Intelligent control of urban traffic signal lights based on MAA3C and LSTM

Aiming at the problems that traditional traffic signal control methods are difficult to adapt to dynamic traffic flow and lack of multi intersection coordination, as well as the limitations that existing adaptive algorithms rely on artificial experience and can not deal with high-dimensional nonlinear state, this paper proposes an intelligent collaborative control model of multi intersection traffic signal. The model integrates the multi-agent asynchronous dominant actor critic algorithm and the long-term and short-term memory network, realizes the distributed decision-making and global parameter sharing through the asynchronous parallel training framework, and uses the long-term and short-term memory network to capture the temporal dependence of traffic flow. On this basis, the model combines the 12 dimensional road network level state space and dynamic multi-objective reward function to optimize the traffic signal control strategy. In the simulation scenario, the model reduced the average waiting time of vehicles from 85.7±4.1s in traditional timed control to 49.0±2.9s, shortened the average queue length from 18.3±2.2 vehicles to 10.4±1.3 vehicles, and improved the traffic efficiency by 33.6%. In the actual road network testing, the model reduced the average delay during morning rush hour from 65.4±5.7 seconds per vehicle to 38.5±3.1 seconds per vehicle, shortened the typical path travel time by 34.5%, and reduced carbon dioxide emissions by 19.3%. Overall, the research model outperforms traditional methods on control accuracy, synergy, and environmental benefits, providing an effective solution for intelligent control of urban traffic signals.