Comparative Analysis of Data Fusion Algorithms for Microscopic Road Traffic State Estimation under Partial-Connected Driving Environment

Connected vehicle data are emerging data sources for estimating urban road performances. However, significant errors can occur at a low market penetration rate of connected vehicles. To improve the accuracy of microscopic road traffic state estimation in a partial-connected driving environment, a generic data fusion framework of roadside fixed detector data and connected vehicle data was proposed. Gaussian Process Regression, Extreme Learning Machine, Random Forest, and Support Vector Machine algorithms were applied to the framework. A comprehensive analysis was conducted and results show that the traffic state estimation accuracy of fusion data is better and the accuracy of the data fusion method is promoted progressively. This method eliminates both the defects of the uneven spatial distribution of fixed detector data and the insufficient representation of low-MPR connected vehicle data. GPR is the most suitable for the data fusion framework and is easily implemented for real-time traffic control and management applications.