output from the traffic volume prediction unit 151. IC1_out (t+Δt)~Q IC1_out (t+MΔt), ..., Q ICX_out (t+Δt)~Q ICX_out Remember (t+MΔt).

[0092] (Learning model generation unit 140) The learning model generation unit 140 acquires event information from the event information storage unit 120, mesh data from the mesh data storage unit 129, and facility differential traffic volume from the traffic volume data storage unit 124. Then, the learning model generation unit 140 generates a model (hereinafter also referred to as the "learning model") based on the acquired event information, mesh data, and facility differential traffic volume.

[0093] (Training data extraction unit 141) The learning data extraction unit 141 extracts learning data by executing the processes in the order of "extraction of traffic congestion data" and "extraction of learning data." These processes will be explained in order below.

[0094] (Extraction of traffic congestion data) The learning data extraction unit 141 extracts the times when traffic congestion occurs (hereinafter also referred to as "congestion times") and the cells where congestion occurs at those times (hereinafter also referred to as "congestion cells") based on the mesh data acquired from the mesh data storage unit 129. The congestion times extracted by the learning data extraction unit 141 may correspond to the first congestion times. The congestion cells extracted by the learning data extraction unit 141 may correspond to the first congestion locations. Note that a cell can also be described as a location on the road or a point on the road.

[0095] The learning data extraction unit 141 extracts combinations of congestion times and congestion cells as congestion data. If the learning data extraction unit 141 extracts multiple combinations of congestion times and congestion cells, it only needs to extract combinations that are consecutive in time or space from among the multiple combinations as congestion data. The congestion data extracted by the learning data extraction unit 141 may correspond to the first congestion data.

[0096] Figure 10 is a magnified view of a portion of the mesh data. Referring to Figure 10, a magnified view of the mesh data surrounding facilities IC2 and IC3 is shown. In Figure 10, the vertical axis represents time, and the horizontal axis represents locations on the road. Locations on the road can be represented by cells. The density of the hatching indicates the speed corresponding to the time and cell.

[0097] Although not shown in Figure 10, facility IC1 is located one position upstream from facility IC2. Also, facility IC4 is located one position downstream from facility IC3. Facilities IC5 to ICX are located sequentially downstream from facility IC4.

[0098] Figure 11 is a diagram illustrating the extraction of congestion data. Referring to Figure 11, congestion data D1 extracted from the mesh data around facility IC2 and facility IC3 shown in Figure 10 is shown. For example, the learning data extraction unit 141 may determine whether the speed is less than or equal to the congestion flow determination speed 641 based on the speed corresponding to each combination of time and cell. Then, if the learning data extraction unit 141 determines that the speed is less than or equal to the congestion flow determination speed 641, it may extract that cell as a congestion cell and extract that time as the congestion time.

[0099] The congestion flow determination speed 641 is a threshold value stored in the master data storage unit 126 beforehand. For example, if the prediction target is the traffic flow on an expressway, the congestion flow determination speed 641 may be set to 40 km / h in accordance with the definition of congestion set by the Ministry of Land, Infrastructure, Transport and Tourism. Congestion data D1 is a combination of multiple combinations of congestion time and congestion cell that are continuous in time or space. In the following explanation, it is mainly assumed that one congestion data D1 is extracted by the learning data extraction unit 141, but multiple congestion data may be extracted by the learning data extraction unit 141.

[0100] (Extraction of training data) Figure 12 shows the traffic congestion data D1. In the traffic congestion data D1 shown in Figure 12, the hatching indicating speed has been removed from within the traffic congestion data D1. The learning data extraction unit 141 performs a facility congestion determination for each facility ICj from facility IC1 to ICX, determining whether or not the congestion in the traffic congestion data D1 has reached the target facility ICj. For example, the learning data extraction unit 141 may determine whether or not the congestion in the traffic congestion data D1 has reached the target facility ICj by determining whether or not the target facility ICj is located in the congestion cell.

[0101] Furthermore, if the learning data extraction unit 141 determines that the congestion in the congestion data D1 has not reached the target facility ICj, it does not need to extract traffic condition data indicating the state of road traffic flow as learning data.

[0102] On the other hand, if congestion reaches the target facility ICj in congestion data D1, the traffic volume flowing out of the target facility ICj may fluctuate. For example, if the prediction target is highway traffic flow, when congestion reaches an interchange, the number of vehicles exiting the highway onto general roads increases, which may cause fluctuations in the traffic volume flowing out of the interchange. The learning data extraction unit 141 also extracts traffic state data as learning data when such traffic volume fluctuations occur. Therefore, if it is determined that congestion has reached the target facility ICj in congestion data D1, it is desirable to extract traffic state data corresponding to congestion data D1 as learning data.

[0103] With this configuration, traffic condition data can be extracted as training data when congestion reaches the target facility ICj. Even if no event has occurred, when congestion reaches the target facility ICj, there is a possibility of sudden traffic volume fluctuations due to natural congestion. Therefore, by using a training model generated based on the training data extracted in this way, it becomes possible to accurately predict traffic volume even when sudden traffic volume fluctuations occur due to natural congestion.

[0104] Specifically, the learning data extraction unit 141 identifies each combination of time and cell included in the congestion data D1 as a learning data extraction range, and extracts the traffic state data corresponding to the identified learning data extraction range as traffic state data corresponding to the congestion data D1. Note that the traffic state data corresponding to the congestion data D1 may correspond to the first traffic state data. Details of the traffic state data corresponding to the congestion data D1 will be explained later.

[0105] In the example shown in Figure 12, cells 37 to 71 are congestion cells, facility IC2 is located in cell 40, and facility IC3 is located in cell 60. Therefore, since facilities IC2 and IC3 are included in the congestion cells, the learning data extraction unit 141 may extract traffic state data corresponding to the congestion data D1 for each of facility IC2 and facility IC3 as learning data.

[0106] On the other hand, since the congestion cell does not include facilities IC1 and IC4-ICX, the learning data extraction unit 141 does not need to extract traffic condition data as learning data for facilities IC1 and IC4-ICX, respectively.

[0107] The learning data extraction unit 141 may, before performing facility congestion determination, determine whether or not an event causing congestion has occurred at the congestion time based on event information. Specific examples of events are as described above. In the example shown in Figure 12, time t-10Δt is the congestion start time of congestion data D1, and cell 71 is the congestion leading cell of congestion data D1. The learning data extraction unit 141 may determine whether or not an event has occurred based on whether or not the congestion start time t-10Δt falls within a predetermined time range based on the event occurrence time included in the event information, and whether or not the congestion leading cell 71 is located within a predetermined distance range based on the event occurrence location included in the event information.

