Information processing device, information processing method, and program
The method improves traffic flow prediction accuracy by using machine learning to adjust spatio-temporal data expansions for real-time traffic conditions, addressing inaccuracies caused by deviations from historical patterns.
Patent Information
- Application Number
- JP2023082116
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-18
- Publication Date
- 2024-11-28
AI Technical Summary
Existing traffic flow prediction technologies struggle with accuracy when traffic jam times and lengths deviate from historical data, leading to inaccurate travel time estimations.
A method that uses machine learning to determine traffic congestion times and predict future traffic densities by expanding spatio-temporal regions based on current traffic densities, adjusting data to account for deviations from historical patterns.
Enhances the accuracy of traffic flow predictions by adapting to real-time changes in traffic conditions, ensuring more precise estimates even when traffic patterns differ from historical data.
Smart Images

Figure 2024165707000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an information processing device, an information processing method, and a program. [Background technology]
[0002] In recent years, various techniques have been known for simulating traffic flow based on sensor data collected by sensors, which may include antennas.
[0003] As an example of a technology for simulating traffic flow, Patent Document 1 discloses, in paragraphs 0017-0019 and 0026-0037, a technology for selecting data from past travel time statistical data that has a similar time fluctuation trend to the travel time actual measurement data, and calculating future travel times based on the selected data.
[0004] As another example of a technology for simulating traffic flow, paragraphs 0032-0038 of Patent Document 2 and other publications disclose a method for obtaining a time-space traffic flow data distribution consisting of a time period from the present time for which prediction is performed back a certain time into the past and small sections on the route, and further obtaining a time-space traffic flow data distribution in the past.The method then discloses a method for extracting a time point in the past with similar traffic conditions by matching these two traffic flow data distributions, accumulating travel times while tracing small sections on the route in sequence, and obtaining an estimated route travel time for the route in the past at a future point. [Prior art documents] [Patent documents]
[0005] [Patent Document 1] JP 2005-196238 A [Patent Document 2] JP 2007-11596 A Summary of the Invention [Problem to be solved by the invention]
[0006] However, the technology described in Patent Document 1 has a problem that when the traffic jam time period or traffic jam length deviates from past data, it is not possible to select data from past data that has a similar tendency of time fluctuation, which deteriorates the accuracy of estimating travel time.Furthermore, the technology described in Patent Document 2 has a problem that it is not possible to extract past traffic conditions similar to the current traffic conditions by matching, which deteriorates the accuracy of estimating travel time.
[0007] Therefore, it is desirable to provide a technique that enables traffic flow to be predicted with higher accuracy. More specifically, it is desirable to predict traffic flow with higher accuracy even when the traffic jam time period or traffic jam length deviates from past data. [Means for solving the problem]
[0008] In order to solve the above problem, according to one aspect of the present invention, an information processing device is provided, comprising: a determination unit that determines whether a first congestion time exists in a first time based on a first average traffic density of a first traffic density at each time belonging to a first time; and a prediction unit that, when it is determined that the first congestion time exists, generates prediction data belonging to the second time based on extending in the time direction a time-space region corresponding to a second traffic density at each time belonging to a second time or a second congestion time of a second average traffic density for each time of the second traffic density, and predicts a third traffic density at a time after the first time based on the prediction data.
[0009] In addition, the third traffic density prediction may use learned model data obtained by machine learning.
[0010] When it is determined that the first congestion time exists, the prediction unit may generate the prediction data based on expanding the time-space domain of the second average traffic density in the time direction.
[0011] The prediction unit may, when the first average traffic density has been increasing since before the first time, expand the time-space region of the second average traffic density in the time direction based on a congestion end time.
[0012] The prediction unit may, when the first average traffic density has been decreasing since before the first time, expand the time-space region of the second average traffic density from a congestion start time to a congestion end time in the time direction based on the congestion start time.
[0013] When it is determined that the first congestion time exists, the prediction unit may generate the prediction data based on expanding the time-space region of the second traffic density in the time direction.
[0014] When it is determined that the first congestion time exists, the prediction unit may generate a distance-direction expanded area based on expanding upstream a first area of the second traffic density that is within the second congestion time and upstream of a specified point, generate a time-direction expanded area based on expanding in the time direction the distance-direction expanded area and a second area of the second traffic density that is within the second congestion time and downstream of the specified point, shift a third area of the second traffic density that corresponds to a time before the first congestion time to a time before the time-direction expanded area, and generate the prediction data based on shifting a fourth area of the second traffic density that corresponds to a time after the second congestion time to a time after the time-direction expanded area.
[0015] The determination unit may determine whether or not the first congestion time exists based on whether or not the first average traffic density is equal to or greater than a predetermined value.
[0016] The prediction unit may generate the prediction data based on expanding the time-space region of the second average traffic density in the time direction when it is determined that the first congestion time exists and when the first average traffic density is smaller than the maximum value of the second average traffic density.
[0017] The prediction unit may generate the prediction data based on expanding the time-space region of the second average traffic density in the time direction when it is determined that the first congestion time exists, and when the first average traffic density is smaller than the maximum value of the second average traffic density, and when the first average traffic density is smaller than or equal to the maximum value of a third average traffic density among the second average traffic densities corresponding to the same time as the time corresponding to the first average traffic density, and larger than or equal to the minimum value of the third average traffic density.
[0018] The prediction unit may generate the prediction data based on expanding the time-space region of the second traffic density in the distance direction and the time direction when it is determined that the first congestion time exists and when the first average traffic density is equal to or greater than a maximum value of the second average traffic density.
[0019] In addition, according to another aspect of the present invention, in order to solve the above-mentioned problem, an information processing method is provided, which includes: determining whether a first congestion time exists in a first time based on a first average traffic density of a first traffic density at each time belonging to a first time; when it is determined that the first congestion time exists, generating prediction data belonging to the second time based on extending in the time direction a time-space region corresponding to a second traffic density at each time belonging to a second time or a second congestion time of a second average traffic density for each time of the second traffic density, and predicting a third traffic density at a time after the first time based on the prediction data.
[0020] In addition, according to another aspect of the present invention, in order to solve the above-mentioned problem, a program is provided that causes a computer to function as an information processing device having a determination unit that determines whether a first congestion time exists in a first time based on a first average traffic density of a first traffic density at each time belonging to a first time, and a prediction unit that, when it is determined that the first congestion time exists, generates prediction data belonging to the second time based on extending in the time direction a time-space region corresponding to a second traffic density at each time belonging to a second time or a second congestion time of a second average traffic density for each time of the second traffic density, and predicts a third traffic density at a time after the first time based on the prediction data. Effect of the Invention
[0021] As described above, according to the present invention, a technique is provided that enables traffic flow to be predicted with higher accuracy. [Brief description of the drawings]
[0022] [Figure 1] 1 is a diagram illustrating an example of a functional configuration of a traffic flow prediction device 1 according to an embodiment of the present invention. [Diagram 2] 2 is a diagram illustrating a detailed configuration example of a learning unit 141. FIG. [Diagram 3] 13 is a diagram illustrating a detailed configuration example of an inference unit 142. FIG. [Figure 4] 13 is a diagram showing an example of master data stored in a master data storage unit 126. FIG. [Diagram 5] 4 is a diagram showing an example of learning parameters stored in a learning parameter storage unit 127. FIG. [Figure 6] 13 is a diagram showing an example of free flow data stored by a free flow data storage unit 123. FIG. [Figure 7] 4 is a diagram showing an example of probe data stored by a probe data storage unit 122. FIG. [Figure 8] FIG. 1 is a diagram showing a set of traffic density (K) and speed (V) and the KV relational expression. [Figure 9]13 is a diagram showing an example of the configuration of KV parameters stored in a learning parameter storage unit 127. FIG. [Figure 10] 13 is a flowchart showing an example of a KV parameter generation process executed by a KV parameter generation unit 1411. [Figure 11] 11 is a diagram for explaining an example of calculation of a passing time at a measurement point of a vehicle that has arrived at an estimation point; FIG. [Figure 12] 11 is a diagram for explaining an example of calculating a traffic volume at an estimation point from a traffic volume at a measurement point; FIG. [Figure 13] 13 is a flowchart showing an example of a traffic density calculation process executed by a traffic density calculation unit 1412. [Figure 14] FIG. 2 is a diagram showing a schematic diagram of traffic density data used for learning. [Figure 15] FIG. 1 is a diagram for explaining an overview of percentile learning. [Figure 16] A figure showing an example of average traffic density percentile data 731 as an example of learning model data 730. [Figure 17] A figure showing an example of mesh traffic density percentile data 732 as an example of learning model data 730. [Figure 18] 13 is a flowchart showing an example of a percentile learning operation by percentile learning unit 1414. [Figure 19] FIG. 2 is a diagram illustrating a traffic density and speed to be predicted. [Figure 20] 13 is a flowchart showing an example of a percentile prediction operation by the inference unit 142 according to the embodiment of the present invention. [Figure 21] 13 is a flowchart showing details of the extension process by the extension processing unit 1422. [Figure 22] 13 is a diagram showing an overview of the space-time expansion process performed by the space-time expansion unit 1424. FIG. [Figure 23] 13 is a flowchart showing details of the time-space expansion process performed by the time-space expansion unit 1424. [Figure 24]13 is a diagram showing an overview of the time extension process performed by a time extension unit 1426. FIG. [Diagram 25] 13 is a flowchart showing details of the time extension process performed by a time extension unit 1426. [Figure 26] 11 is a diagram showing a comparison of actual measurement results of congestion lengths not included in the learning data, prediction results according to a comparative example, and prediction results according to the present embodiment. FIG. [Figure 27] 13 is a diagram showing a comparison of actual measurement results of traffic congestion during time periods not included in the learning data, prediction results according to a comparative example, and prediction results according to the present embodiment. FIG. [Figure 28] FIG. 1 is a diagram showing a hardware configuration of an information processing device 900 as an example of a traffic flow prediction device 1 according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0023] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In this specification and the drawings, components having substantially the same functional configurations are designated by the same reference numerals, and duplicated explanations will be omitted.
