Short-Term Speed Prediction for Expressway Considering Adaptive Selection of Spatiotemporal Dimensions and Similar Traffic Features

Abstract

Drivers on the expressway prefer to acquire more traffic information so they can select the best driving route and avoid a traffic jam. Therefore, it is necessary to conduct a study about real-time speed prediction and notify the drivers via their information screen. In this study, the candidate domains of spatial neighborhoods and time windows were first determined considering the spatiotemporal correlation among road sections. Then, a two-dimensional (2D) spatiotemporal matrix for the prediction model was developed. Based on the search characteristics of the nearest neighbor algorithm, we extracted the historical traffic features similar to the current traffic state and reconstructed a training set for each traffic state. Finally, the support vector regression algorithm was used to finish the short-term speed prediction. The case study was conducted using data collected from the expressway of Changchun, China. The space mean speed in each interval was calculated through matching the vehicle information between the two adjacent video detectors. The standard deviation was used to get rid of outliers. After comparison with four other models, the proposed model was proved to have the best performance in single-step as well as multistep prediction.