Memory-Guided Trust-Region Bayesian Optimization (MG-TuRBO) for High Dimensions

TL;DR

This paper introduces MG-TuRBO, a memory-guided trust-region Bayesian optimization method, which outperforms traditional methods in high-dimensional traffic simulation calibration tasks with noisy, nonconvex relationships.

Contribution

The paper proposes MG-TuRBO, a novel Bayesian optimization approach that effectively handles high-dimensional, noisy, and nonconvex calibration problems, demonstrating superior performance over existing methods.

Findings

MG-TuRBO outperforms genetic algorithms in high-dimensional calibration.

BOMs with adaptive strategies converge faster than traditional methods.

MG-TuRBO shows significant advantages in 84D traffic simulation calibration.

Abstract

Traffic simulation and digital-twin calibration is a challenging optimization problem with a limited simulation budget. Each trial requires an expensive simulation run, and the relationship between calibration inputs and model error is often nonconvex, and noisy. The problem becomes more difficult as the number of calibration parameters increases. We compare a commonly used automatic calibration method, a genetic algorithm (GA), with Bayesian optimization methods (BOMs): classical Bayesian optimization (BO), Trust-Region BO (TuRBO), Multi-TuRBO, and a proposed Memory-Guided TuRBO (MG-TuRBO) method. We compare performance on 2 real-world traffic simulation calibration problems with 14 and 84 decision variables, representing lower- and higher-dimensional (14D and 84D) settings. For BOMs, we study two acquisition strategies, Thompson sampling and a novel adaptive strategy. We evaluate performance using final calibration quality, convergence behavior, and consistency across runs. The results show that BOMs reach good calibration targets much faster than GA in the lower-D problem. MG-TuRBO performs comparably in our 14D setting, it demonstrates noticeable advantages in the 84D problem, particularly when paired with our adaptive strategy. Our results suggest that MG-TuRBO is especially useful for high-D traffic simulation calibration and potentially for high-D problems in general.