[0024] 1. DETAILED DESCRIPTION OF THE EMBODIMENTS The following describes the details of the embodiments of the present invention.
[0025] (1-1. Configuration of the traffic flow prediction device) First, a configuration example of a traffic flow prediction device 1 according to an embodiment of the present invention will be described. Fig. 1 is a diagram showing a functional configuration example of the traffic flow prediction device 1 according to an embodiment of the present invention. The traffic flow prediction device 1 can be an example of an information processing device.
[0026] Referring to FIG. 1, vehicles M1 to M3 are shown as examples of vehicles traveling on a road. Referring further to FIG. 1, vehicles M1 and M2 are traveling in the back lane (vehicle M1 is traveling following vehicle M2), and vehicle M3 is traveling in the front lane in the opposite direction to vehicles M1 and M2 in the back lane. In this way, the embodiment of the present invention mainly assumes a case where the road is made up of multiple lanes, but the road may be made up of a single lane. Each of vehicles M1 to M3 is equipped with an on-board device. Note that a lane can also be referred to as a "lane."
[0027] The traffic flow prediction device 1 includes a driving history data memory unit 121, a probe data memory unit 122, a free flow data memory unit 123, a traffic volume data memory unit 124, a traffic density data memory unit 125, a master data memory unit 126, a learning parameter memory unit 127, and a prediction result memory unit 128.
[0028] The traffic flow prediction device 1 also includes a statistical processing unit 131, a traffic volume calculation unit 133, and a processing unit 140. The processing unit 140 includes a learning unit 141 and an inference unit 142.
[0029] Fig. 2 is a diagram showing a detailed configuration example of the learning unit 141. As shown in Fig. 2, the learning unit 141 includes a KV parameter creation unit 1411, a traffic density calculation unit 1412, and a percentile learning unit 1414. These blocks included in the learning unit 141 mainly operate in the learning stage. Detailed functions of these blocks included in the learning unit 141 will be described later.
[0030] Fig. 3 is a diagram showing a detailed configuration example of the inference unit 142. As shown in Fig. 3, the inference unit 142 includes an extension processing unit 1422. The extension processing unit 1422 also includes a space-time extension unit 1424 and a time extension unit 1426. These blocks included in the inference unit 142 mainly operate in the inference stage. Detailed functions of these blocks included in the inference unit 142 will be described later.
[0031] The driving history data storage unit 121 is connected to the probe antenna 112 , and the free flow data storage unit 123 is connected to the free flow antenna 114 .
[0032] The statistical processing unit 131, the traffic volume calculation unit 133, and the processing unit 140 each include a calculation device such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit), and the functions of each unit can be realized by the calculation device expanding a program stored in a ROM (Read Only Memory) into a RAM and executing the program. At this time, a computer-readable recording medium on which the program is recorded can also be provided.
[0033] Alternatively, the statistical processing unit 131, the traffic volume calculation unit 133, and the processing unit 140 may be configured by dedicated hardware, or may be configured by a combination of multiple hardware. Data necessary for the calculation by the calculation device is appropriately stored in a storage unit (not shown).
[0034] 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 traffic density data storage unit 125, the master data storage unit 126, the learning parameter storage unit 127, and the prediction result storage unit 128 are realized by a storage unit (not shown). Such a storage unit may be configured by a memory such as a RAM (Random Access Memory), a hard disk drive, or a flash memory.
[0035] (Master data storage unit 126) The master data storage unit 126 stores various master data in advance. The master data is data input by a user to the traffic flow prediction device 1 and managed by the traffic flow prediction device 1. Here, an example of the master data will be described with reference to FIG. 4.
[0036] Fig. 4 is a diagram showing an example of master data stored by the master data storage unit 126. As shown in Fig. 4, the master data storage unit 126 stores, as examples of master data, a driving history time width 611, a driving history section width 612, a simulation execution time interval 621, a simulation execution cell width 622, a congestion flow determination speed 641, and a congestion wave propagation speed 645. Details of these master data will be described later.
[0037] (Learning parameter storage unit 127) The learning parameter storage unit 127 can store various learning parameters. The learning parameters are parameters obtained by learning, and are stored in the learning parameter storage unit 127 in the learning stage. Also, the learning parameters are obtained from the learning parameter storage unit 127 in the inference stage. Here, examples of the learning parameters stored by the learning parameter storage unit 127 will be described with reference to FIG. 5.
[0038] Fig. 5 is a diagram showing examples of learning parameters stored by the learning parameter storage unit 127. As shown in Fig. 5, the learning parameter storage unit 127 may store KV parameters 710 and learning model data 730 as examples of learning parameters. Details of these learning parameters will be described later.
[0039] (Free Flow Antenna 114) The free flow antenna 114 functions as an example of a vehicle detection unit that detects vehicles traveling at each position on a road. That is, in the embodiment of the present invention, a case is mainly assumed in which the vehicle detection unit includes the free flow antenna 114. As a result, an ETC free flow antenna of an already constructed ETC (Electronic Toll Collection) system can be used as a vehicle detection unit, so there is no need to provide a new vehicle detection unit. However, instead of the free flow antenna 114, other vehicle detection units (for example, a vehicle detector, an infrared sensor, or an ultrasonic sensor) may be used.
[0040] More specifically, the free flow antenna 114 detects a vehicle in real time by receiving vehicle identification information (vehicle ID) from the vehicle through communication with an on-board unit mounted on the vehicle. In the embodiment of the present invention, a case is mainly assumed in which an ETC on-board unit is used as an example of the on-board unit. It is assumed that the free flow antenna 114 is compatible with a plurality of versions of ETC on-board units, whereas the probe antenna 112 described later is compatible with only a specific version of ETC on-board units. In other words, the free flow antenna 114 can detect more vehicles than the probe antenna 112.
[0041] In addition, the "positions" on the road where the vehicle is detected by the free flow antenna 114 are not particularly limited as long as they are vehicle positions that can be detected by the vehicle detection unit. In the following, each position on the road where the vehicle can be detected by the free flow antenna 114 may be simply referred to as a "free flow antenna position."
[0042] When the free flow antenna 114 detects a vehicle in real time by receiving a vehicle ID through communication with an on-board device mounted on the vehicle, it outputs the vehicle detection result (hereinafter also referred to as "free flow data") in real time to the free flow data storage unit 123. Then, the vehicle detection result output in real time from the free flow antenna 114 to the free flow data storage unit 123 can be used in real time by the traffic volume calculation unit 133.
[0043] Here, "real-time" may mean any short time between when a vehicle arrives at the free flow antenna position and when the traffic conditions on the road change. If a vehicle is detected at this time, some measure can be taken according to the traffic conditions at the time when the vehicle is detected before the traffic conditions on the road change.
[0044] (Free flow data storage unit 123) FIG. 6 is a diagram showing 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 through the free flow antenna position. As shown in FIG. 6, the free flow data is made up of a "vehicle ID" and a "passing time" associated with each other. The "vehicle ID" is vehicle identification information received by the free flow antenna 114 from the vehicle through communication between the free flow antenna 114 and an onboard device mounted on the vehicle. The "passing time" is the time when the vehicle ID is received from the vehicle by the free flow antenna 114, and may correspond to the time when the vehicle passed the free flow antenna position.