[0108] The learning data extraction unit 141 may perform a facility congestion determination if no event occurs during the congestion time. On the other hand, if an event occurs during the congestion time, the learning data extraction unit 141 may extract traffic state data corresponding to the congestion data D1 as learning data without performing a facility congestion determination.

[0109] With this configuration, traffic condition data from when events causing congestion occur can be extracted as training data. Therefore, by using a learning model generated based on the training data extracted in this way, it becomes possible to accurately predict traffic volume when events occur.

[0110] The learning data extraction unit 141 may determine whether the congestion in the congestion data D1 has reached multiple facilities, including the target facility ICj, based on whether it has determined that the congestion has reached the target facility ICj in the congestion data D1, or whether it has determined that an event causing congestion has occurred at the time of congestion. If the learning data extraction unit 141 determines that the congestion in the congestion data D1 has not reached multiple facilities, it may extract traffic state data corresponding to the congestion data D1 as learning data.

[0111] On the other hand, if the learning data extraction unit 141 determines that congestion has reached multiple facilities in the congestion data D1, it may determine whether there is an upstream time when the target facility ICj is located at the upstream end of the multiple facilities. In this case, when congestion has reached multiple facilities, there is a tendency for fluctuations to occur in the traffic volume flowing out from the facility located at the upstream end.

[0112] Therefore, if the learning data extraction unit 141 determines that there is an upstream time when the target facility ICj is located at the upstream end, it may extract traffic state data corresponding to the upstream time and the target facility ICj from the congestion data D1 as learning data. On the other hand, when congestion reaches multiple facilities, there is a tendency for the traffic volume flowing out from facilities not located at the upstream end to be less variable. Therefore, if the learning data extraction unit 141 determines that there is no upstream time, it does not need to extract traffic state data from the congestion data D1 as learning data.

[0113] In the example shown in Figure 12, cells 37 to 71 are congestion cells, facility IC2 is located in cell 40, and facility IC3 is located in cell 60. Therefore, since facilities IC2 and IC3 are included in the congestion cells, the learning data extraction unit 141 may determine that the congestion has reached multiple facilities.

[0114] Facility IC2 is located at the uppermost reaches of multiple facilities at times t-6Δt and t-5Δt. Therefore, when the target facility ICj is facility IC2, the learning data extraction unit 141 may extract traffic state data corresponding to the target facility ICj and the times t-6Δt and t-5Δt, when the target facility ICj is located at the uppermost reaches of multiple facilities, as learning data corresponding to the uppermost reaches time and the target facility ICj.

[0115] For example, when the target facility ICj is facility IC2, the learning data extraction unit 141 may identify combinations of time t-6Δt and time t-5Δt where the target facility IC2 is located at the uppermost position among multiple facilities and the target facility IC2, as well as combinations of time and cell that exist around that combination, and extract traffic state data corresponding to these combinations as learning data corresponding to the uppermost time and the target facility IC2.

[0116] For example, the learning data extraction unit 141 may identify the area from facility IC1, located one stop upstream from the target facility IC2, to facility IC3, located one stop downstream from the target facility IC2, as the spatial learning data extraction range. Furthermore, the learning data extraction unit 141 may identify the area from the time t-9Δt, when congestion first reaches facility IC3, located one stop downstream from the target facility IC2, to the time t, when congestion first ceases to reach facility IC3, as the temporal learning data extraction range.

[0117] The learning data extraction unit 141 may then identify a learning data extraction range for combinations of time and cell within the traffic congestion data D1 (i.e., combinations of traffic congestion time and traffic congestion cell) that fall within the spatially identified learning data extraction range and the temporally identified learning data extraction range. The learning data extraction unit 141 may then extract traffic state data corresponding to the identified learning data extraction range as learning data.

[0118] On the other hand, facility IC2 is not located at the uppermost position among the multiple facilities from time t-9Δt to time t-7Δt, and from time t-4Δt to time t-Δt. Therefore, when the target facility ICj is facility IC2, the learning data extraction unit 141 does not need to extract traffic state data corresponding to the target facility IC2 from time t-9Δt to time t-7Δt, and from time t-4Δt to time t-Δt.

[0119] Facility IC3 is located at the uppermost reaches of multiple facilities from time t-9Δt to time t-7Δt, and from time t-4Δt to time t-Δt. Therefore, when the target facility ICj is facility IC3, the learning data extraction unit 141 may extract traffic state data corresponding to the uppermost reaches from time t-9Δt to time t-7Δt, and from time t-4Δt to time t-Δt, as well as the target facility IC3, as learning data.

[0120] On the other hand, facility IC3 is not located at the uppermost position among the multiple facilities from time t-6Δt to time t-5Δt. Therefore, when the target facility ICj is facility IC3, the learning data extraction unit 141 does not need to extract traffic state data corresponding to the period from time t-6Δt to time t-5Δt and the target facility IC3 as learning data.

[0121] Furthermore, at time t-10Δt and time t (current time), neither facility IC2 nor facility IC3 is located at the uppermost reaches. Therefore, at time t-10Δt and time t (current time), the learning data extraction unit 141 does not need to extract traffic state data corresponding to time t-10Δt and time t (current time) and the target facility ICj as learning data, regardless of whether the target facility ICj is facility IC2 or facility IC3.

[0122] The learning data extraction unit 141 can include various types of data in the traffic state data corresponding to the learning data extraction range.

[0123] For example, the learning data extraction unit 141 may calculate the time required for a vehicle to travel from the target facility ICj to the downstream facility ICj+1, which is located downstream of the target facility ICj, based on each combination of time and cell included in the learning data extraction range (hereinafter also simply referred to as "inter-facility travel time"), and include the calculated inter-facility travel time for each time in the traffic state data corresponding to the learning data extraction range. For example, the inter-facility travel time can be calculated by dividing the length of each cell by the speed corresponding to each cell from the target facility ICj to the downstream facility ICj+1 for each time and summing them up.

[0124] Alternatively, the learning data extraction unit 141 may calculate, based on each combination of time and cell included in the learning data extraction range, the difference between the time taken between the target facility ICj and the downstream facility ICj+1 at that time and the time taken between the target facility ICj and the downstream facility ICj+1 at the time immediately preceding that time, and include the calculated difference in the traffic state data corresponding to the learning data extraction range. This allows the degree of traffic flow recovery based on past traffic flow to be used for learning.

[0125] The learning data extraction unit 141 may calculate the length of congestion for each time period based on each combination of time and cell included in the learning data extraction range, and may include the calculated length of congestion for each time period in the traffic state data corresponding to the learning data extraction range. The length of congestion for each time period may be a count result obtained by counting the number of combinations of time and cell included in the learning data extraction range for each time period, or it may be a multiplication result obtained by multiplying the count result by the length of each cell for each time period.

