Traffic flow prediction method based on deep learning

By constructing a multi-dimensional traffic flow prediction model and configuring intelligent controllers on road sections for edge computing, the problems of insufficient accuracy and low computing efficiency in the prior art are solved, and more accurate and efficient traffic flow prediction is achieved.

CN119028153APending Publication Date: 2024-11-26BEIJING TECH & BUSINESS UNIV
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411136296.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-19
Publication Date
2024-11-26

AI Technical Summary

Technical Problem

Existing deep learning models cannot accurately predict traffic flow when weather changes, and only consider the historical traffic data of the current section, resulting in insufficient prediction accuracy. At the same time, cloud servers have large computing volume and low computing efficiency.

Method used

Build a traffic flow prediction model based on neural network, combine historical traffic flow data, weather data, spatiotemporal feature data and traffic flow data of upstream sections for training, and configure intelligent controllers for edge computing on each section for lightweight processing and Knowledge distillation to achieve multi-dimensional traffic flow prediction.

Benefits of technology

It improves the accuracy and speed of traffic flow prediction, alleviates the computing pressure of cloud servers, and reduces the computing and storage needs of smart controllers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119028153A_ABST
    Figure CN119028153A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of traffic flow prediction, in particular to a traffic flow prediction method based on deep learning, which comprises the following steps: constructing a traffic flow prediction model based on a neural network; combining actual traffic flow data, weather data and spatial-temporal characteristic data of different road sections in a target road network at different historical moments and traffic flow data of an upstream road section to serve as a data set sample, and training a traffic flow prediction model; wherein the spatial-temporal characteristic data of the current road section comprises holiday and festival time characteristics at the current moment, road section characteristics and surrounding supporting facility characteristics; the traffic flow data of the upstream road section comprises actual traffic flow data of the upstream road section and the number of vehicles which are about to drive into the current road section from the upstream road section at the next moment; and predicting the traffic flow data of the target road section at the next moment based on the trained traffic flow prediction model. According to the invention, rapid and accurate prediction of the traffic flow of each road section can be realized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of traffic flow prediction, and more specifically to a traffic flow prediction method based on deep learning. Background Art

[0002] With the sharp increase in urban motor vehicle data, urban traffic congestion has intensified. Accurate traffic prediction can help optimize traffic signal control in real time, thereby improving road utilization, reducing traffic congestion, and ultimately improving urban traffic operation efficiency.

[0003] Deep learning can deeply mine and capture the corresponding rules of data sequences from massive data. Deep learning has also been applied to road traffic flow prediction, but the current deep learning model can only accurately predict traffic flow under normal weather conditions. Once the weather changes, the prediction results will no longer be accurate. At the same time, the current deep learning model only considers the historical traffic flow data of the current road section, and considers fewer dimensions. Its prediction accuracy still needs to be improved.

[0004] In addition, currently, a cloud server is usually used to make a unified prediction of the traffic flow of each section in the target road network. Due to the huge amount of data, the cloud server's computing power is too large and the computing efficiency is not high. Therefore, how to improve the prediction rate has also become an issue that technical personnel in this field need to consider. Summary of the invention

[0005] In view of this, the present invention provides a traffic flow prediction method based on deep learning, which can solve at least one technical problem in the background technology.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] A traffic flow prediction method based on deep learning includes the following steps:

[0008] Construct a traffic flow prediction model based on neural network;

[0009] The actual traffic flow data, weather data, spatiotemporal feature data of different sections in the target road network at different historical moments, and the traffic flow data of the upstream sections are combined as data set samples to train the traffic flow prediction model; the spatiotemporal feature data of the current section include: the holiday time characteristics, section characteristics and surrounding supporting facilities characteristics at the current moment; the traffic flow data of the upstream section include: the actual traffic flow data of the upstream section and the number of vehicles that will enter the current section from the upstream section at the next moment;

[0010] Obtain the spatiotemporal characteristic data, weather forecast data, and traffic flow data of the upstream section of the target section at the current moment, input the trained traffic flow prediction model, and predict the traffic flow data of the target section at the next moment.

