Traffic condition prediction device and traffic condition prediction method

Abstract

To predict future traffic situations in a plurality of zones of a road through simple processing.SOLUTION: A traffic situation prediction device comprises: a learning processing unit which generates a prediction model by performing machine learning based on teacher data defining, as input data, past traffic situation data before a past predetermined time point for each group obtained by dividing a plurality of zones resulting from dividing a road where a vehicle travels into groups based on similarity of traffic situation data and defining, as output data, past traffic situation data after the predetermined time point, in which a zone having a higher possibility to be a bottleneck of congestion is weighted heavier among the plurality of zones within a group; an acquisition unit which acquires present traffic situation data collected by a road information collection terminal for each zone; and a prediction processing unit which defines the present traffic situation data as input data for each zone and acquires future traffic situation data as output data based on the prediction model corresponding to a group to which the zone belongs.SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] An embodiment of the present invention relates to a traffic condition prediction device and a traffic condition prediction method. [Background technology]

[0002] Conventionally, congestion information has been provided to drivers of vehicles traveling on roads as information on the traffic conditions on planned routes. Regarding congestion information, there is a technology in which, for example, a traffic control center or the like performs machine learning using past traffic condition data as training data to generate a prediction model, and then predicts future traffic conditions using the prediction model. This technology makes it possible to predict future traffic conditions (e.g., vehicle speed, congestion level, etc.). Furthermore, by predicting future traffic conditions, it is possible to implement appropriate measures to maintain good traffic conditions. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] International Publication No. 2017 / 033443 [Patent Document 2] Patent No. 3240505 [Patent Document 3] Patent No. 4115373 [Patent Document 4] Japanese Patent Publication No. 2020-135231 [Patent Document 5] Japanese Patent Application Publication No. 2018-147037 [Patent Document 6] Patent No. 6235266 Summary of the Invention [Problem to be solved by the invention]

[0004] In the above-mentioned conventional technology, when predicting future traffic conditions, a separate prediction model is generally generated independently for each road section. However, this method requires large-scale calculations when predicting future traffic conditions for many sections, such as when expanding the traffic condition prediction range to a wide area, which results in large physical and time costs for learning and realizing prediction.

[0005] Therefore, an object of this embodiment is to provide a traffic condition prediction device and a traffic condition prediction method that can predict future traffic conditions in multiple sections of a road through simple processing. [Means for solving the problem]

[0006] In one embodiment, a traffic condition prediction device generates a prediction model by performing machine learning based on teacher data for each group, where multiple sections obtained by dividing a road on which a vehicle travels are grouped based on the similarity of traffic condition data, and where past traffic condition data before a predetermined time in the past is used as input data and past traffic condition data after the predetermined time is used as output data.In this case, the device is equipped with a learning processing unit that generates the prediction model by weighting multiple sections within a group more heavily, such as sections that are more likely to become bottlenecks in congestion; an acquisition unit that acquires current traffic condition data collected by a road information collection terminal for each section; and a prediction processing unit that uses the current traffic condition data as input data for each section and acquires future traffic condition data as output data based on the prediction model associated with the group to which the section belongs. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram schematically showing a road in the first embodiment. [Figure 2] FIG. 2 is a diagram showing the overall configuration of the traffic condition prediction system according to the first embodiment. [Figure 3] FIG. 3 is a functional configuration diagram of the traffic condition prediction device of the first embodiment. [Figure 4]FIG. 4 is a diagram schematically showing a road with a merging road in the first embodiment. [Figure 5] FIG. 5 is a graph showing an example of changes in speed over time for each road section in the first embodiment. [Figure 6] FIG. 6 is a diagram schematically showing an example of grouping road sections in the first embodiment. [Figure 7] FIG. 7 is an explanatory diagram of weighting of each section during learning in the first embodiment. [Figure 8] FIG. 8 is a diagram schematically illustrating an overview of the learning process performed by the traffic condition prediction device of the first embodiment. [Figure 9] FIG. 9 is a diagram schematically showing an example of displaying the congestion levels of roads at a plurality of future points in time in the first embodiment. [Figure 10] FIG. 10 is a flowchart and the like showing the learning process performed by the traffic condition prediction device in the first embodiment. [Figure 11] FIG. 11 is a flowchart and the like showing the prediction process by the traffic condition prediction device in the first embodiment. [Figure 12] FIG. 12 is a diagram showing combinations of a target section and surrounding sections in the second embodiment. [Figure 13] FIG. 13 is an explanatory diagram of how to use a combination of a target section and surrounding sections in the second embodiment. [Figure 14] FIG. 14 is an explanatory diagram of calculation of arrival time at an end point when departing from a starting point at a predetermined departure time in the second embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0008] Hereinafter, embodiments (first embodiment and second embodiment) of a traffic condition prediction device and a traffic condition prediction method of the present invention will be described with reference to the drawings.