[0126] The learning data extraction unit 141 may calculate the position of the end of the congestion for each time period based on each combination of time and cell included in the learning data extraction range, and may include the calculated position of the end of the congestion for each time period in the traffic state data corresponding to the learning data extraction range. The position of the end of the congestion for each time period may be the result of identifying the cell number located at the uppermost reaches included in the learning data extraction range for each time period, or it may be the result of multiplication obtained by multiplying the cell number by the length of each cell for each time period.

[0127] The learning data extraction unit 141 may calculate the average speed corresponding to a cell for each time period based on each combination of time and cell included in the learning data extraction range, and may include the calculated average speed corresponding to the cell for each time period in the traffic condition data corresponding to the learning data extraction range.

[0128] Furthermore, the scale of congestion may change depending on the event. Therefore, if the event information storage unit 120 stores information about an event, the learning data extraction unit 141 may retrieve information about the event from the event information storage unit 120 and include the retrieved event information in the traffic state data corresponding to the learning data extraction range.

[0129] For example, if event information associated with the event occurrence time and location is stored in the event information storage unit 120, the learning data extraction unit 141 may acquire event information associated with the event occurrence time and location if the conditions are met such that the congestion start time t-10Δt falls within a predetermined time range based on the event occurrence time, and the congestion leading cell 71 is located within a predetermined distance range based on the event occurrence location included in the event information. The event information may also include the number of lanes where traffic was restricted due to an accident or traffic regulation.

[0130] (Traffic Volume Fluctuation Learning Unit 142) The traffic volume fluctuation learning unit 142 calculates the facility difference traffic volume R for facilities IC1 to ICX stored in the traffic volume data storage unit 124. IC1 (t)~R IC1 (t-hΔt), ..., R ICX (t)~R ICX From (t-hΔt), the facility difference traffic volume of the target facility ICj corresponding to the time included in the learning data extraction range is obtained. Then, the traffic volume fluctuation learning unit 142 generates a learning model based on the learning data extracted by the learning data extraction unit 141 and the obtained facility difference traffic volume. Note that the facility difference traffic volume of the target facility ICj may correspond to the first difference.

[0131] More specifically, the traffic volume fluctuation learning unit 142 generates a learning model by analyzing the acquired facility-differential traffic volume, using the learning data as explanatory variables and the acquired facility-differential traffic volume as the dependent variable. Since facility-differential traffic volume reflects sudden traffic volume fluctuations, the traffic volume fluctuation learning unit 142 can learn how much sudden traffic volume fluctuations occur according to the explanatory variables by analyzing such facility-differential traffic volume as the dependent variable.

[0132] Furthermore, the type of analysis used by the traffic volume fluctuation learning unit 142 is not limited. For example, if the learning data includes multiple types of traffic condition data, the traffic volume fluctuation learning unit 142 may use any regression analysis that can treat each of the multiple types of traffic condition data included in the learning data as an explanatory variable.

[0133] (Learning model memory unit 127) The learning model storage unit 127 stores the learning model generated by the traffic volume fluctuation learning unit 142.

[0134] (Prediction unit 150) The prediction unit 150 retrieves the learning model from the learning model storage unit 127. Furthermore, the prediction unit 150 calculates the speed V1(t)~V corresponding to time t (current time). n (t) is obtained from the mesh data storage unit 129. The prediction unit 150 calculates the speed V1(t)~V corresponding to time t (current time). nBased on (t), traffic condition data for the target facility ICj is extracted as prediction data, and based on the extracted prediction data and the learning model, facility outflow traffic volume Q for facilities IC1 to ICX is calculated. IC1_out (t+Δt)~Q IC1_out (t+MΔt), ..., Q ICX_out (t+Δt)~Q ICX_out Predict (t+MΔt).

[0135] Here, the traffic condition data for the target facility ICj may fall under the category of second traffic condition data. Also, the facility outflow traffic volume Q for the target facility ICj... ICj_out (t+Δt)~Q ICj_out (t+MΔt) could be the second traffic volume data.

[0136] Furthermore, the speed corresponding to the current time used by the prediction unit 150 is the speed V1(t)~V corresponding to the time t (current time) used by the learning model generation unit 140. n Data other than (t) may be used. For example, a virtual speed calculated by traffic flow simulation may be used as the speed corresponding to the current time used by the prediction unit 150.

[0137] (Judgment unit 1521) The determination unit 1521 determines the speed V1(t)~V corresponding to time t (current time). n Based on (t), a congested cell is extracted, which is a cell where congestion is occurring at time t (current time). Time t (current time) may correspond to a second congestion time. Also, the congested cell extracted by the determination unit 1521 may correspond to a second congestion location.

[0138] The determination unit 1521 extracts combinations of time t (current time) and congestion cells as congestion data D2 (Figures 11 and 12). If the determination unit 1521 extracts multiple combinations of time t (current time) and congestion cells, it only needs to extract spatially consecutive combinations from among these multiple combinations as congestion data D2. Alternatively, the determination unit 1521 may extract congestion data corresponding to time t (current time) from the congestion data D1 extracted by the learning data extraction unit 141 as congestion data D2. The congestion data D2 extracted by the determination unit 1521 may correspond to second congestion data.

[0139] The determination unit 1521 performs a facility congestion determination for each facility ICj from facility IC1 to ICX, determining whether or not congestion has reached the target facility ICj in the extracted congestion data D2. For example, the determination unit 1521 may determine whether or not congestion has reached the target facility ICj in the congestion data D2 by determining whether or not the target facility ICj is located in the congestion cell. In the example shown in Figure 12, congestion has not reached any of the facilities IC1 to ICX in the congestion data D2.

[0140] Then, if the determination unit 1521 determines in the congestion data D2 that the congestion has not reached the target facility ICj, then it determines the facility outflow traffic volume Q of the target facility ICj. ICj_out (t+Δt)~Q ICj_out It is determined that no correction is needed for (t+MΔt), and therefore, traffic condition data does not need to be extracted as prediction data. On the other hand, if the congestion in the congestion data D2 reaches the target facility ICj, then the facility outflow traffic volume Q of the target facility ICj is determined. ICj_out (t+Δt)~Q ICj_out It is determined that a correction to (t+MΔt) is necessary, and traffic condition data corresponding to congestion data D2 is extracted as prediction data for the target facility ICj.

[0141] Specifically, the determination unit 1521 identifies each combination of time and cell included in the congestion data D2 as a prediction data extraction range, and then extracts the traffic condition data corresponding to the identified prediction data extraction range as traffic condition data corresponding to the congestion data D2. Details of the traffic condition data corresponding to the congestion data D2 will be explained later.

