CN-117523867-B - Traffic signal lamp control method for multi-main-body heterogeneous road right allocation

CN117523867BCN 117523867 BCN117523867 BCN 117523867BCN-117523867-B

Abstract

The invention discloses a traffic signal lamp control method for multi-main-body heterogeneous road right distribution, which comprises the steps of firstly realizing traffic main body feature extraction and coding based on a graph neural network, then realizing road right distribution based on ADVANTAGE ACTOR-Critic (A2C) model, and finally realizing signal lamp control based on multi-traffic main body road right distribution. The invention can process different types of main bodies, can dynamically control the signal lamp, reduce the congestion degree of the road and improve the running efficiency of the road network.

Inventors

REN YILONG
CHENG MENGYUE
YU HAIYANG
FU XIANG

Assignees

北京航空航天大学合肥创新研究院

Dates

Publication Date: 20260508
Application Date: 20231124

Claims (6)

1. The traffic signal lamp control method for multi-main-body heterogeneous road right distribution is characterized by comprising the following steps of: Step 1, acquiring characteristic information of each traffic main body, and establishing a characteristic vector of each traffic main body based on the characteristic information; Taking each traffic body as a node and an adjacent matrix of any pair of traffic bodies as sides, respectively giving weight to each side to obtain a weight matrix, and thus constructing a traffic body diagram with the weight matrix; Processing the feature vector of each node based on the weight matrix in the traffic main map to obtain the initialized feature representation of each node; then, inputting the characteristic representation of each node into a graph neural network for processing to obtain a low-dimensional characteristic representation of each node, namely a traffic body; step 2, obtaining a road right allocation strategy based on an A2C algorithm model, wherein the process is as follows: constructing a state matrix by using the low-dimensional characteristic representation of each traffic body obtained in the step 1, and defining the selection of each traffic body on the road weight distribution scheme as a selection action; estimating the state matrix by adopting a strategy network in the A2C algorithm model to obtain probability distribution of various road right distribution actions of each traffic main body, wherein the road right distribution actions are possible traffic actions of the traffic main body, and the road right distribution action with the highest probability of each traffic main body can be obtained and selected according to the probability distribution matrix to obtain a road right distribution strategy; a return function is also constructed in the A2C algorithm model to represent the instant return obtained after the traffic main body collects the selection action under the state matrix, and an advantage function is constructed according to the difference value of the strategy network output and the value network output; The method comprises the steps of training and updating an A2C algorithm model, calculating strategy gradients according to a dominance function during the training and updating, processing the strategy gradients by using a gradient ascending method to update parameters of a strategy network, simultaneously calculating mean square errors of the value network by adopting accumulated returns and output of the value network, processing the mean square errors of the value network by using a gradient descending method to update the parameters of the value network, establishing an overall loss function finally based on gradient ascending processing results of the strategy gradients and gradient descending processing results of mean square errors of the value network, and ending the training when the overall loss function reaches convergence conditions, thereby outputting road weight distribution strategies of each traffic body through the A2C algorithm model after the training is completed; And step 3, obtaining the road weight distribution weight of each traffic body based on the road weight distribution strategy of each traffic body obtained in the step 2, and calculating the green light signal lamp duration of each direction light of the traffic signal lamp according to the road weight distribution weight of each traffic body.
2. The traffic light control method for multi-body heterogeneous road weight distribution according to claim 1, wherein in step 1, the weight of each side is calculated according to the distance and the interaction frequency between traffic bodies corresponding to the side.
3. The traffic light control method for multi-body heterogeneous road weight distribution according to claim 1, wherein in step 1, the initialized feature representation of each node is obtained after the feature vector of each node is linearly embedded and non-linearly mapped based on the weight matrix.
4. The traffic light control method for multi-body heterogeneous road weight distribution according to claim 3, wherein in step 1, the feature vector of each node is normalized, and then the feature vector of each normalized node is linearly embedded and mapped.
5. The traffic light control method for multi-body heterogeneous road weight distribution according to claim 1, wherein in step 1, the graph neural network processes the feature representation of each node by using a multi-layer graph convolution layer, each layer of graph convolution layer combines the feature representation of each node with the feature representation of an adjacent node to obtain a new feature representation, and the output of each layer in the multi-layer graph convolution layer is used as the input of the next layer, thereby obtaining the low-dimensional feature representation of each node.
6. The traffic light control method according to claim 1, wherein in step2, the real-time return is calculated by a return function, and then the real-time return is accumulated to obtain the accumulated return.