[0011] Furthermore, the method further includes: performing lightweight processing on the neural network; the lightweight processing process includes:

[0012] Sort the filters and convolution kernels according to the sum of the binary norms, and prune the filters and convolution kernels that are ranked lower;

[0013] The knowledge distillation method is used to transfer the effect of the pruned network.

[0014] Furthermore, an intelligent controller is configured on each road section, and the trained traffic flow prediction model is configured in each intelligent controller, and all intelligent controllers in the target road network interact with the cloud server for data;

[0015] The intelligent controller on each road section obtains weather forecasts and holiday time characteristics through the network, and collects the actual traffic flow data of the road section at different times and sends it to the cloud server;

[0016] The cloud server is associated with the map software; at every preset time interval, the cloud server publishes the section characteristics of each section in the target road network, the characteristics of the surrounding supporting facilities, the actual traffic flow data of the upstream section, and the number of vehicles that will enter the current section from the upstream section at the next moment to the intelligent controller of the corresponding section.

[0017] Furthermore, the upstream section of a certain road section is determined in the following manner: a road section having a topological connection relationship with the current road section is determined according to the direction of the lane line as the upstream section of the road section.

[0018] Furthermore, the road section characteristics of the current road section include: one or more of the inherent characteristics of the road section, traffic accidents and road construction, which are represented as L1, L2 and L3 respectively; wherein the inherent characteristics of the road section include speed limit information, road width, number of lanes and terrain information;

[0019] The surrounding supporting facilities of the current road section include: schools, hospitals, commercial streets, office buildings, scenic spots or residential buildings, which are represented as M1, M2, M3, M4, M5 and M6 respectively;

[0020] The holiday time characteristics of the current road section include: working hours on weekdays, morning peak hours and evening peak hours on weekdays, other time on weekdays, daytime time on holidays and other time on holidays, which are represented as T1, T2, T3, T4 and T5 respectively.

[0021] Furthermore, weather data includes weather conditions and wind conditions. Weather conditions are divided into four levels: level one is sunny or cloudy, level two is light rain, level three is moderate rain or mist, and level four is heavy rain or fog, which are represented by W1, W2, W3, and W4 respectively;

[0022] Wind conditions are divided into three levels: calm, light breeze and strong wind, denoted as A1, A2 and A3 respectively.

[0023] Furthermore, the representation form of each data set sample is {W, A, L, M, T, Q1, Q2, Q3}, where W represents the weather conditions of the current section at the current moment, A represents the wind conditions of the current section at the current moment, L represents the section characteristics of the current section at the current moment, M represents the characteristics of the surrounding supporting facilities of the current section, T represents holiday characteristics, Q1 represents the actual traffic flow data of the upstream section, Q2 represents the number of vehicles that will enter the current section from the upstream section at the next moment, and Q3 represents the actual traffic flow data of the current section at the current moment.

[0024] Furthermore, the number of vehicles that will enter the current section from the upstream section at the next moment is determined as follows:

[0025] The vehicle’s driving trajectory at different times in the past is recorded through the vehicle terminal and shared to the cloud server;

[0026] If the vehicle turns on the navigation service, the cloud server obtains the navigation information of each vehicle on each road section in the target road network at the current moment through the map software, and determines the next driving section of each vehicle based on the navigation information;

[0027] If the vehicle does not have navigation service turned on, the next driving section is determined based on the vehicle's historical driving trajectory;

[0028] According to the next travel section of each vehicle in each section, the number of vehicles entering each section from the upstream section at the next moment is determined.

[0029] Furthermore, the basic network architecture of the traffic flow prediction model adopts LSTM network.

[0030] It can be seen from the above technical solutions that, compared with the prior art, the present invention has the following beneficial effects:

[0031] When predicting traffic flow, the present invention comprehensively considers the spatiotemporal characteristic data such as the holiday time characteristics of the current road section, the road section characteristics and the characteristics of the surrounding supporting facilities, as well as the traffic flow information of the upstream road section, and trains the traffic flow model from multiple dimensions in combination with weather conditions, deeply explores the inherent laws of the traffic flow data of each road section, and captures the impact of different weather on traffic flow, thereby improving the accuracy of model prediction.

