JP7451303B2 - Traffic situation prediction device and traffic situation prediction method - Google Patents

Description

Embodiments of the present invention relate to a traffic situation prediction device and a traffic situation prediction method.

2. Description of the Related Art Conventionally, there has been a technology for predicting traffic conditions (such as traffic jam conditions) on roads such as expressways by performing machine learning using past traffic condition data as training data.

Furthermore, when predicting the traffic situation on a main road on a road that has a merge or a branch, the prediction has been made by taking into consideration, for example, the number of vehicles flowing into the merge section or the number of vehicles flowing out from the branch section. In addition, when the structure of a road is complex, such as the Metropolitan Expressway, in order to cope with the fact that traffic conditions and how they change differ depending on the section of the road, for example, the road can be divided into multiple sections, and each section can be divided into sections. There was a method to use machine learning to construct a learning model for predicting each traffic situation.

Patent No. 4783414 Patent No. 6516660 Patent No. 3742361 International Publication No. 2017/094267 International Publication No. 2018/012414 Japanese Patent Application Publication No. 2011-186940 Japanese Patent Application Publication No. 2010-191614

However, with the above-mentioned conventional technology, there is room for improvement in terms of prediction accuracy of traffic conditions.

Therefore, an object of the present embodiment is to provide a traffic situation prediction device and a traffic situation prediction method that can predict road traffic conditions with high accuracy based on machine learning.

The traffic situation prediction device of the embodiment is based on a machine learning algorithm that learns how vehicle congestion occurring on the road is propagated upstream based on past traffic situation data for each section of the road. a generation unit that generates a learning model for predicting traffic conditions for each section; an acquisition unit that acquires current traffic condition data collected by a road information collection terminal for the road; When predicting the traffic situation, the traffic situation is predicted for the first section, which is the predetermined section, based on the learning model, the maximum accommodating number of vehicles, and the traffic situation data, and a prediction unit that predicts a traffic situation for a second section adjacent to the second section based on a prediction result of the traffic situation of the first section, the learning model, the maximum number of vehicles accommodated, and the traffic situation data; Be prepared.

FIG. 1 is a diagram for explaining how traffic congestion propagates on a road. FIG. 2 is an overall configuration diagram of the traffic situation prediction system according to the embodiment. FIG. 3 is a functional configuration diagram of the traffic situation prediction device according to the embodiment. FIG. 4 is a graph showing an example of the relationship between congestion level labels and speeds in the embodiment. FIG. 5 is a diagram for explaining the amount of traffic congestion in the embodiment. FIG. 6 is a flowchart illustrating an example of overall processing by the traffic situation prediction device according to the embodiment. FIG. 7 is a flowchart illustrating an example of learning processing by the traffic situation prediction device according to the embodiment. FIG. 8 is a flowchart illustrating an example of prediction processing by the traffic situation prediction device according to the embodiment. FIG. 9 is a graph showing an example of prediction results in the embodiment. FIG. 10 is a graph showing an example of prediction results in the embodiment. FIG. 11 is a graph showing an example of prediction results in the embodiment.

Hereinafter, a traffic situation prediction system according to an embodiment of the present invention will be described with reference to the drawings. Note that in the following embodiment, a case where the road is an expressway will be taken as an example. Furthermore, the traffic situation is a concept that includes the state of traffic congestion on roads (location of traffic jam, date and time, length of time, etc.) and the type of traffic congestion (for example, natural traffic congestion, accident traffic congestion, construction traffic congestion, spectator traffic congestion, etc.). be. In the following embodiment, a traffic situation will be taken as an example of a traffic situation.

First, we will explain how congestion propagates on roads.

FIG. 1 is a diagram for explaining how traffic congestion propagates on a road. As shown in FIG. 1, a road having an up line and a down line is divided into sections 1, 2, 3, 4, and so on. In addition, the maximum number of vehicles that can be accommodated (the maximum number of vehicles that can travel) is set for each section on the up and down lines.

