New MG-TuRBO Method Enhances Traffic Simulation Optimization

The study presents a new optimization method named Memory-Guided Trust-Region Bayesian Optimization (MG-TuRBO) for traffic simulation calibration. The method is evaluated against traditional genetic algorithms (GA) and other Bayesian optimization methods across two real-world traffic scenarios with differing dimensions (14 and 84 decision variables). Results indicate that MG-TuRBO achieves superior calibration quality faster than GA in high-dimensional settings, with a specific focus on adaptive acquisition strategies.

The implications of MG-TuRBO extend beyond traffic simulations, offering potential for advancements in calibration problems characterized by numerous parameters. With high-dimensional optimization remaining a significant challenge in various fields, MG-TuRBO could offer a new pathway for improving efficiency and accuracy in related applications, thereby enhancing the overall capability in complex systems modeling.