[0032] The present invention configures an intelligent controller on each road section to perform edge calculation on the traffic flow of the road section, which can alleviate the computing pressure of the cloud server and improve the prediction rate; at the same time, the neural network is lightweight processed to reduce the computing and storage requirements of the intelligent controller. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0034] Figure 1 A flow chart of a traffic flow prediction method based on deep learning provided by the present invention;

[0035] Figure 2 Schematic diagram of data interaction between the intelligent controller and the cloud server provided by the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] like Figure 1 As shown, the embodiment of the present invention discloses a traffic flow prediction method based on deep learning, comprising the following steps:

[0038] S1. Construct a traffic flow prediction model based on neural network;

[0039] S2. Combine the actual traffic flow data, weather data, spatiotemporal feature data of different sections in the target road network at different historical moments, and the traffic flow data of the upstream section as data set samples to train the traffic flow prediction model; the spatiotemporal feature data of the current section includes: the holiday time characteristics, section characteristics, and surrounding supporting facilities characteristics at the current moment; the traffic flow data of the upstream section includes: the actual traffic flow data of the upstream section and the number of vehicles that will enter the current section from the upstream section at the next moment;

[0040] S3. Obtain the spatiotemporal characteristic data, weather forecast data, and traffic flow data of the target road section at the current moment, input the trained traffic flow prediction model, and predict the traffic flow data of the target road section at the next moment.

[0041] The above steps are further explained below.

[0042] S1. Build a traffic flow prediction model based on the LSTM network as the basic network architecture. In order to reduce the computing and storage requirements, the model is more compatible with the hardware and more hardware-friendly. It also includes: lightweight processing of the neural network; the lightweight processing process includes:

[0043] Sort the filters and convolution kernels according to the sum of the binary norms, and prune the filters and convolution kernels that are ranked lower;

[0044] The knowledge distillation method is used to transfer the effect of the pruned network.

[0045] S2. Combine the actual traffic flow data, weather data, spatiotemporal characteristic data of different sections in the target road network at different historical times, and the traffic flow data of upstream sections as data set samples to train the traffic flow prediction model.

[0046] Among them, the spatiotemporal characteristic data of the current road section include: road section characteristics, holiday time characteristics and surrounding supporting facilities characteristics. Specifically, the road section characteristics include: one or more of the inherent characteristics of the road section, traffic accidents and road construction, which are represented as L1, L2 and L3 respectively; among them, the inherent characteristics of the road section include speed limit information, road width, number of lanes and terrain information, which can be represented as L11, L12, L13 and L14.

[0047] The surrounding supporting facilities characteristics of the current road section include: schools, hospitals, commercial streets, office buildings, scenic spots or residential buildings, which are represented by M1, M2, M3, M4, M5 and M6 respectively; since schools, hospitals, commercial streets, office buildings, scenic spots or residential buildings have their own independent characteristics, this will cause the road sections near them to have different traffic flow patterns.

[0048] The holiday time characteristics of the current road section include: working hours on weekdays, morning peak hours and evening peak hours on weekdays, other time on weekdays, daytime time on holidays and other time on holidays, which are represented as T1, T2, T3, T4 and T5 respectively.

[0049] Weather data includes weather conditions and wind conditions. Weather conditions are divided into four levels: level one is sunny or cloudy, level two is light rain, level three is moderate rain or mist, and level four is heavy rain or fog, which are represented by W1, W2, W3, and W4 respectively;

[0050] Wind conditions are divided into three levels: calm, light breeze and strong wind, denoted as A1, A2 and A3 respectively.

[0051] Traffic flow data of the upstream road section includes: the actual traffic flow data of the upstream road section and the next road section information of each vehicle. The upstream road section usually refers to: the road section that has a topological connection relationship with the current road section according to the direction of the lane line, which is used as the upstream road section of the road section. Taking a certain road section as a reference, in the longitudinal direction, the upper road section with the same lane line direction as the road section is regarded as one of the upstream roads of the road section, and in the transverse direction, each road section intersecting with the road section is also regarded as the upstream road section of the road section.

