ELM-based driver torque demand prediction and real-time optimal energy management strategy for HEVs

Abstract

In hybrid electric vehicles, the energy economy depends on the coordination between the internal combustion engine and the electric machines under the constraint that the total propulsion power satisfies the driver demand power. To optimize this coordination, not only the current power demand but also the future one is needed for real-time distribution decision. This paper presents a prediction-based optimal energy management strategy. Extreme learning machine algorithm is exploited to provide the driver torque demand prediction for realizing the receding horizon optimization. With an industrial used traffic-in-the-loop powertrain simulation platform, an urban driving route scenario is built for the source data collection. Both of one-step-ahead and multi-step-ahead predictions are investigated. The prediction results show that for the three-step-ahead prediction, the 1st step can achieve unbiased estimation and the minimum root-mean-square error can achieve 100, 150 and 160 of the 1st, 2nd and 3rd steps, respectively. Furthermore, integrating with the learning-based prediction, a real-time energy management strategy is designed by solving the receding horizon optimization problem. Simulation results demonstrate the effect of the proposed scheme.