[0009] First, with reference to FIG. 1, a road R in the first embodiment will be described. FIG. 1 is a diagram schematically illustrating a road R in the first embodiment. The road R on which the vehicle C travels is, for example, a highway. The road R is divided into sections #1, #2, #3, ... as management units. A vehicle detector 2 (details will be described later) is installed in each section. Note that, hereinafter, the reference number for the road R may be omitted and the road may be referred to as "road."

[0010] Next, the overall configuration of the traffic condition prediction system S of the first embodiment will be described with reference to FIG. 2 is an overall configuration diagram of a traffic condition prediction system S according to the first embodiment. The traffic condition prediction system S includes a traffic condition prediction device 1, a vehicle detector 2 (road information collection terminal), a road traffic control system 3, a weather data management device 4, and an event data management device 5.

[0011] Vehicle detectors 2 (see Figure 1) are installed on the roadside at each section of the expressway and collect information (traffic condition data) such as traffic volume [vehicles / hour], average speed [km / h], vehicle density [vehicles / km], and occupancy rate [%]. The vehicle detectors 2 transmit the collected traffic condition data to the road traffic control system 3.

[0012] The device that collects traffic condition data may also be a mobile terminal (such as a smartphone) carried by a passenger of a vehicle traveling on a road, or an on-board device of the vehicle.

[0013] The road traffic control system 3 is a computer system that comprehensively monitors and manages the actual traffic conditions on the roads that it controls, and transmits traffic condition data received from the vehicle detectors 2 to the traffic condition prediction device 1. The road traffic control system 3 also instructs the traffic condition prediction device 1 to predict future traffic conditions (details will be described later).

[0014] The weather data management device 4 is a computer system that manages weather data (for example, temperature data, humidity data, sunny / cloudy / rainy / snowy data, etc.) collected by various sensors, etc., and transmits the weather data to the traffic condition prediction device 1.

[0015] The event data management device 5 is a computer system that manages data on events (for example, exhibitions, concerts, etc.) that affect road traffic conditions, and transmits the event data to the traffic condition prediction device 1.

[0016] Next, the functional configuration of the traffic condition prediction device 1 of the first embodiment will be described with reference to FIG. 3 is a functional configuration diagram of the traffic condition prediction device 1 of the first embodiment. The traffic condition prediction device 1 is a computer device, and includes a processing unit 11, a storage unit 12, an input unit 13, a display unit 14, and a communication unit 15.

[0017] In the first embodiment, the traffic condition prediction device 1 is described as being configured by a single computer device for the sake of simplicity, but is not limited to this. The traffic condition prediction device 1 may be realized by, for example, multiple computers, or may be realized by a cloud server.

[0018] The storage unit 12 is a storage device such as a hard disk drive (HDD) or a solid state drive (SSD) and stores various types of information. The storage unit 12 stores, for example, road data 121, traffic condition data 122, weather data 123, event data 124, training data 125, a prediction model 126, and a prediction result 127.

[0019] The road data 121 is information relating to the structure of a road, such as section identification information, length (section length), maximum number of vehicles that can be accommodated, number of lanes, interchanges, and information on the locations of parking areas.

[0020] The traffic condition data 122 is information such as traffic volume [vehicles / h], average speed [km / h], vehicle density [vehicles / km], and occupancy [%] collected by the vehicle detector 2 and obtained from the road traffic control system 3. Note that, hereinafter, past traffic condition data used to generate the prediction model 126 will be referred to as past traffic condition data, and current (including the most recent) traffic condition data used to predict the traffic condition will be referred to as current traffic condition data.

[0021] The weather data 123 is weather data acquired from the weather data management device 4.

[0022] The event data 124 is event data acquired from the event data management device 5.

