Two-stage Calibration and Optimization Method for Microscopic Traffic Simulation Model Parameters Based on Neural Network Surrogate Models

Authors

• Yijia Liu, School of Civil Engineering and Transportation, Guangzhou University, Guangzhou 510700, China
• Chenjing Zhou, School of Civil Engineering and Transportation, Guangzhou University, Guangzhou 510700, China
• Dong Pan, School of Civil Engineering and Transportation, Guangzhou University, Guangzhou 510700, China
• Jian Rong, School of Civil Engineering and Transportation, Guangzhou University, Guangzhou 510700, China
• Yang Xiao, Guangxi Computing Center Co., Ltd., Nanning 530000, China

Abstract

A two-stage calibration and optimization method is proposed to address the problem that parameter calibration methods for microscopic traffic simulation models are time-consuming. In the first stage, a surrogate model based on neural networks is trained to establish the mapping relationship between model parameters and evaluation indicators, and a genetic algorithm (GA) is combined to screen candidate parameters. In the second stage, after obtaining the approximate optimal parameters, by employing this set of parameters as initial values, a genetic algorithm is re-executed by combining the real simulation model for optimization to further improve calibration accuracy. Experimental results show that the basic optimal parameter combination obtained in the first stage can achieve substantial reductions in calibration time while ensuring simulation accuracy, while the second stage enables quick convergence of GA based on the basic optimal parameter combination. The proposed method improves parameter optimization efficiency by 79.7% without compromising prediction accuracy, providing a foundation for the automation of microscopic traffic simulation model parameters.

Recommended Citation

Liu, Yijia; Zhou, Chenjing; Pan, Dong; Rong, Jian; and Xiao, Yang (2026) "Two-stage Calibration and Optimization Method for Microscopic Traffic Simulation Model Parameters Based on Neural Network Surrogate Models," Journal of System Simulation: Vol. 38: Iss. 6, Article 17.
DOI: 10.16182/j.issn1004731x.joss.25-FZ0635
Available at: https://dc-china-simulation.researchcommons.org/journal/vol38/iss6/17

First Page

1711

Last Page

1721

CLC

TP391.9; U491

Recommended Citation

Liu Yijia, Zhou Chenjing, Pan Dong, et al. Two-stage Calibration and Optimization Method for Microscopic Traffic Simulation Model Parameters Based on Neural Network Surrogate Models[J]. Journal of System Simulation, 2026, 38(6): 1711-1721.

Corresponding Author

Zhou Chenjing

DOI

10.16182/j.issn1004731x.joss.25-FZ0635