[0052] Among them, the actual traffic flow data of the upstream section is directly collected and calculated by the on-site hardware, which is represented by Q1.

[0053] The traffic flow data of the vehicles entering the current road section from the upstream road section at the next moment is represented by Q2, which is determined according to the next entry road section information of each vehicle in the upstream road section. The specific determination method is:

[0054] The vehicle’s driving trajectory at different times in the past is recorded through the vehicle terminal and shared to the cloud server;

[0055] If the vehicle turns on the navigation service, the cloud server obtains the navigation information of each vehicle on each road section in the target road network at the current moment through the map software, and determines the next driving section of each vehicle based on the navigation information;

[0056] If the vehicle does not have navigation service turned on, the next driving section is determined based on the vehicle's historical driving trajectory;

[0057] According to the next travel section of each vehicle in each section, the number Q2 of vehicles entering each section from the upstream section at the next moment is determined.

[0058] So far, the representation of each data set sample is {W, A, L, M, T, Q1, Q2, Q3}, where W represents the weather conditions of the current section at the current moment, A represents the wind conditions of the current section at the current moment, L represents the section characteristics of the current section at the current moment, M represents the characteristics of the surrounding supporting facilities of the current section, T represents holiday characteristics, Q1 represents the actual traffic flow data of the upstream section, Q2 represents the number of vehicles that will enter the current section from the upstream section at the next moment, and Q3 represents the actual traffic flow data of the current section at the current moment.

[0059] Finally, the traffic flow prediction model is trained based on the above dataset samples. The mean absolute error (MAE) is used as the loss function during the training process, and the Adam algorithm is used for training. The traffic flow is predicted based on the trained model, and the mean absolute percentage error (MAPE), mean absolute error (MAE) and root mean square error (RMSE) are used to evaluate the model and compare the prediction performance of each model.

[0060] S3. After the traffic flow prediction model is trained, the spatiotemporal characteristic data, weather forecast data, and traffic flow data of the target section at the current moment are obtained, and the trained traffic flow prediction model is input to predict the traffic flow data of the target section at the next moment.

[0061] Specifically, if Figure 2 As shown, an intelligent controller is configured on each road section, and the trained traffic flow prediction model is configured in each intelligent controller, and all intelligent controllers in the target road network interact with the cloud server for data;

[0062] The intelligent controller on each road section obtains weather forecast and time calendar information through the network, and determines the time characteristics of holidays based on the time calendar information. At the same time, each intelligent controller is also equipped with a traffic flow meter, which counts the actual traffic flow data of the road section at different times and sends it to the cloud server;

[0063] The cloud server is associated with the map software and communicates with each vehicle terminal; at each preset interval, the cloud server publishes the section characteristics of each section in the target road network, the characteristics of the surrounding supporting facilities, the actual traffic flow data of the upstream section, and the number of vehicles that will enter the current section from the upstream section at the next moment to the intelligent controller of the corresponding section.

[0064] The present invention can quickly predict the traffic flow of each road section through edge-cloud collaborative computing, and simultaneously consider multiple dimensions of information such as the current road section, upstream road section and weather, thereby improving the accuracy of the prediction results.

[0065] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0066] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A traffic flow prediction method based on deep learning, characterized in that: The following steps are involved: Construct a traffic flow prediction model based on neural network; The actual traffic flow data, weather data, spatiotemporal feature data of different sections in the target road network at different historical moments, and the traffic flow data of the upstream sections are combined as data set samples to train the traffic flow prediction model; the spatiotemporal feature data of the current section include: the holiday time characteristics, section characteristics and surrounding supporting facilities characteristics at the current moment; the traffic flow data of the upstream section include: the actual traffic flow data of the upstream section and the number of vehicles that will enter the current section from the upstream section at the next moment; Obtain the spatiotemporal characteristic data, weather forecast data, and traffic flow data of the upstream section of the target section at the current moment, input the trained traffic flow prediction model, and predict the traffic flow data of the target section at the next moment.