[0023] The teacher data 125 is correct answer data (actual values) of input data and output data for generating the prediction model 126. For example, the input data is past traffic condition data before a predetermined time point in the past, and the output data is past traffic condition data after the predetermined time point.

[0024] The prediction model 126 is a model generated by the learning processing unit 112. The prediction model 126 is generated for each group obtained by grouping a plurality of sections based on the similarity of traffic condition data. Details of the prediction model 126 will be described later.

[0025] The prediction result 127 is the result of the traffic situation prediction by the prediction processing unit 113 .

[0026] The processing unit 11 includes, for example, a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory). The CPU comprehensively controls the operation of the traffic condition prediction device 1. The ROM is a storage medium for storing various programs and data. The RAM is a storage medium for temporarily storing various programs and rewriting various data. The CPU uses the RAM as a work area to execute programs stored in the ROM, the storage unit 12, etc.

[0027] The processing unit 11 includes, as functional components, an acquisition unit 111, a learning processing unit 112, a prediction processing unit 113, a calculation unit 114, a display control unit 115, and a transmission control unit .

[0028] The acquisition unit 111 acquires various types of information from external devices. For example, the acquisition unit 111 acquires current traffic condition data collected by the vehicle detectors 2 for each section from the road traffic control system 3. The acquisition unit 111 also acquires current weather data from the weather data management device 4. The acquisition unit 111 also acquires event data from the event data management device 5.

[0029] The learning processing unit 112 groups multiple sections of multiple roads based on the similarity of their traffic condition data, and manages information related to the grouping using, for example, a correspondence table that associates sections with groups. Furthermore, the learning processing unit 112 generates a prediction model 126 by performing machine learning for each group based on training data in which past traffic condition data before a predetermined time point in the past is used as input data and past traffic condition data after the predetermined time point is used as output data. In this case, for example, the learning processing unit 112 generates the prediction model 126 by weighting sections within a group more heavily, with the more likely they are to become a bottleneck in traffic congestion (details will be described later using FIG. 7).

[0030] Furthermore, for example, the learning processing unit 112 generates a prediction model 126 by performing machine learning based on the teacher data of the interval of interest and its surrounding intervals (details will be described later with reference to FIG. 8).

[0031] Furthermore, when generating the prediction model 126, the learning processing unit 112 may further use past weather data, past event data, and the like.

[0032] The prediction processing unit 113 receives current traffic condition data as input data for each section, and acquires future traffic condition data as output data based on the prediction model 126 associated with the group to which the section belongs.

[0033] For example, for each section of interest, the prediction processing unit 113 receives current traffic condition data for the section of interest and its surrounding sections as input data, and obtains future traffic condition data as output data based on a prediction model associated with the group to which the section of interest belongs and the group to which the surrounding sections belong.

[0034] The calculation unit 114 calculates, for example, for each section, the length of traffic congestion, the time lost due to traffic congestion, the travel time, the amount of traffic congestion, and the like, based on future traffic condition data.

[0035] The display control unit 115 causes the display unit 14 to display various types of information.

[0036] The transmission control unit 116 transmits various information to an external device. For example, the transmission control unit 116 transmits the prediction result 127 by the prediction processing unit 113 to the road traffic control system 3, a vehicle traveling on the road, an information board installed on the side of the road, or the like, and displays it.

[0037] The input unit 13 is an input device that accepts user operations on the traffic condition prediction device 1, and is, for example, a keyboard, a mouse, or the like.

[0038] The display unit 14 is realized by a liquid crystal display (LCD (Liquid Crystal Display)), an organic EL (Electro-Luminescence) display, or the like.

[0039] The communication unit 15 is a communication interface for communicating with an external device.

[0040] This concludes the description of the configuration of each device, and the following describes the processing performed by the traffic condition prediction device 1. First, a more suitable method for learning and prediction regarding the prediction model 126 will be described with reference to FIG. Fig. 4 is a diagram showing a schematic diagram of a road with a merging section in the first embodiment. Learning and prediction can be performed for each section, but it is preferable to use data not only for that section but also for its upstream section (surrounding section) and downstream section (surrounding section). In particular, performing learning and prediction in this manner for merging sections is preferable in terms of improving accuracy.