[0142] Before performing a facility congestion determination, the determination unit 1521 may determine, based on event information, whether or not an event causing congestion has occurred at the congestion time. In the example shown in Figure 12, cell 70 is the congestion leading cell in congestion data D2. The determination unit 1521 may determine whether or not an event has occurred based on whether or not the conditions are met that time t (current time) falls within a predetermined time range based on the event occurrence time included in the event information, and congestion leading cell 70 is located within a predetermined distance range based on the event occurrence location included in the event information.

[0143] The determination unit 1521 may perform a facility congestion determination if no event occurs during the congestion time. On the other hand, if an event occurs during the congestion time, the determination unit 1521 will not perform a facility congestion determination and will instead determine the facility outflow traffic volume Q of the target facility ICj. ICj_out (t+Δt)~Q ICj_out If it is determined that a correction to (t+MΔt) is necessary, traffic condition data corresponding to congestion data D2 may be extracted as prediction data for the target facility ICj.

[0144] The determination unit 1521 may determine whether the congestion in the congestion data D2 has reached multiple facilities, including the target facility ICj, based on the determination that the congestion has reached the target facility ICj in the congestion data D2, or the determination that an event causing congestion has occurred at the time of congestion. If the determination unit 1521 determines that the congestion in the congestion data D2 has not reached multiple facilities, it may determine the facility outflow traffic volume Q of the target facility ICj. ICj_out (t+Δt)~Q ICj_outIf it is determined that a correction to (t+MΔt) is necessary, traffic condition data corresponding to congestion data D2 may be extracted as prediction data for the target facility ICj.

[0145] On the other hand, if the determination unit 1521 determines in the congestion data D2 that congestion has reached multiple facilities, it may determine whether there is an upstream time when the target facility ICj is located at the upstream end of the multiple facilities. If the determination unit 1521 determines that there is an upstream time when the target facility ICj is located at the upstream end, it may determine the facility outflow traffic volume Q of the target facility ICj. ICj_out (t+Δt)~Q ICj_out If it is determined that a correction to (t+MΔt) is necessary, traffic condition data corresponding to the upstream time and the target facility ICj may be extracted from the congestion data D2 as prediction data for the target facility ICj.

[0146] As an example, the determination unit 1521 may identify a prediction data extraction range corresponding to the upstream time and the target facility ICj using a method similar to the method used by the learning data extraction unit 141 to identify a learning data extraction range corresponding to the upstream time and the target facility ICj. The determination unit 1521 may then extract traffic state data corresponding to the identified prediction data extraction range as prediction data. On the other hand, if the determination unit 1521 determines that there is no upstream time, it does not need to extract traffic state data from the congestion data D2 as prediction data for the target facility ICj.

[0147] The determination unit 1521 can include various types of data in the traffic condition data corresponding to the prediction data extraction range of the target facility ICj.

[0148] For example, the determination unit 1521 may calculate the inter-facility travel time, which is the time required for a vehicle to travel from the target facility ICj at time t (current time) to the downstream facility ICj+1, which is a facility located downstream of the target facility ICj, based on the combination of time t (current time) and cell included in the prediction data extraction range of the target facility ICj, and include the calculated inter-facility travel time in the traffic condition data corresponding to the prediction data extraction range of the target facility ICj.

[0149] Alternatively, the determination unit 1521 may calculate the difference between the inter-facility travel time from the target facility ICj to the downstream facility ICj+1 at time t (current time) and the inter-facility travel time from the target facility ICj to the downstream facility ICj+1 at the time t-Δt, which is one time prior to time t, based on the combination of time t (current time) and cell included in the prediction data extraction range of the target facility ICj, and include the calculated difference in the traffic condition data corresponding to the prediction data extraction range of the target facility ICj.

[0150] The determination unit 1521 may calculate the length of congestion at time t (current time) based on the combination of time t (current time) and cell included in the prediction data extraction range, and may include the calculated length of congestion at time t (current time) in the traffic condition data corresponding to the prediction data extraction range of the target facility ICj.

[0151] The determination unit 1521 may calculate the position of the end of the congestion at time t (current time) based on the combination of time t (current time) and cell included in the prediction data extraction range, and may include the calculated position of the end of the congestion at time t (current time) in the traffic condition data corresponding to the prediction data extraction range of the target facility ICj.

[0152] The determination unit 1521 may calculate the average speed corresponding to the cell at time t (current time) based on the combination of time t (current time) and cell included in the prediction data extraction range, and may include the calculated average speed corresponding to the cell at time t (current time) in the traffic condition data corresponding to the prediction data extraction range of the target facility ICj.

[0153] Furthermore, if event information is stored in the event information storage unit 120, the determination unit 1521 may retrieve event information from the event information storage unit 120 and include the retrieved event information in the traffic condition data corresponding to the prediction data extraction range of the target facility ICj.

[0154] As an example, when information about an event associated with an event occurrence time and an event occurrence location is stored in the event information storage unit 120, the determination unit 1521 may acquire information about the event associated with the event occurrence time and the event occurrence location when the following conditions are satisfied: the time t (current time) belongs to a predetermined time range based on the event occurrence time, and the leading cell 70 of the traffic jam data D2 is located within a predetermined distance range based on the event occurrence location included in the event information.

[0155] (Traffic volume change prediction unit 1522) Based on the learning model and the prediction data of the target facility ICj (hereinafter, also referred to as "target facility ICj for correction") for which it is determined that correction of the facility outflow traffic volume is necessary, the traffic volume change prediction unit 1522 predicts the facility differential traffic volume R ICj (t + Δt) to R ICj (t + MΔt) of the target facility ICj for correction. Note that the facility differential traffic volume R ICj (t + Δt) to R ICj (t + MΔt) may correspond to the second difference.

[0156] More specifically, the traffic volume change prediction unit 1522 performs regression analysis using the learning model with the prediction data of the target facility ICj for correction as an input value to the learning model, and obtains the predicted values at times t + Δt to t + MΔt obtained by the regression analysis as the facility differential traffic volume R ICj (t + Δt) to R ICj (t + MΔt) of the target facility ICj for correction.

[0157] (Traffic volume correction unit 1523) Based on the facility differential traffic volume R ICj (t + Δt) to R ICj (t + MΔt) of the target facility ICj for correction, the traffic volume correction unit 1523 corrects the facility outflow traffic volume Q ICj_out (t + Δt) to Q ICj_out (t + MΔt) of the target facility ICj for correction. Thereby, the traffic volume correction unit 1523 corrects the facility outflow traffic volume QICj_out (t + Δt) to Q ICj_out Obtain the corrected facility outflow traffic volume at (t + MΔt).

