A KPI prediction approach with JITL for vehicular Cyber Physical System

Intelligent transportation is a hot research field in Cyber-Physical System (CPS). In order to improve the driving safety, many studies have been conducted to predict collision probability and send out warning signal timely. However, most of these studies are model based with limited prediction accuracy. Moreover, the abundant historical data is leave-off. In this paper, a data-driven method is proposed to achieve the same objective, which could acquire a more satisfactory result and provide an accurate prediction for two key performance indicator(i.e. throttle and brake). A vehicle cyber-physical system (VCPS) benchmark is built on the professional software CarSim. The algorithm just-in time learning (JITL) would process motivation data produced by the benchmark and compute out the prediction result. For testifying the advantages of the proposed method, the other two fitting algorithms (i.e. PLS and KPLS) are compared with it. The simulation results prove that JITL could consume much lesser time and receive a more precise prediction.