2. The traffic flow prediction method based on deep learning according to claim 1 is characterized in that: It also includes: lightweight processing of the neural network; the lightweight processing process includes: Sort the filters and convolution kernels according to the sum of the binary norms, and prune the filters and convolution kernels that are ranked lower; The knowledge distillation method is used to transfer the effect of the pruned network.

3. The traffic flow prediction method based on deep learning according to claim 1 is characterized in that: An intelligent controller is configured on each road section, and the trained traffic flow prediction model is configured in each intelligent controller. All intelligent controllers in the target road network interact with the cloud server for data. The intelligent controller on each road section obtains weather forecasts and holiday time characteristics through the network, and collects the actual traffic flow data of the road section at different times and sends it to the cloud server; The cloud server is associated with the mapping software; At every preset time interval, the cloud server publishes the section characteristics of each section in the target road network, the characteristics of the surrounding supporting facilities, the actual traffic flow data of the upstream section, and the number of vehicles that will enter the current section from the upstream section at the next moment to the intelligent controller of the corresponding section.

4. The traffic flow prediction method based on deep learning according to claim 1 is characterized in that: The upstream section of a certain road section is determined in the following manner: a section having a topological connection relationship with the current road section is determined according to the direction of the lane line as the upstream section of the road section.

5. The traffic flow prediction method based on deep learning according to claim 1, characterized in that: The road section characteristics of the current road section include: one or more of the inherent characteristics of the road section, traffic accidents and road construction, which are represented as L1, L2 and L3 respectively; wherein the inherent characteristics of the road section include speed limit information, road width, number of lanes and terrain information; The surrounding supporting facilities of the current road section include: schools, hospitals, commercial streets, office buildings, scenic spots or residential buildings, which are represented as M1, M2, M3, M4, M5 and M6 respectively; The holiday time characteristics of the current road section include: working hours on weekdays, morning peak hours and evening peak hours on weekdays, other time on weekdays, daytime time on holidays and other time on holidays, which are represented as T1, T2, T3, T4 and T5 respectively.

6. The traffic flow prediction method based on deep learning according to claim 5 is characterized in that: Weather data includes weather conditions and wind conditions. Weather conditions are divided into four levels: level one is sunny or cloudy, level two is light rain, level three is moderate rain or mist, and level four is heavy rain or fog, which are represented by W1, W2, W3, and W4 respectively; Wind conditions are divided into three levels: calm, light breeze and strong wind, denoted as A1, A2 and A3 respectively.

7. The traffic flow prediction method based on deep learning according to claim 6 is characterized in that: The representation form of each data set sample is {W, A, L, M, T, Q1, Q2, Q3}, where W represents the weather conditions of the current section at the current moment, A represents the wind conditions of the current section at the current moment, L represents the section characteristics of the current section at the current moment, M represents the characteristics of the surrounding supporting facilities of the current section, T represents holiday characteristics, Q1 represents the actual traffic flow data of the upstream section, Q2 represents the number of vehicles that will enter the current section from the upstream section at the next moment, and Q3 represents the actual traffic flow data of the current section at the current moment.

8. The traffic flow prediction method based on deep learning according to claim 3 is characterized in that: The number of vehicles that will enter the current section from the upstream section at the next moment is determined as follows: The vehicle’s driving trajectory at different times in the past is recorded through the vehicle terminal and shared to the cloud server; If the vehicle turns on the navigation service, the cloud server obtains the navigation information of each vehicle on each road section in the target road network at the current moment through the map software, and determines the next driving section of each vehicle based on the navigation information; If the vehicle does not have navigation service turned on, the next driving section is determined based on the vehicle's historical driving trajectory; According to the next travel section of each vehicle in each section, the number of vehicles entering each section from the upstream section at the next moment is determined.

9. The traffic flow prediction method based on deep learning according to claim 1, characterized in that: The basic network architecture of the traffic flow prediction model adopts LSTM network.