[0045] (Traffic volume calculation unit 133) The traffic volume calculation unit 133 acquires the free flow data from the free flow data storage unit 123. Then, based on the free flow data (vehicle IDs and passing times), the traffic volume calculation unit 133 calculates the number of vehicles that pass the free flow antenna position per unit time set in advance (in other words, the number of vehicle IDs detected per unit time by the free flow antenna 114) as the traffic volume at the free flow antenna position. The traffic volume calculation by the traffic volume calculation unit 133 may be repeatedly performed for each unit time.
[0046] Note that not all vehicles that pass the free flow antenna position are necessarily equipped with an on-board device capable of communicating with the free flow antenna 114. Therefore, not all vehicles that pass the free flow antenna position are necessarily detected by the free flow antenna 114. Therefore, if the ratio of vehicles equipped with on-board devices capable of communicating with the free flow antenna 114 to the total number of vehicles (including vehicles not equipped with on-board devices capable of communicating with the free flow antenna 114) is preset as the "on-board device installation ratio" and the number of vehicles detected by the free flow antenna 114 is set as the "number of detected vehicles," it is desirable for the traffic volume calculation unit 133 to more accurately estimate the traffic volume at the free flow antenna position using the following formula (1).
[0047] (Estimated traffic volume) = (Number of detected vehicles) ÷ (Proportion of vehicles equipped with on-board devices) (1)
[0048] However, such estimation may be omitted in cases where in-vehicle devices capable of communicating with the free flow antenna 114 are sufficiently widespread. The traffic volume estimated by the traffic volume calculation unit 133, the measurement time which is the end of the time corresponding to the traffic volume, and the free flow antenna position corresponding to the traffic volume are output to the traffic volume data storage unit 124 as traffic volume data every time the traffic volume is estimated.
[0049] (Traffic volume data storage unit 124) The traffic volume data storage unit 124 stores the traffic volume data output from the traffic volume calculation unit 133 .
[0050] (Probe Antenna 112) The probe antenna 112 functions as an example of a driving history data acquisition unit that acquires driving history data of a vehicle. That is, in the embodiment of the present invention, a case is mainly assumed in which the driving history data acquisition unit includes the probe antenna 112. As a result, an 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 provide a new driving history data acquisition unit. However, instead of the probe antenna 112, another driving history data acquisition unit (for example, a mobile base station, etc.) may be used.
[0051] More specifically, when the probe antenna 112 acquires driving history data from the vehicle through communication with an on-board device mounted on the vehicle, the probe antenna 112 outputs the acquired driving history data to the driving history data storage unit 121 .
[0052] (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 by the driving history data storage unit 121 can be used by the statistical processing unit 131.
[0053] The driving history data includes the speed of each vehicle traveling for each section (between kilometer posts) on the road defined by the driving history section width 612 (e.g., 100 m) and for each driving history time width 611 (e.g., 1 minute), and the collection time when the speed was collected by the probe antenna 112.
[0054] (Statistical processing unit 131) The statistical processing unit 131 acquires the driving history data from the driving history data storage unit 121, performs statistical processing on the driving history data, and outputs the driving history data after the statistical processing 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 be mainly used by the learning unit 141.
[0055] For example, the statistical processing unit 131 performs a predetermined statistical process (e.g., averaging process, etc.) on the speed of one or more vehicles that have traveled for each section (between kilometer posts) on the road and for each driving history time range 611. This provides statistics on the vehicle speed for each section on the road and for each driving history time range 611. An example of the averaging process is a process of taking a harmonic mean.
[0056] (Probe data storage unit 122) Fig. 7 is a diagram showing an example of the probe data stored by the probe data storage unit 122. As shown in Fig. 7, the probe data includes a "passing time", a "kilometer post", a "speed", and a "collection time" that are associated with each other.
[0057] "Passing time" is the end time when a vehicle passes a distance marker (kilometer post) from the starting point of a road. For example, "Passing time: 2019-01-01 10:24" indicates that the end time when a vehicle passes a distance marker (kilometer post) is "2019-01-01 10:24." A "kilometer post" is a distance marker from the starting point of a road.
[0058] "Speed" is a statistic (e.g., average speed) of one or more vehicle speeds that traveled each section (between kilometer posts) on the road for each travel history time span 611 (between passing times). For example, "Speed: 80 km / h" is a statistic of the speed of one or more vehicles that traveled from "Kilopost: 100" to "Kilopost: 100.1" from "Passing Time: 2019-01-01 10:23" to "2019-01-01 10:24".
[0059] The “collection time” is the time when the vehicle speed was collected by the probe antenna 112 .
[0060] (KV parameter creation unit 1411) The KV parameter creation unit 1411 acquires, for each predetermined period, the speed and passing time for each free flow antenna position from the probe data storage unit 122. Furthermore, the KV parameter creation unit 1411 acquires, for each predetermined period, traffic volume data (traffic volume and measurement time) for each free flow antenna position from the traffic volume data storage unit 124.
[0061] The KV parameter creation unit 1411 creates parameters indicating the correspondence between traffic density (K) and speed (V) (in the following example, an approximation equation) for each free flow antenna position through machine learning based on the traffic volume and measurement time for a specified period of time, and the speed and passing time for the specified period of time.
[0062] More specifically, the KV parameter creation unit 1411 associates the traffic volume (Q) and the speed (V) corresponding to the same time and the same free flow antenna position. Then, the KV parameter creation unit 1411 calculates the traffic density (K) for each associated traffic volume (Q) and speed (V) by the following formula (2).
[0063] K = Q × 1 / (V × 60) (2)
[0064] where K indicates traffic density (vehicles / km), Q indicates traffic volume (vehicles / min), and V indicates speed (km / hour). The KV parameter creation unit 1411 approximates a pair of corresponding traffic density (K) and speed (V) in a predetermined relationship, and creates the parameters of the approximation equation (KV relational equation) obtained by the approximation as KV parameters.
[0065] Here, it is assumed that the KV parameter creation unit 1411 creates parameters of an approximation equation corresponding to a congestion flow. More specifically, the KV parameter creation unit 1411 creates parameters of a KV relational equation corresponding to a congestion flow by approximating a pair of a speed (V) equal to or lower than the congestion flow determination speed 641 (e.g., 55 km / h on a highway) and a traffic density (K) corresponding to that speed (V) in a predetermined relationship. For example, the KV relational equation corresponding to a congestion flow may be an exponential function expressed by the following equation (3).
[0066] V = a × exp(-b × K) (3)
[0067] where K indicates traffic density (vehicles / km), V indicates speed (km / hour), and a and b are KV parameters. In this specification, congested flow is used as a concept that is included in jammed flow, and can mean a state in which the speed is greater than a certain value even in a jammed flow.
[0068] Fig. 8 is a diagram showing pairs of traffic density (K) and speed (V) and the KV relational expression. In the example shown in Fig. 8, the horizontal axis indicates traffic density (K) and the vertical axis indicates speed (V). In the KV diagram, each pair of traffic density (K) and speed (V) corresponding to each other is plotted as each point. Also, referring to Fig. 8, an exponential function showing the KV relational expression obtained by approximating these pairs is shown.
[0069] The KV parameter generating unit 1411 stores the generated KV parameters in the learning parameter storage unit 127 as the KV parameters 710. Typically, the predetermined period may be one month. However, the predetermined period is not limited to one month. For example, the predetermined period may be one day.
[0070] Fig. 9 is a diagram showing an example of the configuration of KV parameters stored in the learning parameter storage unit 127. As shown in Fig. 9, the KV parameters include "parameter a" and "parameter b." Furthermore, since "parameter a" and "parameter b" are expected to differ depending on the road shape, the number of lanes, etc., it is desirable to create them for each vehicle detection position (each free flow antenna position).
[0071] Therefore, as shown in Fig. 9, it is preferable that "parameter a" and "parameter b" are associated with a "route code" and a "vehicle detection position ID." The "route code" is a code for identifying a route (a road from a starting point to an end point). The vehicle detection position ID is a code for identifying a free flow antenna position.
[0072] Furthermore, standard KV parameters according to the number of lanes may be further stored in the learning parameter storage unit 127. For example, when the number of lanes at the free flow antenna position is acquired, the standard KV parameters according to the number of lanes may be used as the initial values of the KV parameters corresponding to the free flow antenna position.
[0073] 10 is a flowchart showing an example of a KV parameter creation process executed by the KV parameter creation unit 1411. First, the KV parameter creation unit 1411 acquires speeds and passing times for a predetermined period for each free flow antenna position from the probe data stored in the probe data storage unit 122 (S11). In addition, the KV parameter creation unit 1411 acquires traffic volume data (traffic volume and measurement time) for a predetermined period for each free flow antenna position from the traffic volume data storage unit 124 (S12).