[0041] As shown in FIG. 4, sections #11 to #14 are provided on a road, and section #12 is a section where vehicles from section #15 merge. In this case, section #12 is a so-called bottleneck section where congestion is likely to occur. In other words, in section #12, the number of vehicles is likely to approach the maximum number of vehicles that can be accommodated, and congestion occurs when the number of vehicles approaches a certain level. Vehicles that cannot fit into section #12 spill over into section #13, the adjacent upstream section, and the congestion propagates upstream. In addition to merging sections, bottleneck sections include sections that include entrances and exits (interchanges) to public roads.

[0042] Next, the grouping of sections will be described with reference to FIG. Fig. 5 is a graph showing an example of changes in speed over time for each section of a road in the first embodiment, in which (a) to (d) are graphs showing changes in speed over time for different sections.

[0043] Graphs (a) and (b) have similar general shapes. Graphs (c) and (d) also have similar general shapes. Therefore, sections (a) and (b) are grouped together, and sections (c) and (d) are grouped together. Specifically, this type of classification can be achieved using a technique called clustering.

[0044] Since the traffic condition data for sections belonging to the same group are similar to each other, a common prediction model 126 is created through learning and used for prediction. This allows for a significant reduction in the amount of processing compared to conventional techniques that use separate prediction models for each section. This effect becomes more pronounced as the number of sections to be predicted increases.

[0045] Next, an example of grouping road sections will be described with reference to FIG. FIG. 6 is a diagram schematically showing an example of grouping road sections in the first embodiment. The right side of FIG. 6 shows an example of a route on which vehicles travel in one direction. Here, sections 1 to 15 are shown, with sections 3 and 8 being junctions of multiple routes. When clustering is performed on these sections 1 to 15, the sections are classified into groups 1 to 3. In this example, group 1 consists of 8 sections, group 2 consists of 2 sections, and group 3 consists of 5 sections. A prediction model 126 that can be used in common for each of these three groups is created.

[0046] In conventional techniques, a large number of intervals requires large-scale calculations, which increases the physical and time costs for learning and prediction. On the other hand, according to the method of the first embodiment, even if the number of intervals is large, the accuracy of learning and prediction can be maintained at the same level as in conventional techniques by grouping the intervals, and the physical and time costs can be significantly reduced.

[0047] Next, weighting of each section during learning will be described with reference to Fig. 7. Fig. 6 will also be referenced. FIG. 7 is an explanatory diagram of weighting of each section during learning in the first embodiment. In FIG. 7, the horizontal axis indicates the number of prediction errors (difference between prediction and actual) during learning (when creating the prediction model 126). For example, there are two sections in group 2. If the total number of prediction errors in sections 2 and 3 is set to be equal to or less than the number of standard errors for a group of two sections, the number of errors in sections 2 and 3 will not necessarily be equal, and an imbalance may occur, such as one having few errors and the other having many errors. However, as shown in FIG. 6, section 3 is a junction and is a section where congestion is likely to occur, so it is undesirable for the prediction accuracy of section 3 to deteriorate.

[0048] For this reason, a permutation of sections within each group for which prediction accuracy is desired is created. In FIG. 7, the higher the section in each group, the higher the degree of need for prediction accuracy. In this case, for example, the frequency of congestion in each section may be tallied and the order may be determined based on the magnitude of the frequency of occurrence. In particular, since sections 3 and 8 are merging sections where congestion is likely to occur, a weighting coefficient for each section is set in the loss function during learning so that ensuring accuracy in these sections is given priority over other sections. In this way, the learning processing unit 112 generates the prediction model 126 by weighting sections within a group more heavily, with the section more likely to become a congestion bottleneck.

[0049] Next, an overview of the learning process will be described with reference to FIG. 8 is a diagram schematically illustrating an overview of the learning process performed by the traffic condition prediction device 1 according to the first embodiment. A prediction model 126 is generated by learning for each group. The prediction model 126 is, for example, a prediction network of LSTM (Long Short-Term Memory), which is a type of neural network that handles time-series data.

[0050] Here, the prediction networks for group 1 and group 3 are used. Of the five intervals shown in the upper left, intervals d1 and b0 belong to group 1, and intervals u1, u2, and u3 belong to group 2.

[0051] For example, when making a prediction for section b0, the past traffic condition data for b0 is input, and a label corresponding to a future predicted value (e.g., congestion level) for every fixed time period (e.g., 5 minutes) is output. In addition, since information on surrounding sections d1, u1, u2, and u3 is also useful for predicting b0, the information is expanded to include it as input. In other words, the learning processing unit 112 generates the prediction model 126 by performing machine learning based on the training data for the section of interest and its surrounding sections.

