Hybrid Electric Vehicle Powertrain Control Based on Reinforcement Learning

Hybrid Electric Vehicles (HEVs) achieve better fuel economy than conventional vehicles by employing two different power sources: a mechanical engine and an electrical motor. These power sources have conventionally been controlled by a rule-based algorithm or optimization-based control. Besides these conventional approaches, reinforcement learning-based control algorithms have actively been studied recently. To investigate the benefits of the reinforcement learning-based approach, a model-free control algorithm for an HEV is proposed in this article using a Twin Delayed Deep Deterministic policy gradient (TD3), which is an online, off-policy Deep Reinforcement Learning (DRL) algorithm. The effectiveness of the proposed algorithm is studied by applying the TD3 algorithm to a 48V mild HEV (MHEV) model and the optimal operating strategy is obtained for maximum fuel economy. The simulation results show that the proposed TD3-based algorithm improves the average fuel economy by 1.89% on standard driving cycles and 2.20% on real-world driving cycles when compared to the Deep Deterministic Policy Gradient (DDPG) algorithm.