[0074] The KV parameter creation unit 1411 associates the traffic volume (Q) and the speed (V) corresponding to the same time and the same free flow antenna position. Then, the KV parameter creation unit 1411 calculates the traffic density (K) for each associated traffic volume (Q) and speed (V) (S13).
[0075] The KV parameter creation unit 1411 creates parameters of a KV relational equation that indicates the relationship between traffic density (K) and speed (V) by machine learning (S14). More specifically, the KV parameter creation unit 1411 approximates a pair of corresponding traffic density (K) and speed (V) with a predetermined relationship (e.g., an exponential function, etc.), and creates the parameters of the KV relational equation obtained by the approximation as KV parameters. Here, it is assumed that the KV parameter creation unit 1411 creates parameters of an approximation equation that corresponds to a congestion flow.
[0076] The KV parameter generating unit 1411 stores the generated KV parameters as the KV parameters 710 in the learning parameter storage unit 127 (S15).
[0077] (Traffic density calculation unit 1412) The traffic density calculation unit 1412 acquires the speed and passing time for each section stored at a time later than the previous acquisition time of the probe data used to calculate the traffic density at a preset simulation execution time interval 621 (e.g., 5-minute intervals) from the probe data storage unit 122. In addition, the traffic density calculation unit 1412 acquires the traffic volume data (traffic volume and measurement time) for each free-flow antenna position stored at a time later than the previous acquisition time of the traffic volume data used to calculate the traffic density at the simulation execution time interval 621 from the traffic volume data storage unit 124.
[0078] The traffic density calculation unit 1412 calculates the traffic density for each section based on at least the speed. Here, the traffic density for each section may be calculated in any specific manner.
[0079] For example, the traffic density calculation unit 1412 may obtain KV parameters from the learning parameter memory unit 127 and calculate the traffic density in a target section on a road based on the speed in the target section and the KV relational equation defined by the KV parameters.
[0080] More specifically, the traffic density calculation unit 1412 may calculate the traffic density in a section that is a congested flow, that is, a target section where the speed is equal to or lower than the congested flow determination speed 641 (threshold value), based on the speed in the target section and the KV relational equation.
[0081] In addition, the traffic density calculation unit 1412 may set a point on the road corresponding to the target section as an estimated point, and calculate the traffic density in the target section based on the traffic volume at the estimated point and the speed in the target section. For example, the estimated point may be the starting point of the target section.
[0082] More specifically, the traffic density calculation unit 1412 may calculate the traffic density in the target section where the speed exceeds the congested flow judgment speed 641 based on the traffic volume at the estimation point and the speed in the target section. For example, the traffic density calculation unit 1412 may calculate the traffic volume at the estimation point based on the traffic volume at the measurement point calculated by the traffic volume calculation unit 133 and the speed for each section from the measurement point to the target section. For example, the measurement point may be a free flow antenna position upstream from the target section (for example, the free flow antenna position closest upstream from the target section).
[0083] An example of a method for calculating the traffic density in a target section based on the traffic volume at an estimation point and the speed in the target section will be described with reference to Figs.
[0084] Fig. 11 is a diagram for explaining an example of calculation of the passing time at a measurement point of a vehicle that has arrived at an estimated point. In the example shown in Fig. 11, the horizontal axis indicates distance on the road (the right direction is the downstream direction on the road), and the vertical axis indicates time (the downward direction is the direction of time passage). The speed corresponding to a section (between kilometer posts) and unit time (i.e., the time from a passing time to the passing time immediately before that passing time) in the probe data is indicated by the darkness of the color within a rectangle (space-time range) corresponding to that section and unit time. In this example, the darker the color within the rectangle, the lower the speed.
[0085] The traffic density calculation unit 1412 calculates a trajectory from the estimated point and the passing time at the estimated point to the measurement point upstream and in the past, based on the speed corresponding to each section and unit time. More specifically, the traffic density calculation unit 1412 calculates a trajectory from the estimated point and the time at the estimated point so that the vehicle moves within the rectangle along a straight line with a slope according to the speed corresponding to each section and unit time, and moves between adjacent rectangles at the boundary between the rectangles.
[0086] In this way, the traffic density calculation unit 1412 can obtain the time at the measurement point where the trajectory arrives as the passing time at the measurement point. Referring to Fig. 11, the trajectories of two vehicles are illustrated. Here, the traffic volume between the arrival times of the two vehicles at the estimated point and the traffic volume between the passing times of the two vehicles at the measurement point can be considered to be the same.
[0087] Fig. 12 is a diagram for explaining an example of calculating the traffic volume at the estimation point from the traffic volume at the measurement point. Referring to Fig. 12, the traffic volume Q between the arrival times at the estimation point and the traffic volume Q' between the passing times at the measurement point are shown. Here, the traffic volume between the passing times at the measurement point is stored in the traffic volume data storage unit 124. Therefore, the traffic density calculation unit 1412 can obtain the traffic volume Q' between the passing times at the measurement point from the traffic volume data storage unit 124.
[0088] The traffic density calculation unit 1412 may calculate the traffic volume at the estimation point from the traffic volume Q' between the passing times at the measurement point by the ratio of the arrival time difference at the estimation point to the passing time difference at the measurement point. For example, if the passing time difference at the measurement point is t1 and the arrival time difference at the estimation point is t2, the traffic density calculation unit 1412 may calculate the traffic volume Q at the estimation point by multiplying the traffic volume Q' between the passing times at the measurement point by t1 / t2. Hereinafter, the method of calculating the traffic volume Q at the estimation point from the traffic volume Q' at the measurement point in this way is also referred to as a traffic volume calculation method by "vehicle tracking".
[0089] 13 is a flowchart showing an example of a traffic density calculation process executed by the traffic density calculation unit 1412. First, the traffic density calculation unit 1412 acquires, from the probe data storage unit 122, probe data (speed and passing time for each section) stored at a time later than the previous acquisition time of the probe data used to calculate the traffic density, during a preset simulation execution time interval 621 (S21).
[0090] In addition, the traffic density calculation unit 1412 acquires traffic volume data (traffic volume and measurement time) for each free flow antenna position that is stored at a time later than the previous acquisition time of the traffic volume data used to calculate the traffic density during the simulation execution time interval 621 from the traffic volume data storage unit 124 (S22). The traffic density calculation unit 1412 determines whether the speed in the target section is the speed of a congested flow, that is, a speed equal to or lower than the congested flow determination speed 641 (S23).
[0091] If the speed in the target section is the congestion flow speed (i.e., a speed that is equal to or lower than the congestion flow judgment speed 641) ("YES" in S23), the traffic density calculation unit 1412 calculates the traffic density in the target section based on the speed in the target section and the KV relational equation (S24).
[0092] On the other hand, if the speed in the target section is the free flow speed (i.e., a speed higher than the congested flow judgment speed 641) ("NO" in S23), the traffic density calculation unit 1412 calculates the traffic volume at the measurement point by vehicle tracking, and calculates the traffic density in the target section based on the calculated traffic volume at the measurement point and the speed of each section from the measurement point to the target section (S25). The traffic density calculation unit 1412 outputs traffic density data in which the measurement time of the traffic volume, the section, and the traffic density are associated with each other to the traffic density data storage unit 125.
[0093] (Traffic density data storage unit 125) The traffic density data storage unit 125 stores the traffic density data output from the traffic density calculation unit 1412.
[0094] (Traffic density for study) In the above manner, the traffic density data in which the time, the section, and the traffic density are associated with each other is stored. In the following learning, the traffic density data in which the time, the cell, and the traffic density are associated with each other is used.
[0095] Here, the width of the section is defined by the driving history section width 612. On the other hand, the width of the cell is defined by the simulation execution cell width 622.
[0096] The width of the cell may be the same as the width of the section, or may be different from the width of the section. For example, the width of the section may be 100 m, whereas the width of the cell may be 500 m. As in this example, when the width of the cell is different from the width of the section, the learning unit 141 may identify a plurality of sections belonging to the cell, perform statistical processing on the traffic density corresponding to each of the plurality of sections, and obtain the traffic density after the statistical processing as the traffic density corresponding to the cell.
[0097] The time interval corresponding to the section is defined in a driving history time span 611. On the other hand, the time interval corresponding to the cell is defined in a simulation execution time interval 621.
[0098] The time interval corresponding to the cell may be the same as the time interval corresponding to the section, or may be different from the time interval corresponding to the section. For example, the time interval corresponding to the section may be 1 minute, whereas the time interval corresponding to the cell may be 5 minutes. As in this example, when the time interval corresponding to the cell is different from the time interval corresponding to the section, the learning unit 141 may identify the time corresponding to the section that belongs to the time corresponding to the cell, perform statistical processing on the traffic density corresponding to the identified time, and obtain the traffic density after statistical processing as the traffic density for the time corresponding to the cell.
