Regional intersection signal control method based on PPO and graph convolutional neural network

Abstract

The invention discloses a regional intersection signal control method based on PPO and a graph convolutional neural network. The method comprises the following steps: constructing an intersection coordination control region, and determining the state, action and reward information of reinforcement learning and a feature matrix of the graph convolutional neural network; constructing a regional intersection layered signal control model; constructing a playback experience pool, and processing and extracting training data and test data; training a regional intersection layered signal control model; and carrying out overall planning and combined control on the regional intersections. According to the method, a multi-layer signal control model is established for a control area, and a multi-agent control model is constructed for a lower-layer model based on a PPO algorithm; and the upper-layer model performs overall coordination control on each intersection based on the graph convolutional neural network. By constructing a two-layer control structure, the calculation burden of a single-point control model is reduced, overall optimal control over the control area is achieved, and the vehicle running efficiency in the control area is improved.

Description

technical field [0001] The invention relates to the field of coordinated control of adaptive traffic signal lights, in particular to a method for controlling regional intersection signals based on PPO and graph convolutional neural networks. Background technique [0002] With the rise of artificial intelligence, computer vision technology continues to improve, and intelligent new algorithms such as reinforcement learning and graph convolutional neural networks continue to rise. Real-time acquisition of intersection traffic information and fast and stable calculation of adaptive traffic based on data-driven algorithms The signal control scheme becomes an effective new method to realize efficient and stable signal control at intersections. The multi-agent method combined with edge computing reduces the computing burden of the hardware, making the control more real-time and reliable. [0003] Reinforcement learning algorithm goes back and rewards in the continuous interaction ...

AI Technical Summary

Problems solved by technology

Generally speaking, the current application of reinforcement learning algorithms to the field of intersection signal control is mostly independent control of single-point intersections or intersections within a region, and it is rarely possible to coordinate the joint coordinated control of individual intersections as a whole, and the traffic network With its specific network topology, graph convolutional neural networks can be well applied to the overall control of regional intersection signal lights, and there are few researches on this aspect

Images