Traffic State Estimation on Swedish Highways : Model Comparison using Multisource Data

Abstract: Due to the escalating demand for traffic information and management, the significance of traffic state estimation, which involves the assessment of traffic conditions on road segments with limited measurement data, is increasing. Two primary estimation methods are model-driven and data-driven. The former uses traffic flow models, while the latter relies on extensive historical data to explore relationships between traffic states. Due to the uninterrupted nature of highway traffic flow, conventional model-driven approach is adopted in the study to estimate traffic information from sensing data. Data-driven approach is applied to enhance the estimation results. The project mainly focuses on comparing the estimation performance between the Particle Filter and the commonly used Extended Kalman Filter. These two methods are implemented in combination with two typical traffic flow models: Cell Transmission Model and METANET. Moreover, the project investigates the potential of using vehicle-to-everything (V2X) data in traffic state estimation, either alone or combined with traditional inductive loop detector (ILD) data. Being an emerging traffic data source, V2X communication has been recently installed and tested on the motorways near Stockholm. This study provides essential insights into how V2X data can benefit existing traffic information estimation and its performance. To evaluate the models mentioned above, the estimation algorithms and traffic flow models are implemented in a self-developed platform, which may be useful for further work. Results from simulation experiments show that Particle Filter can carry out traffic state estimation with comparable accuracy to Extended Kalman Filter. While standalone V2X speed data falls short, effective fusion methods are implemented to combine both data types, ultimately achieving the desired accuracy. These fusion methods encompass direct filtering, weighted averaging, and linear regression. Future investigations could broaden their scope to include new data sources, such as unmanned aerial vehicles (UAVs), and delve into advanced data fusion techniques, such as deep learning.