[0158] In the following, the facility outflow traffic volume Q of the facility ICj to be corrected ICj_out (t + Δt) to Q ICj_out Let the corrected facility outflow traffic volume at (t + MΔt) be Q'. ICj_out (t + Δt) to Q' ICj_out It is also denoted as (t + MΔt). In this way, the traffic volume correction unit 1523 can predict the facility outflow traffic volume Q' ICj_out to Q' ICj_out (t + MΔt). Note that the corrected facility outflow traffic volume Q' ICj_out to Q' ICj_out (t + MΔt) may correspond to the second traffic volume data.

[0159] More specifically, for each time from t + Δt to t + MΔt, the traffic volume correction unit 1523 subtracts the correction amount corresponding to the facility differential traffic volume R of the facility ICj to be corrected from the facility outflow traffic volume Q ICj to correct the facility outflow traffic volume Q ICj_out Furthermore, more specifically, for each time from t + Δt to t + MΔt, the traffic volume correction unit 1523 may increase the correction amount corresponding to the facility differential traffic volume R as the facility differential traffic volume R ICj_out is larger. ICj is larger. ICj

[0160] The traffic volume correction unit 1523 corrects the facility outflow traffic volume Q IC1_out (t + Δt) to Q IC1_out (t + MΔt), ···, Q ICX_out (t + Δt) to Q ICX_out among the facility outflow traffic volumes Q ICj_out to Q ICj_out (t + MΔt) of the facilities IC1 to ICX stored by the prediction result storage unit 128 into the corrected facility outflow traffic volume Q' ICj_out to Q' ICj_out ​The data is updated at (t+MΔt). As a result, the prediction result storage unit 128 stores the predicted traffic volume outflow for each of the target facilities IC1 to ICX at times t+Δt to t+MΔt.

[0161] The configuration of the traffic volume prediction device 1 according to an embodiment of the present invention has been described above.

[0162] (1-2. Example of operation) Next, referring to Figures 13 to 19 (and Figures 1 to 12 as appropriate), we will summarize examples of the operation of the traffic volume prediction device 1 according to an embodiment of the present invention. Note that the operation examples shown in Figures 13 to 19 are merely examples of the operation of the traffic volume prediction device 1. Therefore, the operation of the traffic volume prediction device 1 is not limited to the operation examples shown in Figures 13 to 19.

[0163] Figure 13 is a flowchart showing an example of the overall operation of the traffic volume prediction device 1 according to an embodiment of the present invention. As shown in Figure 13, the traffic volume prediction device 1 according to an embodiment of the present invention performs data acquisition processing (S201).

[0164] Specifically, the traffic volume prediction device 1 acquires event information, and the acquired event information is stored in the event information storage unit 120. The traffic volume prediction device 1 also acquires probe data, and the acquired probe data is stored in the probe data storage unit 122. Furthermore, the traffic volume prediction device 1 acquires mesh data, and the acquired mesh data is stored in the mesh data storage unit 129. Finally, the traffic volume prediction device 1 acquires facility outflow traffic volume and facility differential traffic volume, and the acquired facility outflow traffic volume and facility differential traffic volume are stored in the traffic volume data storage unit 124.

[0165] The traffic volume prediction unit 151 obtains the facility exit traffic volume from the traffic volume data storage unit 124 and performs traffic volume prediction processing to predict future facility exit traffic volume based on the facility exit traffic volume (S202).

[0166] Specifically, the traffic volume prediction unit 151 predicts the outflow traffic volume Q from facility IC1 to ICX.IC1_out (t) ~ Q IC1_out (t-hΔt), ..., Q ICX_out (t) ~ Q ICX_out Based on (t-hΔt), Q IC1_out (t+Δt)~Q IC1_out (t+MΔt), ..., Q ICX_out (t+Δt)~Q ICX_out Predict (t+MΔt).

[0167] The prediction result storage unit 128 stores the facility outflow traffic volume Q for facilities IC1 to ICX, which is output from the traffic volume prediction unit 151. IC1_out (t+Δt)~Q IC1_out (t+MΔt), ..., Q ICX_out (t+Δt)~Q ICX_out (t+MΔt) is stored. Next, the learning model generation unit 140 performs the learning model generation process (S203). The details of the learning model generation process will be explained with reference to Figure 14.

[0168] Figure 14 is a flowchart detailing the learning model generation process. As shown in Figure 14, the learning model generation unit 140 performs data acquisition processing (S301). Specifically, the learning model generation unit 140 acquires event information from the event information storage unit 120, mesh data from the mesh data storage unit 129, and facility differential traffic volume from the traffic volume data storage unit 124. Here, facility differential traffic volume is the facility differential traffic volume R of facilities IC1 to ICX. IC1 (t)~R IC1 (t-hΔt), ..., R ICX (t)~R ICX (t-hΔt)

[0169] The training data extraction unit 141 performs the training data extraction process (S302). The details of the training data extraction process will be explained with reference to Figures 15 and 16.

[0170] Figure 15 is a flowchart (first half) showing the details of the training data extraction process. As shown in Figure 15, the training data extraction unit 141 extracts the times when congestion occurs as congestion times, and also extracts the cells where congestion occurs at those times as congestion cells, based on the mesh data acquired from the mesh data storage unit 129.

[0171] The learning data extraction unit 141 extracts combinations of congestion times and congestion cells as congestion data (S401). If the learning data extraction unit 141 extracts multiple combinations of congestion times and congestion cells, it only needs to extract combinations that are consecutive in time or space from among the multiple combinations as congestion data. The learning data extraction unit 141 extracts one or more congestion data.

[0172] The learning data extraction unit 141 repeats the process for each facility ICj from facility IC1 to ICX (S402 to S416). In the facility-specific iteration, the learning data extraction unit 141 repeats the process for each traffic congestion data (S403 to S415). In the traffic congestion data iteration, the learning data extraction unit 141 determines, based on event information, whether or not an event causing traffic congestion occurred at the time of congestion (S404).

[0173] If the learning data extraction unit 141 determines that an event causing congestion occurred during the congestion time (S404:YES), it proceeds to S406. On the other hand, if the learning data extraction unit 141 determines that no event causing congestion occurred during the congestion time (S404:NO), it performs a facility congestion determination to determine whether or not the congestion has reached the target facility ICj (S405).

[0174] Then, if the learning data extraction unit 141 determines in the congestion data that the congestion has not reached the target facility ICj (S405: NO), it proceeds to S415 (Figure 16) without extracting the traffic state data as learning data. On the other hand, if the learning data extraction unit 141 determines in the congestion data that the congestion has reached the target facility ICj (S405: YES), it proceeds to S406.

[0175] The learning data extraction unit 141 determines in the congestion data whether the congestion reaches multiple facilities including the target facility ICj (S406). If the learning data extraction unit 141 determines in the congestion data that the congestion does not reach multiple facilities including the target facility ICj (S406: NO), it extracts traffic state data corresponding to the congestion data as learning data (S407) and proceeds to S415 (Figure 16). On the other hand, if the learning data extraction unit 141 determines in the congestion data that the congestion reaches multiple facilities including the target facility ICj (S406: YES), it proceeds to S411 (Figure 16).