For example, on the up line, when the number of vehicles approaches the maximum number of vehicles in section 4, congestion occurs, and vehicles that cannot fit in section 4 spill into the adjacent upstream section 3, causing congestion to propagate upstream. Here, the section 4 is a section including, for example, a confluence section, a branch section, a junction, etc. that are likely to become a starting point (bottleneck) for the occurrence of traffic congestion.

Similarly, for the outbound line, for example, when the number of vehicles approaches the maximum capacity in section 1, congestion occurs, and as vehicles that cannot fit in section 1 protrude into the adjacent upstream section 2, congestion increases upstream. propagate.

In addition, in the conventional technology, congestion prediction was performed based on the idea that an increase or decrease in the inflow of vehicles from branch lines connected to the main line directly causes an increase or decrease in the length of congestion at each point on the main line. On the other hand, in this embodiment, the main line is divided into multiple sections, the length of the congestion is estimated for each section, and changes in the length of the congestion are determined by the increase or decrease in the length of the congestion on the entire main line. Congestion prediction is performed based on the idea of causing changes in traffic congestion.

FIG. 2 is an overall configuration diagram of the traffic situation prediction system S of the embodiment. The traffic situation prediction system S includes a traffic situation prediction device 1, a vehicle sensor 2, a road traffic control system 3, and a weather data management device 4.

In FIG. 2, the vehicle sensor 2 is installed on the roadside of the expressway, and has traffic volume [vehicles/h], average speed [km/h], vehicle density [vehicles/km], and occupancy rate (occupancy) [%]. It is a sensor that collects information such as traffic situation data (traffic situation data), and transmits the sensed information to the traffic situation prediction device 1.

Note that the device for collecting traffic situation data may also be a mobile terminal (such as a smartphone) carried by an occupant of a vehicle traveling on the road, an on-vehicle device of the vehicle, or the like.

The road traffic control system 3 is a computer system that comprehensively monitors and manages the actual traffic situation on the road to be controlled, and transmits traffic situation data received from the vehicle sensor 2 to the traffic situation prediction device 1.

The weather data management device 4 is a computer system that manages weather data collected by various sensors and the like, and transmits the weather data received from the various sensors to the traffic situation prediction device 1.

FIG. 3 is a functional configuration diagram of the traffic situation prediction device 1 according to the embodiment. The traffic situation prediction device 1 is a computer device, and includes a processing section 11, a storage section 12, an input section 13, a display section 14, and a communication section 15.

In addition, in this embodiment, in order to simplify the description, the traffic situation prediction device 1 will be described as being configured by one computer device, but the present invention is not limited to this. The traffic situation prediction device 1 may be realized, for example, by a plurality of computer devices, or may be realized by a cloud server.

The storage unit 12 is a storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive), and stores various information. The storage unit 12 stores, for example, road data 121, traffic situation data 122, weather data 123, teacher data 124, learning models 125, and prediction results 126.

The road data 121 is information about roads, such as section identification information, length, maximum number of vehicles accommodated, number of lanes, interchanges, locations of parking areas, and the like.

The traffic situation data 122 includes traffic volume [vehicles/h], average speed [km/h], vehicle density [vehicles/km], occupancy rate ( Information such as occupancy) [%]. Note that, below, the past traffic situation data used to generate the learning model 125 will be referred to as past traffic situation data, and the current traffic situation used to predict the traffic situation will be referred to as current traffic situation or simply traffic situation data. .

The weather data 123 is weather data (temperature data, humidity data, data on sunny/cloudy/rain/snow, etc.) acquired from the weather data management device 4. Note that, hereinafter, past weather data used to generate the learning model 125 will be referred to as past weather data, and current weather data used to predict traffic conditions will be referred to as current weather data or simply weather data.

The teacher data 124 is data in which, for example, a congestion degree label is attached to past traffic situation data.
FIG. 4 is a graph showing an example of the relationship between congestion level labels and speeds in the embodiment. In the graph of FIG. 4, the vertical axis is the congestion degree label, and the horizontal axis is the speed of the vehicle. For example, as shown in FIG. 4, by performing non-linear mapping of traffic congestion level labels and speeds, highly accurate mapping can be performed, for example, in low speed ranges.