[0099] FIG. 14 is a diagram showing a schematic diagram of traffic density data used for learning. Referring to FIG. 14, cells 1 to n are shown from the upstream side to the downstream side of a road. The upstream end of cell 1 is the start point of the road, and the downstream end of cell n is the end point of the road. i is an arbitrary integer satisfying 1≦i≦N, and cell i is a target cell. Δt is a time interval corresponding to a cell. m is an integer equal to or greater than 1.
[0100] Also, referring to FIG. 14, the traffic densities K1(t) to K n (t) (vehicles / km) is shown. For example, the time t may be the current time, but the time t is not limited to the current time. Similarly, the traffic densities K1(t-Δt) to K n (t-Δt) is shown, and the traffic density K1(t-mΔt)~K n (t-mΔt) is shown.
[0101] In the learning by the learning unit 141, traffic density data in which times t to t-mΔt, cells 1 to N, and traffic density K are associated with each other is used. As an example of the learning by the learning unit 141, percentile learning by the percentile learning unit 1414 can be given. Such percentile learning will be described below.
[0102] (Percentile Learning Section 1414) The percentile learning unit 1414 functions as an example of a learning unit and performs percentile learning. Here, an overview of percentile learning will be described with reference to Figs.
[0103] FIG. 15 is a diagram for explaining an overview of percentile learning. Referring to FIG. 15, a speed corresponding to a time and a distance from a starting point is indicated by the color density of a rectangular area corresponding to the time and the distance from the starting point. The distance from the starting point may correspond to a cell. In the following, the rectangular area (space-time range) corresponding to the time and the distance from the starting point is also called a "mesh."
[0104] For example, according to percentile learning, a time change G1 in traffic density at a distance C1 from the origin is stored in the learning parameter storage unit 127 as an example of learning model data 730. Similarly, a time change G2 in traffic density at a distance C2 from the origin is stored in the learning parameter storage unit 127 as an example of learning model data 730, and a time change G3 in traffic density at a distance C3 from the origin is stored in the learning parameter storage unit 127 as an example of learning model data 730.
[0105] Fig. 16 is a diagram showing an example of average traffic density percentile data 731 as an example of the learning model data 730. As shown in Fig. 16, the average traffic density percentile data 731 is configured by associating "point-to-point distance", "time", "average traffic density", and "average traffic density learning percentile value". Each of these items included in the average traffic density percentile data 731 will be described with reference to Fig. 18.
[0106] Fig. 17 is a diagram showing an example of mesh traffic density percentile data 732 as an example of the learning model data 730. As shown in Fig. 17, the mesh traffic density percentile data 732 is configured by associating "cell", "time", "traffic density", and "mesh traffic density learning percentile value". Each of these items included in the mesh traffic density percentile data 732 will be described with reference to Fig. 18.
[0107] 18 is a flowchart showing an example of the operation of percentile learning by percentile learning unit 1414. An example of the operation of percentile learning by percentile learning unit 1414 will be described with reference to FIG.
[0108] First, when the current time (third time) reaches a preset learning time, the percentile learning unit 1414 acquires traffic density data for a predetermined target time period based on the current time from the traffic density data storage unit 125 as new traffic density data (S31). The new traffic density data is traffic density data that has not yet been used for learning. Note that the reference third time does not have to be the current time. The reference third time may be a past time.
[0109] For example, the study time may be 2:00 every Monday. In this case, the predetermined target time may be one week. However, the specific values of the study time and the predetermined target time are not limited. In the following, it is mainly assumed that the predetermined target time based on the current time is a time that starts from a time that is the predetermined target time before the current time and ends at the current time. However, the predetermined target time based on the current time is not limited to such a time.
[0110] The percentile learning unit 1414 calculates the average traffic density for a predetermined processing time width between two predetermined points at a predetermined processing time interval based on new traffic density data acquired from the traffic density data storage unit 125 (S32). The percentile learning unit 1414 stores the average traffic density in the learning parameter storage unit 127 in association with the "point-to-point" and "time" corresponding to the traffic density used to calculate the average traffic density.
[0111] For example, the distance between the points may be between two interchanges. However, the distance between the two points is not limited to this example. The processing time range may be a time range from a time that precedes the measurement time by a predetermined time range (e.g., 30 minutes) to the measurement time. However, the processing time range is not limited to this example. The processing time interval may be a 5-minute interval. However, the processing time interval is not limited to this example.
[0112] In the average traffic density percentile data 731 (FIG. 16), "cells h to j" to which the target cell i belongs are recorded in the "point-to-point" field as an example of a cell corresponding to the traffic density used to calculate the average traffic density. In addition, in the "time" field, the current time t is recorded as an example of a time corresponding to the traffic density used to calculate the average traffic density. In addition, in the "average traffic density", "32 vehicles / km" and "40 vehicles / km" are recorded.
[0113] The percentile learning unit 1414 classifies the average traffic densities so that the average traffic densities between the same points, at the same time, and of the same type of day to which the time belongs are grouped together. The type of day may be the day of the week, or may be a day of the week attribute (e.g., weekday, day before a holiday, first day of a three-day holiday, etc.). The percentile learning unit 1414 then calculates the percentile value of each average traffic density relative to the total average traffic densities for each group (hereinafter also referred to as the "average traffic density learning percentile value") (S34).
[0114] More specifically, the percentile learning unit 1414 sorts all the average traffic densities for each group in ascending order of average traffic density. Then, the percentile learning unit 1414 assigns an average traffic density learning percentile value to each average traffic density for each group in ascending order of average traffic density. For example, the 0th percentile value is assigned to the smallest average traffic density, the 100th percentile value to the largest average traffic density, and the 50th percentile value to the median average traffic density.
[0115] The percentile learning unit 1414 associates the average traffic density learning percentile value with the point-to-point distance, time, and average traffic density, and stores the association in the learning parameter storage unit 127 (S35).
[0116] In the average traffic density percentile data 731 (FIG. 16), a percentile value of "41" for "32 vehicles / km" is recorded as an example of an "average traffic density learning percentile value" corresponding to "average traffic density" = "32 vehicles / km". Also, a percentile value of "42" for "40 vehicles / km" is recorded as an example of an "average traffic density learning percentile value" corresponding to "average traffic density" = "40 vehicles / km".
[0117] Furthermore, the percentile learning unit 1414 classifies each traffic density so that traffic densities in the same cell and for the same type of day to which the time belongs are grouped together based on new traffic density data acquired from the traffic density data storage unit 125. Then, the percentile learning unit 1414 calculates the percentile value of each traffic density relative to the total traffic density (hereinafter also referred to as the "mesh traffic density learning percentile value") for each group (S36).
[0118] The percentile learning unit 1414 associates the point-to-point distance, time, average traffic density, and average traffic density learning percentile value with the cell and mesh traffic density learning percentile value, and stores them in the learning parameter storage unit 127.
[0119] In the mesh traffic density percentile data 732 (FIG. 17), a percentile value of "38" for "34 vehicles / km" is recorded as an example of a "mesh traffic density learning percentile value" corresponding to "traffic density" = "34 vehicles / km". Also, a percentile value of "40" for "38 vehicles / km" is recorded as an example of a "mesh traffic density learning percentile value" corresponding to "traffic density" = "38 vehicles / km".
[0120] (Traffic density and speed to be predicted) As described above, percentile learning generates the learning model data 730. The inference unit 142 can predict the traffic density and speed at a time after the current time by inference using this learning model data 730.
[0121] FIG. 19 is a diagram showing traffic density and speed to be predicted. As shown in FIG. 19, similar to the diagram showing traffic density data used for learning (FIG. 14), traffic densities K1(t) to K n (t) (vehicles / km) is shown. For example, the time t may be the current time, but the time t is not limited to the current time.
[0122] Similarly, the traffic density K1(t+Δt)~K of cells 1~n at time t+Δt n (t+Δt) is shown, and the traffic density K1(t+MΔt)~K n (t+MΔt) is shown.
[0123] In the inference by the inference unit 142, traffic flow characteristics corresponding to times t+Δt to t+MΔt (fourth time) and cells 1 to N are predicted. The following describes the prediction of such traffic flow characteristics. Note that at least one of traffic density K and speed V is given as an example of the traffic flow characteristics.
[0124] (Summary of prediction by inference unit 142) Next, an overview of the prediction by the inference unit 142 will be described. The inference unit 142 predicts traffic flow characteristics corresponding to times t+Δt to t+MΔt and cells 1 to N, based on the learning model data 730 generated as described above.
