JP7204525B2 - Traffic control device and traffic control method - Google Patents

Description

TECHNICAL FIELD Embodiments of the present invention relate to a traffic control device and a traffic control method.

2. Description of the Related Art Conventionally, there is a technique for predicting an accident on a road such as an expressway by performing machine learning using past traffic condition data as teacher data.

JP 2009-193212 A WO2018/180688 JP 2008-209345 A JP 2012-194718 A

However, in the above-described prior art, there is no effective method for determining whether or not the accident prediction result was actually valid.

Therefore, an object of the present embodiment is to provide a traffic control device and a traffic control method that can assist in determining whether or not an accident prediction result for roads based on machine learning is appropriate.

The traffic control device in the embodiment learns accident prediction based on past traffic condition data indicating past vehicle traffic conditions on roads , past accident data indicating past vehicle accidents, and a machine learning algorithm. a machine learning unit that generates a model; an acquisition unit that acquires traffic condition data indicating current traffic conditions of vehicles from a road information collection terminal for the road; and the learning model and the traffic condition data. a calculation unit that calculates an accident occurrence rate indicating the likelihood of a traffic accident occurring for each section of the road; the accident occurrence rate calculated by the calculation unit ; and a comparison unit that compares the data with the traffic condition data indicating the traffic condition and outputs a comparison result indicating the degree of similarity .

FIG. 1 is an overall configuration diagram of a traffic information management system according to an embodiment. FIG. 2 is a diagram schematically showing an example of graphically illustrated accident occurrence degree information according to the embodiment. FIG. 3 is a diagram schematically showing an example of traffic condition graphical information in the embodiment. FIG. 4 is a flowchart showing an example of processing by the traffic control device of the embodiment.

A traffic information management system according to an embodiment of the present invention will be described below with reference to the drawings.
FIG. 1 is an overall configuration diagram of a traffic information management system S according to an embodiment. In addition, in this embodiment, the case of a highway is taken as an example as a road. The traffic information management system S includes a traffic control device 1, a display unit 91 in the traffic control room 9, a terminal device 92, and the like. Note that the traffic control device 1 is implemented by a plurality of computer devices such as an information exchange server and a central processing unit, for example. described as being

The traffic information management system S recognizes traffic conditions, including the degree and types of congestion on highways (for example, natural congestion, accidental congestion, construction congestion, sightseeing congestion, etc.), traffic accidents, etc. It is a system that provides measures such as traffic restrictions and information such as warnings to expressway users.

The traffic control device 1 includes a mobile information terminal 5 such as a vehicle detector 2, a surveillance camera 3, an emergency telephone 4, a mobile phone 5A, a smart phone 5B, or a general vehicle 6 as an information collecting means (road information collecting terminal). Use equipment. In addition to these, the information gathering means may include another device such as a precipitation detector.

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

The monitoring camera 3 is installed on the roadside of the expressway, is a camera for photographing the expressway, and transmits the photographed image to the traffic control device 1 .

Emergency telephones 4 are provided on highway main lines, in tunnels, interchanges, service areas, parking areas, bus stops, emergency parking zones, and the like. In the event of a traffic accident, vehicle failure, or the like, the user can use the emergency telephone 4 to talk to an operator in the traffic control room 9 .

The mobile information terminal 5 is carried by a highway user. When a traffic accident, vehicle failure, or the like occurs, the user can use the mobile information terminal 5 to talk to an operator in the traffic control room 9 . Also, the mobile information terminal 5 can transmit and receive various information (traffic conditions, accident information, etc.) to and from the traffic control device 1 .

The in-vehicle device of the general vehicle 6 transmits probe information (information such as vehicle position, speed, acceleration, steering wheel operation, braking operation, etc.) to the traffic control device 1 .

The traffic control device 1 includes a processing section 11 , a storage section 12 , an input section 13 and a display section 14 . The traffic control device 1 also has a communication unit for communication with external devices and equipment, but illustration and description of the communication unit are omitted for the sake of simplicity.

