Prediction and Feedback Assisted Evolutionary Algorithms for Scheduling Urban Traffic Signals

With the acceleration of urbanization, the traffic congestion issue is becoming more and more prominent in large cities. The effective scheduling of urban traffic signals becomes critical. This study proposes three novel prediction and feedback assisted evolutionary algorithms (PFAEAs) to address the urban traffic signal scheduling problem (UTSSP) with minimizing vehicle delays. First, we construct a mathematical model of UTSSP and design an improved evolutionary algorithm (EA) framework that integrates an eight-phase control strategy based on a vehicle movement relationship graph. Then, by combining a back-propagation neural network (BPNN) and meta-heuristics, we improve the prediction accuracy of the vehicle turning rate for generating high-quality initial solutions. Further, 12 problem-specific search operators (PSSOs) are designed to enhance the exploration capability of EA. Reinforcement learning (RL) algorithms, especially the Q-learning and Sarsa algorithms, are employed to select premium PSSOs dynamically for guiding the search direction of EA. Finally, for 18 cases with different scales, the proposed PFAEAs show significant advantages in reducing vehicle delays compared with the state-of-the-art algorithms. The results validate the competitiveness and practicality of the PFAEAs for UTSSP.