[0125] (Inference part 142) 20 is a flowchart showing an example of the operation of percentile prediction by the inference unit 142 according to the embodiment of the present invention. With reference to FIG. 20, an example of the operation of percentile prediction by the inference unit 142 according to the embodiment of the present invention will be described.
[0126] First, the inference unit 142 acquires probe data for a predetermined target time period (first time period) based on the current time (first time period) from the probe data storage unit 122 as the latest probe data. Furthermore, the inference unit 142 acquires traffic volume data for a predetermined target time period based on the current time from the traffic volume data storage unit 124 as the latest traffic volume data (S51). Note that the first time period used as the reference does not have to be the current time. The first time period used as the reference may be a time in the past.
[0127] For example, the predetermined target time may be one hour. However, the specific value of each predetermined target time is not limited. In the following, the predetermined target time based on the current time is mainly assumed to be a time that starts from a time that is the predetermined target time before the current time and ends at the current time. However, the predetermined target time based on the current time is not limited to such a time.
[0128] Next, the inference unit 142 calculates the traffic density of each cell between two preset points based on the acquired latest probe data and the latest traffic volume data (S52). Note that the traffic density calculation method described with reference to FIG. 13 can be applied to the calculation of the traffic density here.
[0129] Next, the inference unit 142 calculates an average traffic density (first average traffic density) for a predetermined processing time span between points based on the traffic density for a predetermined target time span based on the current time. In the embodiment of the present invention, the inference unit 142 calculates the average traffic density for the processing time span according to the current time (S53).
[0130] For example, the processing time range may be a time range from a time preceding the current time by a predetermined time range (e.g., 30 minutes) to the current time, but the processing time range is not limited to this example.
[0131] Next, the inference unit 142 executes the extension process by the extension processing unit 1422 (S54). This extension process updates the average traffic density and traffic density corresponding to the time that coincides with the current time among the average traffic density and traffic density of the learning model data 730 stored by the learning parameter storage unit 127. The details of the extension process by the extension processing unit 1422 will be described in detail later with reference to Figs. 21 to 25.
[0132] Next, the inference unit 142 predicts the traffic density at a time after the current time based on the calculated average traffic density and the average traffic density, average traffic density learning percentile value, and mesh traffic density learning percentile value stored by the learning parameter memory unit 127.
[0133] For example, the inference unit 142 acquires from the learning parameter storage unit 127 the average traffic density associated with the time that coincides with the reference time and associated with the same point-to-point relationship as the point-to-point relationship for which the average traffic density was calculated.
[0134] Alternatively, the inference unit 142 may obtain from the learning parameter memory unit 127 an average traffic density that is associated with a time that matches the current time and is further associated with a time that belongs to the same type of day as the type of day to which the current time belongs.
[0135] Then, the inference unit 142 obtains a predicted percentile value based on the calculated average traffic density and the acquired average traffic density (S55).
[0136] More specifically, the inference unit 142 may specify, as the predicted percentile value, the average traffic density learned percentile value associated with the acquired average traffic density that is closest to the calculated average traffic density.
[0137] Alternatively, the inference unit 142 may calculate the acquired average traffic density that is greater than and closest to the calculated average traffic density as the upper average traffic density. Furthermore, the inference unit 142 may calculate the acquired average traffic density that is less than and closest to the calculated average traffic density as the lower average traffic density. Then, the inference unit 142 may calculate the predicted percentile value by linear interpolation based on the upper average traffic density and the lower average traffic density.
[0138] The inference unit 142 acquires from the learning parameter memory unit 127 the time, cell, and mesh traffic density learning percentile value associated with the same location-to-location pair for which the average traffic density was calculated from the current time to a predetermined set time later (e.g., two hours later).
[0139] Alternatively, the inference unit 142 may acquire from the learning parameter memory unit 127 a time, cell, and mesh traffic density learning percentile value that is associated with the same location-to-location pair as the location-to-location pair for which the average traffic density was calculated from the current time until a predetermined set time later (e.g., two hours later), and that is associated with a time that belongs to the same type of day as the type of day to which the current time belongs.
[0140] Based on the time, cell, and mesh traffic density learning percentile value obtained from the learning parameter memory unit 127, the inference unit 142 obtains a mesh traffic density learning percentile value corresponding to the predicted percentile value for each cell of the simulation execution cell width at the simulation execution time interval.
[0141] For example, the inference unit 142 may obtain the mesh traffic density learned percentile value that is closest to the predicted percentile value.
[0142] Alternatively, the inference unit 142 may calculate a mesh traffic density learning percentile value that is greater than the predicted percentile value and closest to the predicted percentile value as the upper mesh traffic density learning percentile value. Furthermore, the inference unit 142 may calculate a mesh traffic density learning percentile value that is smaller than the predicted percentile value and closest to the predicted percentile value as the lower mesh traffic density learning percentile value. Then, the inference unit 142 may calculate the predicted percentile value by linear interpolation based on the average traffic density on the upper side and the average traffic density on the lower side or linear interpolation by the harmonic mean.
[0143] The inference unit 142 functions as an example of a prediction unit, and obtains the traffic density associated with the mesh traffic density learning percentile value thus obtained as a predicted traffic density for each cell of the simulation execution cell width and for the simulation execution time interval. The inference unit 142 may also calculate a predicted speed corresponding to the predicted traffic density for each cell of the simulation execution cell width and for the simulation execution time interval (S56). For example, the inference unit 142 may calculate the predicted speed corresponding to the predicted traffic density using the KV relational expression shown by the above formula (3).
[0144] The inference unit 142 outputs the prediction result to the prediction result storage unit 128 (S57).
[0145] (Extension processing unit 1422) Fig. 21 is a flowchart showing details of the extension process by the extension processing unit 1422. Note that this extension process is called from S54 (Fig. 20) in the percentile prediction by the inference unit 142. Details of the extension process by the extension processing unit 1422 will be described with reference to Fig. 21.
[0146] The extension processing unit 1422 acquires the average traffic density calculated in S53 (FIG. 20) and also acquires the average traffic density from the learning model data 730. The extension processing unit 1422 functions as an example of a determination unit, and determines whether or not a traffic jam time (first traffic jam time) exists based on the calculated average traffic density. More specifically, the extension processing unit 1422 determines whether or not a traffic jam time exists based on whether or not the calculated average traffic density is equal to or greater than a predetermined value.
[0147] As an example, the predetermined value is a predetermined threshold value. That is, the extension processing unit 1422 determines whether or not a congestion time exists based on whether or not the calculated average traffic density is equal to or greater than a threshold value (S61). However, the predetermined value is not limited to such an example. For example, the predetermined value may be an average traffic density calculated from a free flow traffic density among the average traffic densities acquired from the learning model data 730.
[0148] If the calculated average traffic density is less than the threshold value ("NO" in S61), the extension processing unit 1422 determines that there is no congestion time and ends the process. On the other hand, if the calculated average traffic density is equal to or greater than the threshold value ("YES" in S61), the extension processing unit 1422 determines whether the calculated average traffic density is less than the maximum value of the average traffic density at all times acquired from the learning model data 730 (S62).
[0149] If the calculated average traffic density is equal to or greater than the maximum value of the average traffic density at all times acquired from the learning model data 730 ("NO" in S62), it is considered that the prediction accuracy will not improve unless the extension in the distance direction is also performed. In such a case, the extension processing unit 1422 performs the time-space extension process (S64) by the time-space extension unit 1424, and ends the process. The details of the time-space extension process (S64) will be described in detail later with reference to Figs. 22 and 23.
[0150] On the other hand, if the calculated average traffic density is smaller than the maximum value of the average traffic density at all times acquired from the learning model data 730 ("YES" in S62), it is considered that the prediction accuracy is improved even without extension in the distance direction. In such a case, the extension processing unit 1422 determines whether the calculated average traffic density is equal to or smaller than the maximum value and equal to or larger than the minimum value of the average traffic density (third average traffic density) corresponding to the same time as the time corresponding to the average traffic density acquired from the learning model data 730 (S63).
[0151] If the calculated average traffic density is equal to or less than the maximum value and equal to or greater than the minimum value of the average traffic density corresponding to the same time as the average traffic density obtained from the learning model data 730 (YES in S63), it is assumed that prediction accuracy can be guaranteed to a certain degree even without extension in the time direction. Therefore, in such a case, the extension processing unit 1422 ends the process.
[0152] On the other hand, if the calculated average traffic density exceeds the maximum value of the average traffic density corresponding to the same time as the time corresponding to the average traffic density obtained from the learning model data 730 and is less than the minimum value ("NO" in S63), it is considered that the prediction accuracy will not improve unless extension in the time direction is performed. Therefore, in such a case, the time extension unit 1426 performs a time extension process (S65) and ends the process. The details of the time extension process (S65) will be described in detail later with reference to Figs. 24 and 25.
