An optimization framework for multi-objective traffic signal control using intersection similarity and reinforcement learning

Urban traffic signal control faces dual challenges of sub-area partitioning difficulties and computational inefficiency. To address this, this study proposes a multi-objective optimization framework based on “similar intersection collaborative optimization.” This framework first introduces a novel Regional Intersection Similarity Aggregation (RISA) algorithm. By integrating static and dynamic multi-source attributes of intersections, it enables cross-physical location grouping of traffic networks, thereby reducing the computational complexity of the optimization problem by approximately 40 %. Second, we propose a Reinforcement Learning guided NSGA-II with Angular Diversity (RL-ADNSGA-II). By incorporating a diversity maintenance strategy based on vector angle and a dynamic crossover operator selection mechanism using Q-learning, this algorithm significantly enhances performance and convergence speed when solving non-convex Pareto fronts. Simulation results demonstrate that under peak traffic conditions, our proposed framework reduces average vehicle delay by 21.59 % and average queue length by 23.57 % compared to the traditional NSGA-II algorithm, validating its effectiveness and advanced capabilities in enhancing the overall operational efficiency of urban transportation networks.