Gao, Qi (Columbia University), Di, Xuan (Columbia University)

Comparative Studies of Neural Operators for Traffic State Estimation

Scheduled for presentation during the Poster Session "Traffic prediction and estimation III" (ThAT13), Thursday, September 26, 2024, 10:30−12:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 3, 2024

Abstract

Traffic State Estimation (TSE) is essential for traffic management, planning, and control by predicting future traffic conditions such as flow density and average velocity. Traditional TSE methods, which employ either numerical techniques or deep learning, often require extensive prior knowledge and are too time-consuming for real-time applications. Recently, Neural Operators (NOs), a novel type of machine learning model, have been recognized for their efficiency in learning complex mappings between functional spaces. NOs offer significant advantages, including the elimination of the need for recalculations or retraining under varying initial or boundary conditions, and they are trained directly with data, without understanding of the underlying physical laws. This capability makes NOs particularly suitable for real-world, time-sensitive applications. Although NOs have been successfully used to model a range of phenomena from simple to complex partial differential equations, their ability to generalize across diverse functional spaces in traffic flow has not been thoroughly explored. This paper uses simulated traffic flow data with varying initial conditions and real-world fraffic flow data from the Next Generation SIMulation (NGSIM) dataset to evaluate the capacity of different NOs to approximate various traffic flow scenarios. Our study aims to guide transportation professionals in selecting suitable NOs and to encourage further advancements in NO research.