High-Dimensional Simulation-Based Optimization for Low-Carbon Traffic Signal Control in Large-Scale Urban Networks

Large-scale urban traffic signal optimization presents substantial challenges arising from high-dimensional decision spaces, expensive microscopic simulations, and competing mobility and environmental objectives. This study formulates large-scale traffic signal timing as a high-dimensional bi-objective black-box optimization problem, in which average waiting time and total CO2 emissions are minimized simultaneously, thereby ensuring a symmetric and preference-neutral treatment of the two conflicting objectives without imposing predefined preference weights. To address the computational challenges, a Pareto–Hypervolume Guided Trust-Region Bayesian Optimization (PHV-TuRBO) method is proposed. The framework integrates Pareto-guided center selection, exploiting front sparsity and surrogate uncertainty, with hypervolume-driven trust-region adaptation to balance exploration and exploitation under limited simulation budgets. Experiments on 100-dimensional DTLZ1 and DTLZ2 benchmarks demonstrate superior convergence and Pareto-front approximation compared with NSGA-II, ParEGO, qEHVI-MOBO, and MORBO. Further validation on a large-scale Beijing traffic network involving 456 continuous decision variables and a budget of 5000 simulator evaluations demonstrates the effectiveness of PHV-TuRBO in identifying diverse efficiency–emission trade-offs. Representative Pareto-optimal solutions achieve substantial improvements in waiting time, travel time, queue length, stop frequency, mean speed, and CO2 emissions relative to the original signal plan. These results establish PHV-TuRBO as a sample-efficient approach to high-dimensional, multi-objective traffic signal optimization for sustainable urban mobility.