[0052] Here, the reference time is set to t0, and the congestion level of a section is predicted in 5-minute increments from 5 minutes to 2 hours later. In this case, traffic condition data (traffic volume, speed, and occupancy rate) for the section of interest and surrounding sections for the 2 hours prior to the reference time is prepared. Then, each prediction model is trained using a predetermined loss function.

[0053] In the prior art, for example, if there were 30 intervals, there would also be 30 prediction models. On the other hand, in the first embodiment, if the 30 intervals are classified into about 4 or 5 groups, for example, the number of prediction models 126 will also be about 4 or 5, which reduces the processing load and also provides the advantage of more learning data for each prediction model 126.

[0054] Furthermore, for multiple intervals belonging to one group, a common prediction model 126 is used, but when performing learning and prediction for each interval, an adjuster (adjustment software) may also be used to reflect information specific to that interval.

[0055] Similarly, predictions for the other sections d1, u1, u2, and u3 are made in combination with peripheral section data.

[0056] Next, with reference to FIG. 9, an example of how to use the information on the congestion level, required time, and speed predicted by the traffic condition prediction device 1 will be described. 9 is a schematic diagram showing an example of congestion levels at multiple future time points in the first embodiment. When displaying the congestion level, for example, a congested section display M may be superimposed on road R. When displaying the required time, for example, a route T may be superimposed on road R, and the required time for route T may be displayed numerically. When displaying speed, for example, a graph with time on the horizontal axis and speed on the vertical axis may be displayed.

[0057] For example, by transmitting and displaying information on congestion levels, travel times, and speeds for the next 30, 60, and 120 minutes to vehicles in motion or on information boards installed on the side of the road, drivers can view this information and use it effectively in their driving plans, etc.

[0058] Next, the learning process performed by the traffic condition prediction device 1 will be described with reference to FIG. FIG. 10 is a flowchart and the like showing the learning process performed by the traffic condition prediction device 1 in the first embodiment. First, in step S11, the learning processing unit 112 selects a road to be learned based on an instruction given by the user using the input unit 13 or the like.

[0059] Next, in step S12, the learning processing unit 112 groups the traffic condition data for each of a plurality of sections on the road to be learned based on the similarity of the traffic condition data (FIG. 6).

[0060] Next, in step S13, the learning processing unit 112 learns a prediction model 126 for each group. That is, the learning processing unit 112 generates the prediction model 126 by performing machine learning for each group based on training data in which past traffic condition data before a predetermined time point in the past is used as input data and past traffic condition data after the predetermined time point is used as output data. In this way, the prediction model 126 for each group can be generated.

[0061] In this case, for example, the learning processing unit 112 generates the prediction model 126 by weighting sections within the group more heavily, the more likely they are to become bottlenecks in traffic congestion (FIG. 7). Also, for example, the learning processing unit 112 generates the prediction model 126 by performing machine learning based on training data for the section of interest and its surrounding sections (FIG. 8).

[0062] Next, the prediction process performed by the traffic condition prediction device 1 will be described with reference to FIG. FIG. 11 is a flowchart and the like showing the prediction process by the traffic condition prediction device 1 in the first embodiment. First, in step S21, the prediction processing unit 113 selects a road to be predicted based on an instruction by the user using the input unit 13 or the like.

[0063] Next, in steps S22 to S24, the prediction processing unit 113 performs processing for each section. In step S23, the prediction processing unit 113 receives current traffic condition data for the target section as input data, and acquires future traffic condition data as output data based on the prediction model 126 associated with the group to which the section belongs. Also, for example, for each section of interest, the prediction processing unit 113 receives current traffic condition data for the section of interest and its surrounding sections as input data, and acquires future traffic condition data as output data based on the prediction model associated with the group to which the section of interest belongs and the group to which the surrounding sections belong.

[0064] After steps S22 to S24, in step S25, the calculation unit 114 calculates and outputs (for example, transmits to the road traffic control system 3) an index value for the road to be predicted. For example, the calculation unit 114 calculates and outputs, as index values, congestion length, time lost due to congestion, travel time, congestion volume, etc., based on future traffic condition data for each section.

