A Data-driven KPI Prediction Method for Vehicular Cyber Physical System

In driving process, providing accurate collision warning and effective advice about acceleration or deceleration in advance is beneficial to traffic safety. Furthermore, it will reduce the probability of vehicle collision. Much studies based on model or infrastructure have been proposed to solve this problem. However, their prediction accuracy is limited and few work utilize the large amount of historical data acquired in driving process. In this paper, a data-driven prediction method of key performance indicator (KPI)-the throttle data and brake data-is proposed. It associates Vehicular cyber physical system (VCPS) with kernel partial least squares (KPLS) algorithm and is implemented in Simulink environment. In order to achieve the prediction objective, getting adequate real-time data (e.g. velocity, location and acceleration) about vehicle motion is necessary. VCPS is an integrated system. It will provide us with comprehensive information about vehicle motion. KPLS is used to predict optimal input of throttle or brake. The other three fitting algorithms are compared with this method. The simulation results prove that the proposed method receives the best prediction results. It is effective and it could provide precise prediction and timely advice for driving.