Reinforcement learning-based energy management for hybrid electric vehicle enhanced by velocity prediction

This paper proposes a novel approach for optimal energy management of engine-battery hybrid electric systems. It addresses two limitations in existing research, namely inaccurate speed prediction and low control accuracy. The proposed method integrates Markov theory and clustering theory to achieve high-precision prediction of future velocity under diverse driving styles. Furthermore, a learning-based energy management strategy is developed using a deep deterministic strategy gradient algorithm. This strategy considers the predicted velocity and aims to minimize engine fuel consumption, battery state fluctuations, and engine start-stop occurrences. Simulation results demonstrate that the proposed algorithm achieves near-optimal performance under specific operating conditions.