Description

Traffic signal lamp control method for multi-main-body heterogeneous road right allocation Technical Field The invention relates to the field of traffic control methods, in particular to a traffic signal lamp control method for multi-main-body heterogeneous road right distribution. Background Along with the development of autonomous traffic systems, the diversity and heterogeneity of traffic subjects are increasing, and various traffic subjects interfere with each other, so that the operation efficiency of the traffic system is seriously affected. In order to improve traffic operation efficiency, signal lamp control oriented to multi-main-body road right distribution becomes a current research hotspot. The traditional road right distribution signal lamp control method usually adopts a single strategy to distribute resources on the basis of fixed rules or preset strategies, the actual traffic environment is dynamically changed and is easily influenced by various factors such as time periods, weather and the like, such as the peak in the morning and evening, the common time period, holidays and daily weather, sunny weather and snowy weather, traffic flow speed are different, actual traffic conditions cannot be flexibly dealt with, meanwhile, the traditional method often does not fully consider the heterogeneity characteristics of traffic main bodies, such as different main bodies of buses, cars, trucks and bicycles, the speed and behavior patterns of different main bodies of the electric vehicles are different, and the optimization result of the traditional road right distribution cannot truly reflect the actual requirements. Therefore, a multi-traffic main body characteristic model needs to be defined, and a dynamic road right distribution network is constructed to control signal lamps so as to realize dynamic road right assignment for various traffic main bodies and improve road network operation efficiency. Disclosure of Invention The invention provides a traffic signal lamp control method for multi-main-body heterogeneous road right distribution, which aims to solve the problem that the traffic signal lamp control method based on road right distribution in the prior art is not suitable for multi-type traffic main bodies. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: A traffic signal lamp control method facing multi-main-body heterogeneous road right distribution comprises the following steps: Step 1, acquiring characteristic information of each traffic main body, and establishing a characteristic vector of each traffic main body based on the characteristic information; Taking each traffic body as a node and an adjacent matrix of any pair of traffic bodies as sides, respectively giving weight to each side to obtain a weight matrix, and thus constructing a traffic body diagram with the weight matrix; Processing the feature vector of each node based on the weight matrix in the traffic main map to obtain the initialized feature representation of each node; then, inputting the characteristic representation of each node into a graph neural network for processing to obtain a low-dimensional characteristic representation of each node, namely a traffic body; step 2, obtaining a road right allocation strategy based on an A2C algorithm model, wherein the process is as follows: constructing a state matrix by using the low-dimensional characteristic representation of each traffic body obtained in the step 1, and defining the selection of each traffic body on the road weight distribution scheme as a selection action; Estimating the state matrix by adopting a strategy network in the A2C algorithm model to obtain probability distribution of various road right distribution actions of each traffic main body, wherein the road right distribution actions are possible traffic actions of the traffic main body, such as straight running, left turning, right turning, stopping running and the like, and the road right distribution action with the highest probability of each traffic main body can be obtained and selected according to the probability distribution matrix to obtain a road right distribution strategy; a return function is also constructed in the A2C algorithm model to represent the instant return obtained after the traffic main body collects the selection action under the state matrix, and an advantage function is constructed according to the difference value of the strategy network output and the value network output; The method comprises the steps of training and updating an A2C algorithm model, calculating strategy gradients according to a dominance function during the training and updating, processing the strategy gradients by using a gradient ascending method to update parameters of a strategy network, simultaneously calculating mean square errors of the value network by adopting accumulated returns and output of the value network, processing the mean square erro