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 control device 1 . The 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 uses the RAM as a work area to execute programs stored in the ROM, the storage unit 12, and the like. Details of the processing unit 11 will be described later.

The storage unit 12 is a storage device such as an HDD (Hard Disk Drive) or an SSD (Solid State Drive). The storage unit 12 stores road information, machine learning information, acquisition information, accident occurrence degree graphical information, traffic situation graphical information, and the like.

The road information is information about expressways, such as sections, number of lanes, interchanges, locations of parking areas, and the like.

The machine learning information is information related to machine learning for accident prediction and traffic jam prediction, and includes, for example, past traffic condition data of roads, machine learning algorithms for accident prediction and traffic jam prediction, learning models for accident prediction and traffic jam prediction, and the like.

Acquired information is information acquired from the vehicle detector 2, the monitoring camera 3, the emergency telephone 4, the mobile information terminal 5, the vehicle-mounted equipment of the general car 6, and the like.

The graphically illustrated information on the degree of occurrence of accidents and the graphically illustrated information on traffic conditions will be described later.

The input unit 13 is an input device that receives a user's operation on the traffic control device 1, such as a keyboard and a mouse.

The display unit 14 is implemented by a liquid crystal display (LCD), an organic EL (Electro-Luminescence) display, or the like.

The processing unit 11 includes a machine learning unit 111 , a calculation unit 112 , an acquisition unit 113 , a generation unit 114 , a comparison unit 115 and a display control unit 116 .

The machine learning unit 111 generates an accident prediction learning model (for example, a self-organizing map-based learning model described in Japanese Patent No. 6045846) based on past traffic condition data and an accident prediction machine learning algorithm for roads. In addition, the machine learning unit 111 generates a traffic jam prediction learning model based on a traffic jam prediction machine learning algorithm.

The calculation unit 112 calculates an accident occurrence rate indicating the likelihood of a traffic accident occurring for each section of the road based on the learning model of accident prediction and traffic condition data. The calculation unit 112 also calculates a degree of congestion that indicates the degree of congestion predicted for each section of the road, based on a traffic congestion prediction learning model or the like.

Acquisition unit 113 acquires traffic condition data for roads from information collecting means such as vehicle detector 2 .

The generation unit 114 generates accident occurrence degree graphically illustrated information, which is information that graphically illustrates the accident occurrence degree for each section of the road. The accident occurrence degree graphically represented information will be described below with reference to FIG. 2 .

FIG. 2 is a diagram schematically showing an example of graphically illustrated accident occurrence degree information according to the embodiment. The graphed information on the degree of accident occurrence shown in FIG. 2 is information representing the magnitude of the degree of accident occurrence as two axes that intersect the section (position) of the road and the time. Specifically, the vertical axis is time (minutes) (the direction of progress of time is downward), and the horizontal axis is road section (position, for example, kilometer post). Although detailed illustration is omitted, the horizontal axis direction includes several to several tens of sections (one section is about 2 km, for example) between points A and B. FIG. Also, the traveling direction of the vehicle is from right to left in FIG.

Regions M1 to M3 are regions calculated by calculation unit 112 as having a “medium” degree of accident occurrence. Areas L1 and L2 are areas where the calculation unit 112 has calculated that the degree of accident occurrence is “high”.

Returning to FIG. 1, the generation unit 114 generates traffic condition graphical information, which is information that graphically depicts traffic conditions, based on traffic condition data. The traffic condition graphical information will be described below with reference to FIG.

FIG. 3 is a diagram schematically showing an example of traffic condition graphical information in the embodiment. The traffic condition graphical information shown in FIG. 3 is information representing at least the degree of traffic congestion of vehicles as two axes that intersect road sections and time. The vertical and horizontal axes are the same as in FIG.

Areas N1 to N4 are areas of natural congestion. Areas A1 to A3 are accident congestion areas. A region C1 is a region of construction congestion. The area S1 is an area of sightseeing congestion.