[0176] Figure 16 is a flowchart (second half) showing the details of the learning data extraction process. As shown in Figure 16, the learning data extraction unit 141 repeats the process for each time interval of the traffic congestion data (S411-414). In each time interval, the learning data extraction unit 141 determines whether the target facility ICj is located at the uppermost position among multiple facilities (S412).

[0177] If the learning data extraction unit 141 determines that the target facility ICj is not located at the uppermost position among multiple facilities (S412: NO), it proceeds to S414. On the other hand, if the learning data extraction unit 141 determines that the target facility ICj is located at the uppermost position among multiple facilities (S412: YES), it extracts traffic condition data corresponding to the time and the target facility ICj as learning data (S413), and proceeds to S414.

[0178] When the time-based repetitions S411-S414 are completed, the traffic congestion data-based repetitions S403-S415 are completed, and the facility-based repetitions S402-S416 are completed, the learning data extraction unit 141 outputs the extracted learning data (S417). Return to Figure 14 and continue the explanation.

[0179] The traffic volume fluctuation learning unit 142 performs learning processing (S303). Specifically, the traffic volume fluctuation learning unit 142 processes the facility difference traffic volume R of facilities IC1 to ICX stored in the traffic volume data storage unit 124. IC1 (t)~R IC1 (t-hΔt), ..., R ICX (t)~R ICX From (t-hΔt), the facility difference traffic volume for the target facility ICj corresponding to the time included in the training data extraction range is obtained.

[0180] The traffic volume fluctuation learning unit 142 then generates a learning model based on the learning data extracted by the learning data extraction unit 141 and the acquired facility difference traffic volume. The traffic volume fluctuation learning unit 142 outputs the generated learning model (S304). The learning model storage unit 127 stores the learning model output by the traffic volume fluctuation learning unit 142. Return to Figure 13 and continue the explanation.

[0181] Next, the correction processing unit 152 performs the correction process (S204). The details of the correction process will be explained with reference to Figure 17.

[0182] Figure 17 is a flowchart showing the details of the correction process. As shown in Figure 17, the correction processing unit 152 performs data acquisition processing (S501). Specifically, the correction processing unit 152 acquires the learning model from the learning model storage unit 127, acquires event information from the event information storage unit 120, and acquires mesh data from the mesh data storage unit 129.

[0183] The determination unit 1521 performs a correction necessity determination process (S502). The details of the correction necessity determination process will be explained with reference to Figures 18 and 19.

[0184] Figure 18 is a flowchart (first half) showing the details of the correction necessity determination process. As shown in Figure 18, the determination unit 1521 sets the correction flag indicating the necessity of correction for each facility IC1 to ICX to "No correction required" (S600). The determination unit 1521 attempts to extract congestion data from the mesh data acquired from the mesh data storage unit 129 (S601). Specifically, the determination unit 1521 attempts to extract congestion data based on the mesh data, depending on whether or not a congestion cell corresponding to time t (current time) exists.

[0185] Furthermore, if the determination unit 1521 extracts multiple combinations of time t (current time) and congestion cells, it only needs to extract spatially consecutive combinations from among these multiple combinations as congestion data.

[0186] If the determination unit 1521 does not extract traffic congestion data (S602: NO), it proceeds to S613 (Figure 19). On the other hand, if the determination unit 1521 does extract traffic congestion data (S602: YES), it proceeds to S603.

[0187] The determination unit 1521 repeats the process for each facility ICj from facility IC1 to ICX (S603 to S613). In each facility iteration, the determination unit 1521 determines, based on event information, whether or not an event causing congestion has occurred at time t (current time) (S604).

[0188] If the determination unit 1521 determines that an event causing congestion occurred at time t (current time) (S604: YES), it proceeds to S606. On the other hand, if the determination unit 1521 determines that no event causing congestion occurred at time t (current time) (S604: NO), it performs a facility congestion determination to determine whether or not the congestion has reached the target facility ICj (S605).

[0189] Then, if the determination unit 1521 determines in the congestion data that the congestion has not reached the target facility ICj (S605: NO), it proceeds to S613 (Figure 19) without setting the correction flag corresponding to the target facility ICj to "Correction Required". On the other hand, if the determination unit 1521 determines in the congestion data that the congestion has reached the target facility ICj (S605: YES), it proceeds to S606.

[0190] The determination unit 1521 determines in the congestion data whether the congestion reaches multiple facilities including the target facility ICj (S606). If the determination unit 1521 determines in the congestion data that the congestion does not reach multiple facilities including the target facility ICj (S606: NO), it sets the correction flag corresponding to the target facility ICj to "Correction Required" (S607), extracts traffic state data corresponding to the congestion data as prediction data for the target facility ICj, and proceeds to S613 (Figure 19). On the other hand, if the determination unit 1521 determines in the congestion data that the congestion reaches multiple facilities including the target facility ICj (S606: YES), it proceeds to S611 (Figure 19).

[0191] Figure 19 is a flowchart (second half) showing the details of the correction necessity determination process. As shown in Figure 19, the determination unit 1521 determines whether the target facility ICj is located at the uppermost position among multiple facilities (S611). If the determination unit 1521 determines that the target facility ICj is not located at the uppermost position among multiple facilities (S611: NO), it proceeds to S613. On the other hand, if the determination unit 1521 determines that the target facility ICj is located at the uppermost position among multiple facilities (S612: YES), it sets the correction flag corresponding to the target facility ICj to "Correction Required" (S612), extracts traffic condition data corresponding to the uppermost time and the target facility ICj as prediction data for the target facility ICj, and proceeds to S613. After the repetition of S603 to S613 for each facility is completed, the correction flag is output (S614). Return to Figure 17 and continue the explanation.

[0192] The traffic volume fluctuation prediction unit 1522 determines whether there are any target facility ICj that require correction based on whether there are any target facility ICj with the correction flag set to "Correction Required" (S503). If there are no target facility ICj that require correction (S503: NO), the traffic volume fluctuation prediction unit 1522 terminates the process. On the other hand, if there are target facility ICj that require correction (S503: YES), the traffic volume fluctuation prediction unit 1522 calculates the facility difference traffic volume R of the target facility ICj based on the learning model and the prediction data of the target facility ICj that requires correction. ICj (t+Δt)~R ICj Predict (t+MΔt) (S504).