The congestion degree label may be, for example, a binary value of 0 (congestion) and 1 (non-congestion), a 3-value (congestion, congestion, free flow), a 5-value, a 10-value, or the like. For example, when attaching a binary congestion degree label to past traffic situation data, for example, if the average speed is less than 20 km/h, it is set as 0 (congestion), and if the average speed is 20 km/h or more, it is set as 1 (non-congestion). And it is sufficient.

Returning to FIG. 2, the learning model 125 is based on past traffic situation data for each section of the road. This is a model generated by the generation unit 112 based on a machine learning algorithm or the like that has learned whether the information propagates to (details will be described later).

The prediction result 126 is the result of prediction of the traffic situation by the prediction unit 113.

The processing unit 11 includes, for example, an MPU (Micro Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory). The MPU centrally controls the operation of the traffic situation prediction device 1. ROM is a storage medium that stores various programs and data. RAM is a storage medium for temporarily storing various programs and rewriting various data. The MPU then executes programs stored in the ROM, the storage unit 12, etc., using the RAM as a work area.

The processing unit 11 includes an acquisition unit 111, a generation unit 112, a prediction unit 113, a display control unit 114, and a transmission control unit 115 as functional configurations.

The acquisition unit 111 acquires various information from an external device. For example, the acquisition unit 111 acquires current traffic situation data collected by the vehicle sensor 2 (road information collection terminal) about the road from the road traffic control system 3. The acquisition unit 111 also acquires current weather data (temperature data, humidity data, data on sunny/cloudy/rain/snow, etc.) from the weather data management device 4 .

The generation unit 112 generates past traffic situation data (information such as past traffic volume [vehicles/h], average speed [km/h], vehicle density [vehicles/km], and occupancy rate (occupancy) [%]), A learning model 125 for predicting traffic conditions in units of sections is generated based on teacher data 124 and a machine learning algorithm that has learned how vehicle congestion occurring on a road propagates upstream. The generation unit 112 may further generate the learning model 125 using past weather data. The following explanation assumes that past weather data and meteorological data are also used.

Note that, in order to generate the above-mentioned learning model 125, the generation unit 112 uses road structure information (for example, which sections are connected to which sections by branching or merging on a major trunk road) based on the road data 121 in the storage unit 12. (e.g.), distance information for each road section, information on bottleneck candidate sections and sections adjacent to that section, etc.

For example, if it is necessary to round (average) traffic situation data collected at 1-minute intervals to 3-minute intervals, 5-minute intervals, etc., the generation unit 112 generates a Perform processing such as averaging.

Further, the generation unit 112 associates the past traffic situation data with the past weather data by, for example, comparing the position information of a sensor that measured the past weather data with the position information of a road section.

For example, when using CAN (Controller Area Network) data (wiper operation information, brake information, inter-vehicle distance information, speed information, etc.), the generation unit 112 compares the position information with the position information of the road section. Then, the past traffic situation data and CAN data are associated with each other.

For example, the generation unit 112 integrates past traffic situation data, past weather data, CAN data, etc. based on these associations, and creates a learning model 125 based on the integrated data, teacher data 124, and a machine learning algorithm. generate.

When predicting the traffic situation for each section, the prediction unit 113 uses the learning model 125 for a first section that is a predetermined section (for example, a section predetermined as a section that is the starting point (bottleneck) of traffic congestion). , the maximum number of accommodated vehicles, traffic situation data 122, weather data 123, etc. to predict the traffic situation.

Further, the prediction unit 113 includes, for a second section adjacent to the first section on the upstream side, a prediction result 126 of the traffic situation of the first section, a learning model 125, the maximum number of accommodated vehicles, traffic situation data 122, Traffic conditions are predicted based on weather data 123 and the like. As described above, in this embodiment, based on the idea that congestion propagates upstream, we first predict traffic congestion in a bottleneck candidate section, and then predict traffic conditions for adjacent upstream sections one after another. By performing prediction using so-called convolution processing, prediction accuracy can be improved.

