An optimization method for single intersection’s signal timing based on SARSA(λ) algorithm

Considering the operating characteristic of intersection signal control system, the reinforcement learning method was introduced to the intersection signal control system due to its powerful adaptability. By defining the state sets, action sets and reward function for a reinforcement learning agent, a new intersection signal control model was established based on reinforcement learning, and an optimization method for single intersection’s signal timing was presented using SARSA(λ) algorithm. Analyses show that this new signal timing optimization method can make the intersection saturation degree approach to the optimum section as close as possible, improve the efficiency of the crossing, increase utilization rate of green light, and decrease intersection traffic delay and number of stops.