[0193] The traffic volume correction unit 1523 receives the facility outflow traffic volume Q of the target facility ICj from the traffic volume data storage unit 124. ICj_out (t+Δt)~Q ICj_out (t+MΔt) is obtained (S505). The traffic volume correction unit 1523 calculates the facility difference traffic volume R of the target facility ICj that requires correction. ICj (t+Δt)~R ICj Based on (t+MΔt), the facility outflow traffic volume Q of the facility ICj that requires correction. ICj_out (t+Δt)~Q ICj_out By correcting (t+MΔt), the corrected facility outflow traffic volume is given by Q' ICj_out (t+Δt)~Q' ICj_out We obtain (t+MΔt) (S506).

[0194] The traffic volume correction unit 1523 processes the facility outflow traffic volume Q for facilities IC1 to ICX, which is stored by the prediction result storage unit 128. IC1_out (t+Δt)~Q IC1_out (t+MΔt), ..., Q ICX_out (t+Δt)~Q ICX_out Of (t+MΔt), the facility outflow traffic volume Q of the facility ICj that requires correction. ICj_out ~Q ICj_out (t+MΔt) is the corrected facility outflow traffic volume Q' ICj_out ~Q' ICj_out The data is updated at (t+MΔt). As a result, the prediction result storage unit 128 stores the predicted traffic volume outflow for each of the target facilities IC1 to ICX at times t+Δt to t+MΔt.

[0195] Returning to Figure 13, the explanation continues. Once the correction process S204 is completed, the operation of the traffic volume prediction device 1 according to the embodiment of the present invention ends.

[0196] The above describes an example of the operation of the traffic volume prediction device 1 according to an embodiment of the present invention.

[0197] (1-3. Effects) As described above, according to the embodiment of the present invention, when congestion reaches the target facility in the congestion data, traffic state data corresponding to the congestion data is extracted as training data. Then, a learning model is generated based on the training data, and traffic volume data at the facility is predicted based on the learning model.

[0198] Even when no event is occurring, if congestion reaches the target facility, sudden traffic volume fluctuations may occur due to natural congestion. Therefore, by using a learning model generated based on the training data extracted in this way, it becomes possible to accurately predict traffic volume even when sudden traffic volume fluctuations occur due to natural congestion.

[0199] Figure 20 is a graph showing the relationship between actual traffic volume, predicted traffic volume (uncorrected), and predicted traffic volume (corrected) and time (Day 1). Figure 21 is a graph showing the relationship between actual traffic volume, predicted traffic volume (uncorrected), and predicted traffic volume (corrected) and time (Day 2). Figure 22 is a graph showing the relationship between actual traffic volume, predicted traffic volume (uncorrected), and predicted traffic volume (corrected) and time (Day 3). Figure 23 is a graph showing the relationship between actual traffic volume, predicted traffic volume (uncorrected), and predicted traffic volume (corrected) and time (Day 4).

[0200] The predicted traffic volume (with correction) is the traffic volume predicted using the correction according to the embodiment of the present invention. On the other hand, the predicted traffic volume (without correction) is the traffic volume predicted without using the correction according to the embodiment of the present invention.

[0201] Referring to Figures 20 to 22, there are time periods in which the actual traffic volume and the predicted traffic volume (uncorrected) change at a similar rate. However, it can be seen that there are parts where the actual traffic volume and the predicted traffic volume (uncorrected) diverge. On the other hand, when comparing the actual traffic volume and the predicted traffic volume (uncorrected), it can be seen that corrections have been made to the parts of the predicted traffic volume (uncorrected) that deviate from the actual traffic volume, and the predicted traffic volume (corrected) is closer to the actual traffic volume.

[0202] On the other hand, referring to the graph showing the relationship between predicted traffic volume (uncorrected) and predicted traffic volume (corrected) and time (Day 4), it can be seen that no major congestion occurred on Day 4. On such days, no sudden fluctuations in traffic volume occur, so no correction is made to the predicted traffic volume (uncorrected). Therefore, in the graph shown in Figure 23, it can be confirmed that the predicted traffic volume (corrected) has not been corrected in areas where correction is unnecessary.

[0203] The effects of the traffic volume prediction device 1 according to the embodiment of the present invention have been described above.

[0204] (2. Hardware Configuration Example) Next, an example of the hardware configuration of the traffic volume prediction device 1 according to an embodiment of the present invention will be described.

[0205] In the following, an example of the hardware configuration of the information processing device 900 will be described as an example of the hardware configuration of the traffic volume prediction device 1 according to an embodiment of the present invention. Note that the example of the hardware configuration of the information processing device 900 described below is merely one example of the hardware configuration of the traffic volume prediction device 1. Therefore, the hardware configuration of the traffic volume prediction device 1 may be modified by removing unnecessary components from the hardware configuration of the information processing device 900 described below, or by adding new components.

[0206] Figure 24 shows the hardware configuration of an information processing device 900 as an example of a traffic volume prediction device 1 according to an embodiment of the present invention. The information processing device 900 includes a CPU (Central Processing Unit) 901, a ROM (Read Only Memory) 902, a RAM (Random Access Memory) 903, a host bus 904, a bridge 905, an external bus 906, an interface 907, an input device 908, an output device 909, a storage device 910, and a communication device 911.

[0207] The CPU 901 functions as both an arithmetic processing unit and a control unit, controlling the overall operation of the information processing unit 900 according to various programs. The CPU 901 may also be a microprocessor. The ROM 902 stores programs and arithmetic parameters used by the CPU 901. The RAM 903 temporarily stores programs used in the execution of the CPU 901 and parameters that change as needed during its execution. These are interconnected by a host bus 904, which consists of a CPU bus and other components.

[0208] The host bus 904 is connected to an external bus 906, such as a PCI (Peripheral Component Interconnect / Interface) bus, via a bridge 905. It is not always necessary to configure the host bus 904, bridge 905, and external bus 906 separately; these functions may be implemented on a single bus.

[0209] The input device 908 consists of input means for the user to input information, such as a mouse, keyboard, touch panel, buttons, microphone, switches, and levers, and an input control circuit that generates input signals based on the user's input and outputs them to the CPU 901. The user operating the information processing device 900 can input various types of data to the information processing device 900 or instruct it to perform processing operations by operating this input device 908.

[0210] The output device 909 includes, for example, display devices such as a CRT (Cathode Ray Tube) display device, a liquid crystal display (LCD) device, an OLED (Organic Light Emitting Diode) device, a lamp, and audio output devices such as a speaker.

[0211] The storage device 910 is a device for storing data. The storage device 910 may include a storage medium, a recording device for recording data on the storage medium, a reading device for reading data from the storage medium, a deleting device for deleting data recorded on the storage medium, and the like. The storage device 910 is constituted of, for example, an HDD (Hard Disk Drive). This storage device 910 drives a hard disk and stores programs executed by the CPU 901 and various data.