[0065] As described above, according to the traffic condition prediction device 1 of the first embodiment, multiple sections of a road are grouped, a prediction model 126 is associated with each group and learned, and for each section, the future traffic conditions are predicted using the prediction model 126 corresponding to the group to which it belongs. This makes it possible to predict the future traffic conditions of multiple sections of a road through simple processing.

[0066] Furthermore, for multiple sections within a group, sections that are more likely to become bottlenecks in traffic congestion are weighted more heavily to generate a prediction model 126 (FIG. 7). This makes it possible to improve the accuracy of the prediction results.

[0067] Furthermore, a prediction model 126 is generated by performing machine learning based on the training data for the interval of interest and its surrounding intervals (FIG. 8), which can improve the accuracy of the prediction results.

[0068] (Second embodiment) Next, a second embodiment will be described. Descriptions of matters similar to those in the first embodiment will be omitted as appropriate. In the second embodiment, multiple combinations of surrounding intervals are used when generating the prediction model 126 by performing machine learning.

[0069] FIG. 12 is a diagram showing combinations of a target section and surrounding sections in the second embodiment. Here, the section of interest is selected, for example, as a merging section that is likely to become a bottleneck in traffic congestion. The selection method may be, for example, based on the shape of the road, or on the history of past traffic congestion. Six combinations A to F are then considered as surrounding sections to be combined with the section of interest. Section (3) is the section of interest.

[0070] Next, Fig. 13 is an explanatory diagram of how combinations of a target interval and surrounding intervals are used in the second embodiment. (a) shows each interval represented by a graph structure. (b) is a matrix showing the degree of influence between intervals using weighting coefficients. The weighting coefficients are expressed as real coefficients between 0 and 1, with 1 representing the strongest influence and 0 representing no influence.

[0071] Create such a matrix for each prediction time (from 5 minutes ahead to 120 minutes ahead). If such a matrix does not change depending on the length of the prediction time, obtain a single matrix. However, if the effect changes depending on the length of the prediction time, use a different matrix for each prediction time.

[0072] For example, as shown in (c), when the prediction time is short, the matrix corresponding to combination A is used, when the prediction time is medium, the matrix corresponding to combinations B and C is used, and when the prediction time is long, the matrix corresponding to combination F is used. When the prediction time is short, increasing the range of the surrounding intervals used will actually increase information equivalent to noise, which tends to degrade prediction accuracy, so it is better to reduce the range of the surrounding intervals. In other words, in this case, for the matrix, the weight components are close to 0 for intervals with long distances, and a matrix that minimizes this influence is acquired through learning.

[0073] On the other hand, when the prediction time is long, it is preferable to refer to information on more distant sections, and learning results can be obtained in which the weight components are not 0 even for distant sections. Also, depending on the situation, the combination of surrounding sections that influence the section of interest may differ for each time period. In such cases, a system is provided in which weight coefficients are obtained according to conditions such as prioritizing upstream sections or downstream sections. By obtaining these weight coefficient matrices for each time period, the optimal combination of surrounding sections can be obtained according to the prediction time, thereby improving prediction accuracy.

[0074] In this way, the learning processing unit 112 generates a prediction model 126 by performing machine learning and weighting each individual section that constitutes the surrounding section for each prediction time that indicates the length of time from the reference time to the predicted time.

[0075] Next, FIG. 14 is an explanatory diagram of calculation of arrival time at an end point when departing from a starting point at a predetermined departure time in the second embodiment. Within the wide-area road network shown in (a), conditions such as specifying a route are set in advance. Here, different routes (1), (2), and (3) are set.

[0076] Next, as shown in (b), the traffic sections of the three corresponding routes are extracted. Sections with the same pattern indicate that they belong to the same group. The direction of the arrow indicates the direction the vehicle is moving. The corresponding relationships on the section map indicate which section the vehicle will move from when changing routes. For each section, prediction results can be obtained for the next 5 to 120 minutes, and this prediction information for each section can be obtained from a small number of group networks.

[0077] Therefore, as shown in (c), by compiling the prediction results for the sections corresponding to the route (1), (2), and (3), it is possible to grasp the prediction results for the entire route up to 120 minutes in advance. This system makes it possible to obtain information that shows the time-space relationship for any route, so it can provide information when there is a risk of an increase in travel time, for example, due to the occurrence of a major traffic jam in the near future.