2 and the traffic condition graphical information of FIG. 3 are considered to be correct when there is a positive correlation. In other words, it is correct that there is a large overlap between the areas in which the degree of occurrence of accidents is "medium" or "high" in FIG. 2 and the areas in which various types of congestion occur in FIG. This is because it is statistically known that there is a positive correlation between the occurrence of congestion and the occurrence of accidents on highways.

Therefore, by comparing the graphically illustrated information on the degree of occurrence of accidents in FIG. 2 and the graphically illustrated information on traffic conditions in FIG. can. For example, the comparison unit 115 compares the graphically illustrated information on the degree of occurrence of accidents and the graphically illustrated information on traffic conditions, and outputs a comparison result. Note that outputting the comparison result includes, for example, displaying the comparison result, storing the comparison result, and instructing other processing (such as re-machine learning) according to the comparison result.

Further, for example, the comparison unit 115 determines whether or not the degree of similarity between the graphically illustrated accident occurrence degree information and the graphically illustrated traffic condition information is equal to or greater than a predetermined threshold. Then, the machine learning unit 111 regenerates the learning model based on the past traffic condition data newer than the past traffic condition data according to the comparison result (if the similarity is less than the predetermined threshold).

In addition, when determining whether or not the degree of similarity between the graphically illustrated accident occurrence degree information and the graphically illustrated traffic condition information is equal to or greater than a predetermined threshold value, the comparison unit 115 performs, for example, a predetermined region size unit (for example, each section of the road). Either a first judgment method for judging information correspondence in units of corresponding cells every 5 minutes or a second judgment method for judging by pattern matching is performed.

In the first determination method, for example, when the graphed information on the occurrence of accidents and the graphed traffic condition information are compared on a cell-by-cell basis and the overlapping rate is greater than or equal to a predetermined ratio threshold, there is no need to regenerate the learning model for accident prediction. should be determined.

In addition, in the second determination method, for example, when the graphically illustrated information on the occurrence of accidents and the graphically illustrated information on traffic conditions are compared by pattern matching of image processing and the degree of similarity is equal to or greater than a predetermined similarity threshold, learning of accident prediction is performed. It is sufficient to determine that regeneration of the model is unnecessary.

When judging the degree of similarity between the graphical accident occurrence level information and the traffic situation graphical information, the comparison unit 115 weights the accident occurrence levels in the graphical accident occurrence level graphical information by "middle" and "high". may In that case, for example, the weight of the "high" accident occurrence level may be set higher than the weight of the "medium" accident occurrence level. Further, the comparison unit 115 may perform weighting according to the type of congestion in the graphically illustrated traffic condition information. In that case, for example, the weight of accidental congestion may be set higher than that of other congestions.

Returning to FIG. 1, the display control unit 116 causes the display unit 14 to display various information. For example, the display control unit 116 causes the display unit 14 to display the graphically illustrated accident occurrence degree information of FIG. 2 and the graphically illustrated traffic situation information of FIG. By doing so, the user who sees it can visually judge the degree of similarity between the two. At that time, the display control unit 116 may also display the comparison results (similarity, etc.) of the two side by side.

In addition, the generation unit 114 generates provision information (traffic information, alert information, detour information, etc.) for each road section or point based on traffic condition data or the like.

The transmission control unit 117 transmits various information to the external device. For example, the transmission control unit 117 transmits the provided information generated by the generation unit 114 to the vehicle-mounted device of the general vehicle 6, the mobile information terminal 5, the information board of the expressway, and the like, and displays the information.

Next, processing by the traffic control device 1 will be described with reference to FIG. FIG. 4 is a flowchart showing an example of processing by the traffic control device 1 of the embodiment. It is assumed that a learning model has already been generated for accident prediction before executing this process.

In step S<b>1 , the acquisition unit 113 acquires traffic condition data for roads from information collecting means such as the vehicle detector 2 .

Next, in step S2, the calculation unit 112 calculates the accident occurrence rate for each section of the road based on the accident prediction learning model and the traffic condition data.

Next, in step S3, the processing unit 11 determines whether or not the timing for determining the necessity of re-machine learning (for example, a fixed timing such as monthly, weekly, or daily) has arrived. If Yes, step S4. If No, proceed to step S8.