Furthermore, examples of traffic conditions to be predicted include the distance of a traffic jam, the length of a traffic jam, and the amount of traffic jam, which is the product of the average distance of a traffic jam and the length of a traffic jam.

FIG. 5 is a diagram for explaining the amount of traffic congestion in the embodiment. As shown in FIG. 5, for example, the product of the average distance of a traffic jam ((1/2)×L) and the length of time of a traffic jam (T) is the traffic jam volume. The area of triangle A1 in FIG. 5 corresponds to the amount of traffic congestion. Furthermore, the area of rectangle A2 also corresponds to the amount of traffic congestion.

In the conventional technology, for example, traffic volume Q (=traffic density k×traffic speed v) was used to understand the traffic situation. On the other hand, in this embodiment, in order to understand the traffic congestion situation as a traffic situation, the amount of traffic congestion, which is the area of the triangle A1 and the area of the quadrangle A2 described above, is used. It is highly likely that the length of a traffic jam is approximately 0 when the traffic jam occurs and when the traffic congestion clears, and reaches its maximum at the time in between, so the area of triangle A1 mentioned above is particularly accurate as an indicator of the traffic congestion situation. be.

Returning to FIG. 2, other possible traffic conditions to be predicted include, for example, the travel time from a predetermined starting point to a predetermined end point on a road, the timing of clearing a traffic jam, and the length of the longest traffic jam.

In addition, in terms of time, the traffic situation to be predicted may be, for example, traffic congestion in the near future from 5 minutes to 2 hours in 5 minute units.

Further, when predicting the traffic situation, for example, it is possible to make predictions by associating information on two sections immediately before the merge with information on the section immediately after the merge. That is, for example, all vehicles in the two sections immediately before merging can be treated as flowing into the section immediately after merging.

The same applies to branches. When predicting the traffic situation, for example, information on two sections immediately after a branch and information on a section immediately before the branch can be correlated and predicted. That is, for example, all vehicles in the section immediately before the branch can be treated as flowing into any section immediately after the branch.

Further, as a method for realizing the prediction unit 113, for example, a LSTM (Long Short-Term Memory) network, which is a type of neural network that handles time-series data, can be considered.

When predicting traffic conditions from 5 minutes to 2 hours in 5-minute increments using an LSTM network, first, traffic condition data 122 and weather data 123 are collected for a certain period of time (for example, 5 minutes) in one section. Let m be the number of feature data updated within a given time period. If this data is prepared for 2 hours, m×24 pieces of input data will be required for one section. Furthermore, if predicted data representing the traffic situation for 2 hours every 5 minutes (for example, 3-value data representing the degree of traffic congestion) is set to be output, 24 predicted values will be output as well.

On the other hand, in order to handle information on multiple sections, for example, when considering a given section of interest and a total of n sections connected to it as input sections, the number of features of the input data described above is (m x 24). ×n, the output is 24×n. As for n, since the influence of each section of interest is limited, it is considered practical to set it to about 10 to 20 at most. Further, as the feature amount, in addition to traffic situation data 122 and weather data 123, event information and the like can be considered.

The display control unit 114 causes the display unit 14 to display various information.

The transmission control unit 115 transmits various information to external devices. For example, the transmission control unit 115 transmits the prediction result 126 by the prediction unit 113 to the road traffic control system 3, a vehicle traveling on the road, an information board installed on the roadside, or the like.

The input unit 13 is an input device that receives user operations on the traffic situation prediction device 1, and is, for example, a keyboard, a mouse, or the like.

The display unit 14 is realized by a liquid crystal display (LCD), an organic EL (electro-luminescence) display, or the like.

The communication unit 15 is a communication interface for communicating with external devices.

FIG. 6 is a flowchart showing an example of overall processing by the traffic situation prediction device 1 of the embodiment. In step S1, the processing unit 11 of the traffic situation prediction device 1 executes a learning process.
After step S1, in step S2, the processing unit 11 performs prediction processing.

FIG. 7 is a flowchart illustrating an example of learning processing by the traffic situation prediction device 1 of the embodiment. In step S11, the generation unit 112 acquires past traffic situation data from the storage unit 12.

