Cloud-Based Vehicle Velocity Prediction Based on Seasonal Autoregressive Integrated Moving Average Processes

Intelligent transportation systems (ITSs) and advanced driver assistance systems (ADASs) are considered as key technologies for improving road safety, fuel economy and driving comfort. For various ITSs and ADASs, e.g. for energy management systems in hybrid electric vehicles and adaptive cruise control systems, the velocity prediction of the ego vehicle and the target vehicles can substantially improve the system performance and is therefore an important building block. In this paper a novel concept for cloud-based vehicle velocity prediction using seasonal autoregressive integrated moving average (SARIMA) processes is proposed. The concept relies on collecting velocity profiles and estimating SARIMA processes using the collected velocity profiles for distinct road segments in a cloud (offboard). When a vehicle enters a road segment, the SARIMA model for the road segment is transmitted from the cloud to the vehicle for velocity prediction (onboard). The actual velocity profiles are transmitted from the vehicle back to the cloud for updating the SARIMA models. For quantifying the prediction uncertainty, an analytical formulation of the prediction bounds is provided. Such an analytical formulation is essential for robust control design but not available in most existing concepts. Throughout the paper the theoretical findings are evaluated utilizing real measurement data from highway driving. Moreover, the proposed concept is compared with a concept from the literature relying on artificial neural networks. The evaluation and comparison indicate that the concept based on SARIMA processes provides a good compromise between prediction accuracy and computational effort. Particularly real-time requirements on velocity prediction in many ITSs and ADASs can be satisfied.