In step S4, the generation unit 114 generates accident occurrence degree graphical information (FIG. 2).

Next, in step S5, the generator 114 generates traffic condition graphical information (FIG. 3).

Next, in step S6, the comparing unit 115 determines whether the degree of similarity between the graphically illustrated accident occurrence degree information and the graphically illustrated traffic condition information is equal to or greater than a predetermined threshold value. If Yes, the process proceeds to step S8, and if No, the process proceeds to step S7.

In step S7, the machine learning unit 111 performs machine learning again. In other words, the machine learning unit 111 regenerates the accident prediction learning model using the new past traffic condition data. After step S7, the process ends.

In step S8, the generation unit 114 generates provision information (traffic information, alert information, detour information, etc.) for each road section or point based on the traffic condition data, etc., and the transmission control unit 117 The provided information is transmitted to and displayed on the vehicle-mounted device of the general car 6, the mobile information terminal 5, the information board of the expressway, or the like. After step S8, the process ends. Such a series of processes are performed, for example, every few minutes.

As described above, according to the traffic control device 1 of the present embodiment, by comparing the graphically illustrated information on the degree of occurrence of accidents (FIG. 2) and the graphically illustrated information on traffic conditions (FIG. 3), it is possible to predict accidents by machine learning about roads. It can assist in determining whether the results were reasonable.

Also, when the degree of similarity between the graphically illustrated information on the occurrence of accidents and the graphically illustrated information on traffic conditions is less than a predetermined threshold, machine learning for accident prediction is performed again, thereby improving the accuracy of machine learning for accident prediction. .

In addition, both the graphed information on the degree of accident occurrence and the graphed information on traffic conditions have two axes that intersect the road section and time, the former being information that expresses the degree of accident occurrence, and the latter being traffic congestion. By using the information indicating the degree of

Further, by using either the first determination method or the second determination method for comparison of the two, effective determination can be made.

Also, by displaying both side by side on the display unit 14, the user who sees it can visually judge the degree of similarity between the two.

In addition, it is possible to accurately determine the timing of re-machine learning for accident prediction. For example, when there is a change in the structure of the road (number of lanes, etc.), there are times when the traffic conditions also change, requiring machine learning again. At this time, the degree of similarity between the two decreases, and by determining No in step S6 of FIG. 4, machine learning can be performed again at an appropriate timing. At that time, by performing machine learning again based on new past traffic condition data, it is possible to reflect the influence of changes in the structure of the road on the learning model.

In the above description, for ease of explanation, the comparing unit 115 compares the graphically illustrated information on the degree of occurrence of accidents and the graphically illustrated information on traffic conditions. However, without being limited to this, the comparison unit 115 may compare the accident occurrence rate and the traffic condition data and output the comparison result for the same purpose.

In the prior art, there is a method of judging the validity of the result of accident prediction by machine learning, for example, using a simulator. However, with this method, it is difficult to select setting values for parameters related to the simulator, and it is difficult to make highly accurate judgments. On the other hand, according to the traffic control device 1 of the present embodiment, it is possible to easily perform a highly accurate judgment without the need for such troubles such as selection of setting values for parameters.

Although the embodiment of the present invention has been described above, the above embodiment is merely an example and is not intended to limit the scope of the invention. The above embodiments can be implemented in various forms, and various omissions, replacements, and modifications can be made without departing from the scope of the invention. The above embodiments are included in the scope and gist of the invention, and are included in the scope of the invention described in the claims and equivalents thereof.

For example, in the above-described embodiment, when the similarity is less than a predetermined threshold as a result of comparison, machine learning is automatically performed again, but the present invention is not limited to this. For example, the graphed information on the occurrence of accidents and the graphed information on traffic conditions are displayed side by side on the display unit 14, and the user visually judges the degree of similarity between the two. may be input to instruct to re-execute machine learning.

Also, the target roads are not limited to highways, and may be other roads such as general roads.

Further, when performing machine learning for accident prediction, past accident data excluding accident data unrelated to traffic congestion may be used as teacher data.