Next, in step S12, the generation unit 112 acquires past weather data from the storage unit 12.

Next, in step S13, the generation unit 112 obtains the teacher data 124 from the storage unit 12.

Next, in step S14, the generation unit 112 generates the learning model 125 based on past traffic situation data, past weather data, teacher data 124, machine learning algorithms, and the like.

FIG. 8 is a flowchart showing an example of prediction processing by the traffic situation prediction device 1 of the embodiment. In step S21, the prediction unit 113 acquires current traffic situation data from the storage unit 12.

Next, in step S22, the prediction unit 113 acquires current weather data from the storage unit 12.

Next, in steps S23 to S25, the prediction unit 113 predicts the traffic situation for each section (step S24). Here, for example, the prediction unit 113 first uses the learning model 125, the maximum number of vehicles accommodated, the traffic situation data 122, and the weather data 123 for a first section that is predetermined as the starting point (bottleneck) of congestion. Predict traffic conditions based on etc.

Next, for a second section adjacent to the first section on the upstream side, the prediction unit 113 predicts the traffic situation prediction result 126 of the first section, the learning model 125, the maximum number of vehicles accommodated, and the traffic situation data 122. , the traffic situation is predicted based on the weather data 123 and the like. Thereafter, in the same manner, the prediction unit 113 predicts traffic conditions for the sections adjacent to each other on the upstream side one after another.

Next, in step S26, the transmission control unit 115 transmits the prediction result 126 of the traffic situation by the prediction unit 113 to the road traffic control system 3, vehicles traveling on the road, information boards installed on the roadside, etc. (Output.

Next, an example of the traffic situation prediction result 126 will be explained.
FIG. 9 is a graph showing an example of the prediction result 126 in the embodiment. For example, the prediction unit 113 predicts the amount of traffic congestion (the product of the average distance of the traffic jam and the length of the traffic jam) in time series for each of sections 1 to 10, and calculates the total (sum) of the amount of traffic congestion for each section. This allows the total amount of traffic congestion to be calculated. It can be seen that the total amount of traffic congestion varies depending on the time of day, and can be used as information to understand changes in traffic conditions. In addition to this, the prediction results 126 may be expressed in a format showing a heat map of the average speed for each time for each section.

FIG. 10 is a graph showing an example of the prediction result 126 in the embodiment. In FIGS. 10(a) to (c), the vertical axis represents the amount of traffic congestion, and the horizontal axis represents time.
In FIG. 10A, a line L11 shows the predicted value and a line L12 shows the correct answer (actual value) for 5 minutes after a certain section. It can be seen that the line L11 and the line L12 almost overlap, indicating that the prediction accuracy is high.

In addition, in FIG. 10(b), a line L21 shows the predicted value and a line L22 shows the correct answer (actual value) for the same section 30 minutes later. It can be seen that the line L21 and the line L22 almost overlap, indicating that the prediction accuracy is high.

In addition, in FIG. 10(c), a line L31 shows the predicted value and a line L32 shows the correct answer (actual value) for the same section two hours later. Line L31 and line L32 overlap considerably, and it can be seen that although the prediction accuracy is slightly lower than that after 5 minutes or 30 minutes, it is still quite high.

FIG. 11 is a graph showing an example of the prediction result 126 in the embodiment. In FIGS. 11(a) and 11(b), the vertical axis shows the congestion level (10 levels), and the horizontal axis shows time (one scale is 5 minutes).
In FIG. 11A, a line L41 indicates a predicted value and a line L42 indicates a correct answer (actual value) for a certain section. Line L41 and line L42 overlap in many parts, indicating that the prediction accuracy is high.

Further, in FIG. 11(b), a line L51 indicates a predicted value and a line L52 indicates a correct answer (actual value) for another section. Line L51 and line L52 overlap in many parts, indicating that the prediction accuracy is high.

In this way, the traffic situation prediction device 1 of the present embodiment predetermines the maximum number of vehicles that can be accommodated for each road section and learns how vehicle congestion that occurs on the road propagates upstream. By predicting the traffic situation sequentially from downstream to upstream for each section based on the learning model 125 generated by the machine learning algorithm, highly accurate prediction can be achieved.

