Surrogate assisted calibration framework for crowd model calibration

• Main Authors: Yi, Wenchao, Zhong, Jinghui, Tan, Singkuang, Cai, Wentong, Hu, Nan
• Other Authors: School of Computer Science and Engineering
• Format: Conference or Workshop Item
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Computer science and engineering; Crowd Simulation; Model Calibration
• Online Access: https://hdl.handle.net/10356/90266; http://hdl.handle.net/10220/48502
• Institution: Nanyang Technological University

Surrogate models are commonly used to approximate the multivariate input or output behavior of complex systems. In this paper, surrogate assisted calibration frameworks are proposed to calibrate the crowd model. To integrate the surrogate models into the evolutionary calibration framework, both the offline and online training based approaches are developed. The offline training needs to generate training set in advance, while the online training can adaptively build and re-build the surrogate model along the evolutionary process. Our simulation results demonstrate that the surrogate assisted calibration framework with the online training is effective and the surrogate model using artificial neural network obtains the best overall performance in the scenario evaluated in the case study.