In the above-described embodiment, the method using the self-organizing map is exemplified as the machine learning method for accident prediction. However, as a method of machine learning for accident prediction, various methods other than the method using the self-organizing map are conceivable. For example, as a relatively simple method, there is a method of holding (accumulating) past traffic situation data at the time of an accident and simply comparing it with the current traffic situation data, or statistically processing a combination of past traffic situation data at the time of an accident. A method of clustering and generating traffic condition data for cases similar to the time of an accident can be considered. As another method, for example, a method using other multivariate analysis such as a Paysian network is also conceivable.

DESCRIPTION OF SYMBOLS 1... Traffic control apparatus, 2... Vehicle detector, 3... Surveillance camera, 4... Emergency telephone, 5... Mobile information terminal, 6... General car, 9... Traffic control room, 11... Processing part, 12... Storage part, 13 Input unit 14 Display unit 91 Display unit 92 Terminal device 111 Machine learning unit 112 Calculation unit 113 Acquisition unit 114 Generation unit 115 Comparison unit 116 Display control unit

Claims (8)

a machine learning unit that generates an accident prediction learning model based on past traffic condition data indicating past vehicle traffic conditions on a road , past accident data indicating past vehicle accidents, and a machine learning algorithm ; ,
an acquisition unit for acquiring traffic condition data indicating the current traffic condition of the vehicle from the road information collection terminal for the road;
a calculation unit that calculates an accident occurrence rate indicating the likelihood of a traffic accident occurring for each section of the road based on the learning model and the traffic condition data;
a comparison unit that compares the accident occurrence rate calculated by the calculation unit with the traffic condition data indicating the current traffic conditions of the vehicle obtained by the acquisition unit, and outputs a comparison result indicating the degree of similarity ;
A traffic control device. The past traffic condition data and the traffic condition data include at least one or more of vehicle traffic volume, average speed, vehicle density, and occupancy rate data,
The traffic control device according to claim 1. The machine learning unit performs the learning based on the past traffic condition data and the past accident data newer than the past traffic condition data and the past accident data when the similarity of the comparison result of the comparing unit is lower than a predetermined threshold. regenerate the model,
The traffic control device according to claim 1. Generating accident occurrence degree graphical information, which is information graphically representing the accident occurrence degree for each section of the road, and generating traffic condition graphical information, which is information graphically representing the traffic condition based on the traffic condition data. further comprising a generator for generating
The comparison unit compares the graphically illustrated information on the degree of accident occurrence and the graphically illustrated information on traffic conditions, and outputs a degree of similarity as the comparison result.
The traffic control device according to claim 1. The accident occurrence degree graphically represented information is information representing the magnitude of the accident occurrence degree as two axes that intersect the road section and time,
The traffic condition graphical information is information that represents at least the degree of vehicle congestion as two axes that intersect the road section and time,
The traffic control device according to claim 4 . When comparing the graphed accident occurrence degree information and the graphed traffic condition information, the comparison unit includes a first judgment method for judging the correspondence of the information in units of a predetermined area size, and a second judgment method for judging by pattern matching. perform one of the two judgment methods,
The traffic control device according to claim 5 . further comprising a display control unit that performs control to display the graphically illustrated accident occurrence degree information and the graphically illustrated traffic situation information side by side on a display unit;
The traffic control device according to claim 4 . The machine learning unit generates an accident prediction learning model based on past traffic condition data indicating past vehicle traffic conditions on roads , past accident data indicating past vehicle accidents, and a machine learning algorithm. a machine learning step that
an acquisition step in which an acquisition unit acquires traffic condition data indicating current vehicle traffic conditions for the road from a road information collection terminal;
a calculation step in which the calculation unit calculates an accident occurrence rate indicating the likelihood of a traffic accident occurring for each section of the road based on the learning model and the traffic condition data;
a comparison step in which the comparison unit compares the accident occurrence rate calculated by the calculation unit with the traffic condition data indicating the current traffic conditions of the vehicle, and outputs a comparison result indicating the degree of similarity ;
traffic control methods, including

Images