XU, Bicheng (southeast university), Bai, Lu (Southeast University), Wong, Wai (University of Canterbury), Liu, Pan (Southeast University), Li, Yongtao (Hebei Provincial Communication Planning, Design and Research Ins)

A Novel Hybrid Induction Model for Traffic State Estimation of Arbitrary-Shaped Missing Area with Physical Constraints

Scheduled for presentation during the Regular Session "S30a-Intelligent Modeling and Prediction of Traffic Dynamics" (TH-LM-T30), Thursday, November 20, 2025, 11:50−12:10, Gold Coast

2025 IEEE 28th International Conference on Intelligent Transportation Systems (ITSC), November 18-21, 2025, Gold Coast, Australia

This information is tentative and subject to change. Compiled on October 18, 2025

Abstract

Traffic states derived from on-road fixed detectors commonly comprise missing information due to detector malfunctions, environmental conditions, objects occlusion and data quality issues. These challenges are inherent in the data acquisition system, and exert a significant impact on spatio-temporal perception of highway. Despite numerous existing traffic state estimation algorithms, concerns persist about estimating accurate, reliable and interpretable values across diverse missing data scenarios. This is crucial for two main reasons. Firstly, the missing spatio-temporal area is arbitrary-shaped with complex and random contours and shapes, and the method needs to learn the hidden patterns in observed data. Secondly, there are complex spatio-temporal correlations among the missing values at different locations, meanwhile, specific correlations exist among different missing values at the same location. To address these issues, a novel hybrid model integrating machine learning methods and macroscopic traffic flow models is proposed for missing density and speed in highway with partial observed density and speed. It works based on attention-based data extraction and deduction module, and the deduction process is conducted from boundary to the center of four directions. Additionally, a physics-informed loss function is designed. The loss function contains the traffic flow theory discrepancy, observed and estimated data absolute bias, and the discrepancy of estimated values across various estimation directions at the same location. Then, several real-world datasets are utilized to evaluate the accuracy and robustness of the proposed hybrid model. The estimation results reveal that the proposed method can effectively recover the missing density and speed.