[0212] The communication device 911 is a communication interface constituted of, for example, a communication device for connecting to a network. Also, the communication device 911 may support either wireless communication or wired communication.

[0213] The hardware configuration example of the traffic volume prediction device 1 according to the embodiment of the present invention has been described above.

[0214] (3. Summary) Although the preferred embodiments of the present invention have been described in detail with reference to the accompanying drawings, the present invention is not limited to such examples. It is obvious that those having ordinary knowledge in the technical field to which the present invention pertains can come up with various modification examples or correction examples within the scope of the technical idea described in the claims, and these are also naturally understood to belong to the technical scope of the present invention.

Explanation of Reference Numerals

[0215] 1 Traffic volume prediction device 112 Probe antenna 114 Free flow antenna 120 Event information storage unit 121 Driving history data storage unit 122 Probe data storage unit 123 Free-flow data storage unit 124 Traffic volume data storage unit 125 Traffic density data storage unit 126 Master Data Storage Unit 127 Learning Model Memory Unit 128 Prediction Result Storage Unit 129 Mesh data storage unit 131 Statistics Processing Department 133 Traffic calculation department 135 Mesh Data Generation Unit 140 Learning Model Generation Unit 141 Training Data Extraction Unit 142 Traffic Volume Fluctuation Learning Department 150 Prediction Section 151 Traffic Volume Forecasting Section 152 Correction Processing Unit 1521 Judgment section 1522 Traffic Volume Fluctuation Prediction Department 1523 Traffic volume correction section 160 Event Information Input Section

Claims

1. A learning data extraction unit extracts first traffic state data corresponding to the first traffic data as learning data, based on the determination that the congestion has reached the target facility in first traffic data which includes a first congestion time which is the time when congestion occurs and a first congestion location which is the location where congestion occurs at the first congestion time. A learning unit that generates a model based on the aforementioned learning data and the first traffic volume data at the target facility, A prediction unit predicts second traffic volume data at the target facility based on the aforementioned model and second traffic condition data, A traffic volume prediction device equipped with the following features.

2. The learning data extraction unit determines whether the congestion has reached the target facility in the first congestion data if no predetermined event causing congestion has occurred at the first congestion time. The traffic volume prediction device according to claim 1.

3. The learning data extraction unit acquires information about the event as the first traffic state data. The traffic volume prediction device according to claim 2.

4. The learning data extraction unit calculates, based on the first traffic congestion data, at least one of the following as the first traffic condition data: the length of the congestion at the first congestion time, the position of the end of the congestion at the first congestion time, and the average value of the vehicle speed corresponding to the first congestion time. The traffic volume prediction device according to claim 1.

5. The learning data extraction unit calculates the difference in time required for a vehicle to travel from the target facility to a downstream facility, which is a facility located downstream of the target facility, at the first congestion time and at a time prior to the first congestion time, as the first traffic condition data. The traffic volume prediction device according to claim 1.

6. The learning unit generates the model by performing an analysis using the learning data as explanatory variables and the first traffic volume data as the dependent variable. The traffic volume prediction device according to claim 1.

7. The learning data extraction unit determines whether a statistical quantity of vehicle speed corresponding to a first position and a first time is below a threshold, and if it determines that the statistical quantity is below the threshold, it extracts the first position as the first congestion location and the first time as the first congestion time. The traffic volume prediction device according to claim 1.

8. The learning data extraction unit determines whether the target facility is located at the first congestion location, thereby determining whether the congestion has reached the target facility in the first congestion data. The traffic volume prediction device according to claim 1.

9. The traffic volume prediction device includes a traffic volume calculation unit that calculates a first difference between the traffic volume flowing into the target facility and the traffic volume flowing out of the target facility during the first congestion time as the first traffic volume data. The traffic volume prediction device according to claim 1.

10. The learning data extraction unit determines, based on whether it has determined in the first traffic congestion data that the congestion has reached the target facility, or whether a predetermined event causing congestion has occurred at the first traffic congestion time, whether the congestion has reached multiple facilities including the target facility in the first traffic congestion data; if it has determined that the congestion has reached multiple facilities in the first traffic congestion data, it determines whether there is an upstream time when the target facility is located at the upstream end of the multiple facilities; and if it has determined that there is an upstream time, it extracts first traffic state data corresponding to the upstream time and the target facility as the learning data. The traffic volume prediction device according to claim 1.

11. The prediction unit, The system includes a determination unit that determines whether the congestion has reached the target facility in second congestion data, which includes a second congestion time, which is the time when congestion occurs, and a second congestion location, which is the location where congestion occurs at the second congestion time, and extracts second traffic state data corresponding to the second congestion data based on the determination that the congestion has reached the target facility in the second congestion data. The traffic volume prediction device according to claim 1.

12. The prediction unit, A traffic volume prediction unit predicts the amount of traffic flowing out of the target facility at a time after the second congestion time, based on the amount of traffic flowing out of the target facility at a second congestion time, which is the time when congestion occurs. A traffic volume fluctuation prediction unit predicts a second difference between the amount of traffic flowing into the target facility at a later time and the amount of traffic flowing out of the target facility at a later time, based on the aforementioned model and the aforementioned second traffic condition data. A traffic volume correction unit predicts the second traffic volume data by correcting the outflow traffic volume from the target facility at a later time based on the second difference, A traffic volume prediction device according to claim 1, comprising:

13. The traffic volume correction unit corrects the traffic volume flowing out of the target facility at a later time by subtracting a correction amount corresponding to the second difference from the traffic volume flowing out of the target facility at a later time. The traffic volume prediction device according to claim 12.

14. The traffic volume correction unit increases the correction amount as the second difference, obtained by subtracting the traffic volume flowing into the target facility from the traffic volume flowing out of the target facility during the second congestion time, increases. The traffic volume prediction device according to claim 13.

15. Based on the determination that the congestion has reached the target facility in the first congestion data which includes a first congestion time which is the time when congestion occurs and a first congestion location which is the location where congestion occurs at the first congestion time, first traffic state data corresponding to the first congestion data is extracted as training data, A model is generated based on the aforementioned training data and the first traffic volume data at the target facility. Based on the aforementioned model and the second traffic condition data, predict the second traffic volume data at the target facility. A computer-based method for predicting traffic volume, including the method described above.

16. Computers, A learning data extraction unit extracts first traffic state data corresponding to the first traffic data as learning data, based on the determination that the congestion has reached the target facility in first traffic data which includes a first congestion time which is the time when congestion occurs and a first congestion location which is the location where congestion occurs at the first congestion time. A learning unit that generates a model based on the aforementioned learning data and the first traffic volume data at the target facility, A prediction unit predicts second traffic volume data at the target facility based on the aforementioned model and second traffic condition data, A program that makes it function as such.

Images