Multi-Intersection Traffic Signal Controls in Smart Cities Based on Graph Convolution Collaborative Communication
TL;DR
A multi-intersection graph topology is constructed and the Graph Convolutional Network (GCN) and Long Short-Term Memory (LSTM) are used to extract features from the graph topology and predict changes in traffic flow and deep reinforcement learning is applied to autonomously control traffic signals.
Abstract
To solve the poor control effect of the traffic signal control model due to low resource sharing and weak node connections in the collaborative communication module of the traffic signal control model, this study first constructed a multi-intersection graph topology to enhance the connections between nodes. Then, the Graph Convolutional Network (GCN) and Long Short-Term Memory (LSTM) are used to extract features from the graph topology and predict changes in traffic flow. Finally, deep reinforcement learning is applied to autonomously control traffic signals. This study tested these control methods. The results showed that after using this method to optimize the traffic signal control model at four multi-intersections in the Traffic Flow Forecasting dataset, the average waiting time of vehicles at these intersections was reduced by 56.4%, 53.0%, 65.6%, and 61.3%, respectively. Moreover, it could reduce the average queue length by 56.2%, 61.9%, 43.3%, and 52.4%, while improving traffic efficiency. The research method can accurately control traffic signals at multiple intersections, reduce average vehicle waiting time, and reduce traffic congestion.