[0078] In this way, the calculation unit 114 calculates the arrival time at a predetermined end point when departing from a predetermined start point on a road at a predetermined departure time, based on future traffic condition data.

[0079] In this way, according to the second embodiment, the neighboring sections to be combined with the section of interest are varied depending on the length of the prediction time (FIG. 13), thereby further improving prediction accuracy.

[0080] Furthermore, based on future traffic condition data (prediction results), it is possible to calculate the arrival time at a predetermined destination when departing from a predetermined starting point at a predetermined departure time (FIG. 14).

[0081] The program executed by the CPU of the traffic condition prediction device 1 of this embodiment may be configured to be provided by being recorded in an installable or executable file format on a computer-readable recording medium such as a CD-ROM, a flexible disk (FD), a CD-R, or a DVD (Digital Versatile Disk).

[0082] Furthermore, the program may be stored on a computer connected to a network such as the Internet and provided by being downloaded via the network. Also, the program executed in this embodiment may be provided or distributed via a network such as the Internet.

[0083] Although the embodiments of the present invention have been described above, the above embodiments are merely examples and are not intended to limit the scope of the invention. The above embodiments can be implemented in various forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. The above embodiments are included within the scope and spirit of the invention, as well as within the scope of the inventions described in the claims and their equivalents.

[0084] For example, the roads to be considered are not limited to expressways, but may be other roads such as general roads. [Explanation of symbols]

[0085] 1...traffic condition prediction device, 2...vehicle detector, 3...road traffic control system, 4...weather data management device, 5...event data management device, 11...processing unit, 12...memory unit, 13...input unit, 14...display unit, 15...communication unit, 111...acquisition unit, 112...learning processing unit, 113...prediction processing unit, 114...calculation unit, 115...display control unit, 116...transmission control unit, 121...road data, 122...traffic condition data, 123...weather data, 124...event data, 125...teaching data, 126...prediction model, 127...prediction result, S...traffic condition prediction system

Claims

1. a learning processing unit that groups a plurality of sections obtained by dividing a road on which a vehicle travels based on the similarity of traffic condition data, and generates a prediction model for each group by performing machine learning based on training data in which past traffic condition data before a predetermined time point in the past is used as input data and past traffic condition data after the predetermined time point is used as output data, and in this case, generates the prediction model by weighting the loss function during learning more heavily for sections that are more likely to become bottlenecks in traffic congestion for the plurality of sections within the group; an acquisition unit that acquires current traffic condition data collected by a road information collection terminal for each section; a prediction processing unit that receives, for each section, the current traffic condition data as input data, and acquires future traffic condition data as output data based on the prediction model associated with the group to which the section belongs; A traffic situation prediction device comprising:

2. the learning processing unit generates the prediction model by performing the machine learning based on the teacher data of the interval of interest and its surrounding intervals; the prediction processing unit receives, for each section of interest, the current traffic condition data of the section of interest and its surrounding sections as input data, and acquires, as output data, future traffic condition data based on the prediction model associated with the group to which the section of interest belongs and the prediction model associated with the group to which the surrounding sections belong. The traffic condition prediction device according to claim 1 .

3. the learning processing unit generates the prediction model by performing the machine learning while weighting each individual section constituting the surrounding section for each prediction time indicating a time length from a reference time to a prediction time. The traffic condition prediction device according to claim 2 .

4. a calculation unit that calculates an arrival time at a predetermined end point when departing from a predetermined start point on the road at a predetermined departure time based on the future traffic condition data, The traffic condition prediction device according to claim 1 further comprises:

5. A computer device comprising: a learning process step in which a plurality of sections obtained by dividing a road on which a vehicle travels are grouped based on the similarity of traffic condition data, and for each group, a prediction model is generated by performing machine learning based on training data in which past traffic condition data before a predetermined time point in the past is used as input data and past traffic condition data after the predetermined time point is used as output data, and in which, for the plurality of sections within a group, the weight of the loss function during learning is assigned heavier to sections that are more likely to become bottlenecks in traffic congestion, thereby generating the prediction model; an acquisition step of acquiring current traffic condition data collected by a road information collection terminal for each section; a prediction processing step for obtaining, for each section, future traffic condition data as output data based on the prediction model associated with the group to which the section belongs, using the current traffic condition data as input data.

Images