Energy management strategy for hybrid electric vehicle integrated with waste heat recovery system based on deep reinforcement learning

Hybrid electric vehicles (HEVs) are acknowledged to be an effective way to improve the efficiency of internal combustion engines (ICEs) and reduce fuel consumption. Although the ICE in an HEV can maintain high efficiency during driving, its thermal efficiency is approximately 40%, and the rest of the fuel energy is discharged through different kinds of waste heat. Therefore, it is important to recover the engine waste heat. Because of the great waste heat recovery performance of the organic Rankine cycle (ORC), an HEV integrated with an ORC (HEV-ORC) has been proposed. However, the addition of ORC creates a stiff and multi-energy problem, greatly increasing the complexity of the energy management system (EMS). Considering the great potential of deep reinforcement learning (DRL) for solving complex control problems, this work proposes a DRL-based EMS for an HEV-ORC. The simulation results demonstrate that the DRL-based EMS can save 2% more fuel energy than the rule-based EMS because the former provides higher average efficiencies for both engine and motor, as well as more stable ORC power and battery state. Furthermore, the battery always has sufficient capacity to store the ORC power. Consequently, DRL showed great potential for solving complex energy management problems.