For example, in the conventional technique, a method of constructing a different learning model for each road section did not have high prediction accuracy due to the large amount of calculations. According to the traffic situation prediction device 1 of this embodiment, by using a single learning model, the amount of calculation can be suppressed and prediction accuracy can be increased.

The highly accurate prediction result 126 of the traffic situation can be utilized by the road traffic control system 3 or vehicles traveling on the road, or can be displayed on an information board installed on the side of the road.

Further, by using the weather data 123 for learning and prediction, even more accurate prediction can be achieved.

In addition, traffic conditions include the distance of a traffic jam, the length of a traffic jam, the amount of traffic jam (the product of the average distance of a traffic jam and the length of a traffic jam), the travel time from a predetermined starting point to a predetermined end point on a road, and traffic congestion. It is highly convenient because it can predict various information such as the timing of resolution.

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 embodiments described above can be implemented in various forms, and various omissions, substitutions, and changes can be made without departing from the gist of the invention. The above-mentioned embodiments are included in the scope and gist of the invention, and are also included in the scope of the invention described in the claims and its equivalents.

For example, the target road is not limited to an expressway, but may be another road such as a general road.

Further, the number of divisions of the congestion degree may be made different between learning and prediction. For example, it is possible to use a 10-rank traffic jam level during learning, make a prediction using a 10-rank traffic jam level during prediction, and then convert it to a 5-rank or 3-rank traffic jam level.

DESCRIPTION OF SYMBOLS 1... Traffic situation prediction device, 2... Vehicle sensor, 3... Road traffic control system, 4... Weather data management device, 11... Processing part, 12... Storage part, 13... Input part, 14... Display part, 15... Communication Part, 111... Acquisition unit, 112... Generation unit, 113... Prediction unit, 114... Display control unit, 115... Transmission control unit, S... Traffic situation prediction system

Claims (3)

Changes in past traffic situation data for each road section and vehicle congestion occurring on the road, such as an increase or decrease in the length of the congestion for each section, cause a change in the length of the congestion on the entire main line. A machine learning algorithm that learns how traffic jams propagate upstream based on the idea of a generation unit that generates a learning model for predicting the amount ;
an acquisition unit that acquires current traffic situation data collected by a road information collection terminal regarding the road;
When predicting the amount of traffic congestion for each section, the learning model, the maximum number of vehicles accommodated, and the traffic situation data are used for the first section, which is the predetermined section and is predetermined as the section where the congestion starts. predicting the amount of traffic congestion based on
Regarding a second section adjacent to the first section on the upstream side, the traffic congestion is calculated based on the prediction result of the traffic situation in the first section, the learning model, the maximum number of vehicles accommodated, and the traffic situation data. A traffic situation prediction device comprising: a prediction unit that predicts traffic volume . The generation unit further generates the learning model using past weather data,
The acquisition unit further acquires current weather data,
The traffic situation prediction device according to claim 1, wherein the prediction unit further uses the weather data to predict the traffic situation in the first section and the traffic situation in the second section. Changes in past traffic situation data for each road section and vehicle congestion occurring on the road, such as an increase or decrease in the length of the congestion for each section, cause a change in the length of the congestion on the entire main line. A machine learning algorithm that learns how traffic jams propagate upstream based on the idea of a machine learning step of generating a learning model for predicting the amount ;
an acquisition step of acquiring current traffic situation data collected by a road information collection terminal for the road;
When predicting the amount of traffic congestion for each section, the learning model, the maximum number of vehicles accommodated, and the traffic situation data are used for the first section, which is the predetermined section and is predetermined as the section where the congestion starts. predicting the amount of traffic congestion based on;
Regarding a second section adjacent to the first section on the upstream side, the traffic congestion is calculated based on the prediction result of the traffic situation in the first section, the learning model, the maximum number of vehicles accommodated, and the traffic situation data. a prediction step for predicting the amount ;
Traffic situation prediction method including.

Images