[0153] (Space-time extension 1424) Fig. 22 is a diagram showing an overview of the space-time expansion processing by the space-time expansion unit 1424. Fig. 23 is a flowchart showing details of the space-time expansion processing by the space-time expansion unit 1424. Note that this expansion processing is called from S64 (Fig. 21) in the expansion processing by the expansion processing unit 1422. Details of the space-time expansion processing by the space-time expansion unit 1424 will be described with reference to Figs. 22 and 23.
[0154] As shown in Fig. 23, the time-space expansion unit 1424 acquires traffic density data (second traffic density) from the traffic density data storage unit 125 (S71). The traffic density data is traffic density for each time and for each cell. Referring to Fig. 22, traffic density data G1 acquired by the time-space expansion unit 1424 is shown. In the traffic density data G1, the darker the color of a time-space range, the higher the traffic density is. The time-space expansion unit 1424 generates prediction data based on expanding a time-space region according to the traffic density congestion time (second congestion time) in the time direction and distance direction.
[0155] The time-space expansion unit 1424 acquires the most upstream point of a point before and after which a speed change occurs in a congestion flow area as a reference point L1 (a predetermined point). Examples of the reference point L1 include an interchange (IC), a junction (JCT), a service area (SA), a parking area (PA), or a point where the number of lanes changes. The location of such a reference point L1 may be registered in advance in a storage unit (not shown) inside or outside the traffic flow prediction device 1. In addition, the time-space expansion unit 1424 acquires the time when the speed converted from the traffic density at the reference point L1 becomes equal to or lower than the congestion flow judgment speed 641 as the congestion time (second congestion time) (S72).
[0156] The traffic density data G1 indicates the "traffic jam time (at reference point L1)," "start time (of the traffic jam time)," and "end time (of the traffic jam time)." The space-time region R11 is a space-time region (first region) located within the traffic jam time at reference point L1 and upstream of reference point L1. The space-time region R21 is a space-time region (second region) located within the traffic jam time at reference point L1 and downstream of reference point L1.
[0157] Furthermore, the time-space region R31a is a time-space region (third region) corresponding to the time before the traffic jam time at the reference point L1. The time-space region R31b is a time-space region (fourth region) corresponding to the time after the traffic jam time at the reference point L1. The time width of the entire region combining the time-space region R31a, the time-space region R11, the time-space region R21, and the time-space region R31b is the second time.
[0158] The space-time expansion unit 1424 generates a space-time region R12 after distance direction expansion in the traffic density data G2 based on expanding the space-time region R11 in the distance direction (upstream direction) (S73). For example, when the reference point L1 is 0 m, data at 0 m, 500 m, and 1000 m points becomes data at 0 m, 750 m, and 1500 m points due to the expansion in the distance direction.
[0159] Next, the space-time expansion unit 1424 generates a space-time region R23 and a space-time region R13 after extension in the time direction based on extending the space-time region (space-time region R12 and region R21) in the congestion time in the time direction (S74). For example, when the congestion time at the reference point L1 is from 12:00 to 13:00, the data from 12:00 to 12:30 to 13:00 is treated as the data from 11:45 to 12:30 to 13:15 by the extension in the time direction.
[0160] Next, the space-time expansion unit 1424 translates the space-time regions outside the congestion time (space-time region R31a and space-time region R31b) in the time direction (S75). More specifically, the space-time expansion unit 1424 shifts the space-time region R31a to a time before the space-time region R23 and space-time region R13 after the extension in the time direction, and shifts the space-time region R31b to a time after the space-time region R23 and space-time region R13 after the extension in the time direction, thereby generating traffic density data G3. For example, by translating in the time direction, data at 11:45, 11:50, and 11:55 before the congestion occurs is made data at 11:30, 11:35, and 11:40.
[0161] Next, the time-space expansion unit 1424 adjusts the traffic density data G3 by applying an existing method such as linear interpolation to the traffic density data G3, and generates prediction data for each simulation execution cell width 622 (e.g., 500 m) at a preset simulation execution time interval 621 (e.g., 5-minute interval) (S76). The time-space expansion unit 1424 stores the prediction data in the traffic density data storage unit 125.
[0162] (Time extension 1426) Fig. 24 is a diagram showing an overview of the time extension process by the time extension unit 1426. Fig. 25 is a flowchart showing the details of the time extension process by the time extension unit 1426. Note that this extension process is called from S65 (Fig. 21) in the extension process by the extension processing unit 1422. Details of the time extension process by the time extension unit 1426 will be described with reference to Figs. 24 and 25.
[0163] Referring to Fig. 24, the past average traffic density H11 "before time adjustment" and the past average traffic density H12 "after time adjustment" are shown. The current average traffic density H21 is also shown. "Traffic jam start" corresponds to the traffic jam start time, and "traffic jam end" corresponds to the traffic jam end time.
[0164] In the "before time adjustment", the current average traffic density H21 is larger than the maximum value of the past average traffic density H11 at the same time as the time corresponding to the current average traffic density H21. Therefore, in the "before time adjustment", the accuracy is not improved even if a prediction is made using the past average traffic density H11 in the "before time adjustment" and the current average traffic density H21.
[0165] On the other hand, in the "after time adjustment", the past average traffic density H11 from "start of traffic jam" to "clearance of traffic jam" is extended toward the current average traffic density H21 based on "clearance of traffic jam". As a result, the current average traffic density H21 is equal to or greater than the minimum value and equal to or less than the maximum value of the past average traffic density H12 at the same time as the time corresponding to the current average traffic density H21. Therefore, the prediction accuracy is improved by using the past average traffic density H12 after "time adjustment" and the current average traffic density H21 to make a prediction.
[0166] 25, first, the time expansion unit 1426 acquires the learning model data 730 associated with the time of the day of the same type as the day type to which the current time belongs from the learning parameter storage unit 127. The time expansion unit 1426 generates prediction data by expanding in the time direction the spatio-temporal region corresponding to the average traffic density and the traffic congestion time (second congestion time) of each traffic density included in such learning model data 730.
[0167] The time extension unit 1426 acquires a traffic jam start time and a traffic jam end time based on the acquired learning model data 730 (S81). The time extension unit 1426 may calculate, as the traffic jam start time, a representative value (e.g., a median value, etc.) obtained by statistically processing the start time of each traffic jam of the average traffic density included in the acquired learning model data 730. Similarly, the time extension unit 1426 may calculate, as the traffic jam end time, a representative value (e.g., a median value, etc.) obtained by statistically processing the end time of each traffic jam of the average traffic density included in the acquired learning model data 730.
[0168] Next, the time extension unit 1426 judges whether the average traffic density has increased from before the current time to the current time. More specifically, the time extension unit 1426 judges whether the average traffic density at the current time has been increasing in the recent past (for example, the past 30 minutes) (S82). This makes it possible to judge whether a traffic jam has occurred or is extending.
[0169] If the average traffic density at the current time is on the rise ("YES" in S82), it is considered that a traffic jam is occurring. Therefore, the time expansion unit 1426 expands the average traffic density and traffic density included in the acquired learning model data 730 in the time direction, using the traffic jam end time as a reference, until the traffic jam start time matches the current time (S83). As a result, the traffic jam area (time-space area corresponding to the traffic jam time) from the traffic jam start time to the traffic jam end time in the average traffic density and traffic density included in the acquired learning model data 730 is expanded in the time direction. Then, the operation proceeds to S85.
[0170] On the other hand, if the average traffic density at the current time is on a decreasing trend ("NO" in S82), it is considered that the congestion is extending. Therefore, the time expansion unit 1426 expands in the time direction the congestion end time of the average traffic density and traffic density included in the acquired learning model data 730, based on the congestion start time, until it matches the current time (S84). As a result, the congestion area (time-space area according to the congestion time) from the congestion start time to the congestion end time of the average traffic density and traffic density included in the acquired learning model data 730 is expanded in the time direction. Then, the operation proceeds to S85.
[0171] Next, the time extension unit 1426 adjusts the average traffic density and traffic density after the time direction extension by applying an existing method such as linear interpolation to the average traffic density and traffic density after the time direction extension, and generates prediction data for each simulation execution cell width 622 (e.g., 500 m) at a preset simulation execution time interval 621 (e.g., 5 minutes) (S85). The prediction data is used to calculate a predicted percentile value in S55 (FIG. 20).
[0172] (Prediction result storage unit 128) The prediction result storage unit 128 stores the prediction result output from the inference unit 142 .
[0173] The details of the embodiment of the present invention have been described above.
