Multi-intersection traffic signal control method based on graph transformation network

The invention provides a multi-intersection traffic signal control method based on a graph transformation network, and the method comprises the steps: S1, constructing a traffic network environment, and constructing a road network graph; s2, reading road network diagram information, establishing a vehicle state set, and obtaining original data observed at each intersection; s3, performing initialization coding on the original data to obtain current state information of the intersection; s4, inputting the current state information of the intersection into the graph transformation network layer, capturing the spatial correlation of the road network graph by using the graph transformation network layer, updating the state information of the intersection, and outputting the state information of the intersection; s5, the updated intersection state information is input into an optimized deep Q network in the deep reinforcement learning model, an optimal signal control strategy is determined, and a traffic signal co