[0174] (1-2. Effects) As described above, according to the embodiment and the modified example of the present invention, traffic density data is calculated based on the driving history data and the traffic volume data, and the average traffic density for a preset time width and section is calculated based on the traffic density data. Also, a learned percentile value is calculated based on the stored past traffic densities and average traffic densities, and the average traffic density is calculated from the latest data acquired in real time.
[0175] At this time, it is determined whether the average traffic density deviates from the learning data, and if the average traffic density deviates from the learning data, data augmentation is applied to the learning model data. Then, based on a comparison between the learning model data after the data augmentation and the calculated average traffic density, a predicted percentile value is calculated, and the speed is predicted from the predicted percentile value. This provides the effect of being able to predict traffic flow with high accuracy even when the traffic flow deviates from the learning data.
[0176] Fig. 26 is a diagram showing a comparison of actual measurement results of traffic jam lengths not included in the learning data, prediction results by a comparative example, and prediction results by this embodiment. Referring to Fig. 26, it can be seen that the "prediction by the comparative example" is far from the "actual measurement" because the learning about the vicinity of the end of the traffic jam is insufficient in the "prediction by the comparative example". On the other hand, it can be seen that the "prediction by this embodiment" is close to the "actual measurement" because the prediction is made based on data that has been expanded in the distance direction.
[0177] Fig. 27 is a diagram showing a comparison of actual measurement results of traffic congestion during time periods not included in the learning data, prediction results by a comparative example, and prediction results by this embodiment. Referring to Fig. 27, it can be seen that the "prediction by the comparative example" is far from the "actual measurement" because learning about the early morning hours is insufficient in the "prediction by the comparative example." On the other hand, it can be seen that the "prediction by this embodiment" is close to the "actual measurement" because the prediction is made based on data that has been expanded in the time direction.
[0178] The effects achieved by the traffic flow prediction device 1 according to the embodiment of the present invention and its modified examples have been described above.
[0179] (2. Hardware configuration example) Next, an example of the hardware configuration of the traffic flow prediction device 1 according to the embodiment of the present invention will be described.
[0180] 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 flow prediction device 1 according to the 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 flow prediction device 1. Therefore, in the hardware configuration of the traffic flow prediction device 1, unnecessary components may be deleted from the hardware configuration of the information processing device 900 described below, or new components may be added.
[0181] 28 is a diagram showing a hardware configuration of an information processing device 900 as an example of the traffic flow 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.
[0182] The CPU 901 functions as an arithmetic processing device and control device, and controls the overall operation of the information processing device 900 in accordance with 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 appropriately during the execution. These are connected to each other by a host bus 904 consisting of a CPU bus etc.
[0183] 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 necessary to configure the host bus 904, bridge 905, and external bus 906 separately, and these functions may be implemented in a single bus.
[0184] The input device 908 is composed of input means such as a mouse, keyboard, touch panel, button, microphone, switch, and lever for the user to input information, and an input control circuit that generates an input signal based on the user's input and outputs it to the CPU 901. A user who operates the information processing device 900 can input various data to the information processing device 900 and instruct the information processing device 900 to perform processing operations by operating this input device 908.
[0185] The output device 909 includes, for example, a display device such as a CRT (Cathode Ray Tube) display device, a Liquid Crystal Display (LCD) device, an OLED (Organic Light Emitting Diode) device, or a lamp, and an audio output device such as a speaker.
[0186] 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 in the storage medium, a reading device for reading data from the storage medium, and a deleting device for deleting data recorded in the storage medium. The storage device 910 is configured, for example, with an HDD (Hard Disk Drive). This storage device 910 drives a hard disk and stores programs executed by the CPU 901 and various data.
[0187] The communication unit 911 is, for example, a communication interface configured with a communication device for connecting to a network, etc. Also, the communication unit 911 may be compatible with either wireless communication or wired communication.
[0188] An example of the hardware configuration of the traffic flow prediction device 1 according to the embodiment of the present invention has been described above.
[0189] (3. Summary) Although the preferred embodiment of the present invention has been described in detail above with reference to the accompanying drawings, the present invention is not limited to such an example. It is clear that a person having ordinary knowledge in the technical field to which the present invention pertains can conceive of various modified or altered examples within the scope of the technical ideas described in the claims, and it is understood that these also naturally belong to the technical scope of the present invention. [Explanation of symbols]
[0190] 1. Traffic flow prediction device 112 Probe Antenna 114 Free Flow Antenna 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 parameter memory unit 128 Prediction result memory unit 131 Statistical Processing Unit 133 Traffic Volume Calculation Department 140 Processing section 141 Learning Department 1411 KV Parameter Creation Section 1412 Traffic density calculation section 1414 Percentile Learning Department 142 Reasoning part 1422 Extended Processing Unit 1424 Time-Space Expansion Unit 1426 Time Extension
Claims
1. a determination unit that determines whether or not a first congestion time exists during a first time period based on a first average traffic density of a first traffic density at each time point belonging to the first time period; a prediction unit which, when it is determined that the first congestion time exists, generates prediction data belonging to the second time based on extending in the time direction a time-space region corresponding to a second congestion time of a second traffic density at each time belonging to the second time or a second average traffic density for each time of the second traffic density, and predicts a third traffic density at a time after the first time based on the prediction data; An information processing device comprising:
2. the prediction unit generates the prediction data based on extending the time-space domain of the second average traffic density in a time direction when it is determined that the first congestion time exists; The information processing device according to claim 1 .
3. The prediction unit extends the time-space region of the second average traffic density in a time direction based on a congestion end time when the first average traffic density has increased since before the first time. The information processing device according to claim 2 .
4. When the first average traffic density has decreased since before the first time, the prediction unit expands the spatio-temporal region of the second average traffic density from a congestion start time to a congestion end time in a time direction based on the congestion start time. The information processing device according to claim 2 .
5. the prediction unit generates the prediction data based on expanding the time-space region of the second traffic density in a time direction when it is determined that the first congestion time exists; The information processing device according to claim 1 .
6. When it is determined that the first congestion time exists, the prediction unit generates a distance-direction extended area based on extending in an upstream direction a first area of the second traffic density that is within the second congestion time and is located upstream of a predetermined point, generates a time-direction extended area based on extending in a time direction a distance-direction extended area and a second area of the second traffic density that is within the second congestion time and is located downstream of the predetermined point, shifts a third area of the second traffic density that corresponds to a time before the first congestion time to a time before the time-direction extended area, and generates the prediction data based on shifting a fourth area of the second traffic density that corresponds to a time after the second congestion time to a time after the time-direction extended area. The information processing device according to claim 5 .
7. The determination unit determines whether or not the first congestion time exists based on whether or not the first average traffic density is equal to or greater than a predetermined value. The information processing device according to claim 1 .
8. the prediction unit generates the prediction data based on extending the time-space region of the second average traffic density in a time direction when it is determined that the first congestion time exists and when the first average traffic density is smaller than a maximum value of the second average traffic density. The information processing device according to claim 1 .
9. the prediction unit generates the prediction data based on extending the time-space region of the second average traffic density in the time direction when it is determined that the first congestion time exists, and when the first average traffic density is smaller than the maximum value of the second average traffic density, and when the first average traffic density is equal to or smaller than the maximum value of a third average traffic density of the traffic densities corresponding to the same time as the time corresponding to the first average traffic density among the second average traffic densities and is equal to or larger than the minimum value of the third average traffic density; The information processing device according to claim 1 .
10. the prediction unit generates the prediction data based on extending the time-space region of the second traffic density in a distance direction and a time direction when it is determined that the first congestion time exists and when the first average traffic density is equal to or greater than a maximum value of the second average traffic density. The information processing device according to claim 1 .
11. determining whether or not a first congestion time exists during a first time period based on a first average traffic density of a first traffic density at each time point belonging to the first time period; When it is determined that the first congestion time exists, generating prediction data belonging to the second time based on extending in the time direction a time-space region corresponding to a second congestion time of a second traffic density at each time belonging to the second time or a second average traffic density for each time of the second traffic density, and predicting a third traffic density at a time after the first time based on the prediction data; An information processing method comprising:
12. Computer, a determination unit that determines whether or not a first congestion time exists during a first time period based on a first average traffic density of a first traffic density at each time point belonging to the first time period; a prediction unit which, when it is determined that the first congestion time exists, generates prediction data belonging to the second time based on extending in the time direction a time-space region corresponding to a second congestion time of a second traffic density at each time belonging to the second time or a second average traffic density for each time of the second traffic density, and predicts a third traffic density at a time after the first time based on the prediction data; A program that causes the information processing device to function as an information processing device having the above-mentioned.
Citation Information
Patent Citations
Travel time prediction method, device and program
JP2005196238A
Apparatus and method for estimating and predicting route trip time
JP2007011596A
