Power Management Based on Reinforcement Learning Integrating SOC Constrain for Hybrid Electric Air and Land Vehicle

Hybrid electric power system (HEPS) with gas turbine (GT) is promising solution for hybrid electric land and air vehicle, and power management strategy (PMS) is key to obtain better performances of HEPS. Reinforcement learning (RL) based PMS needs a series of interaction with environment for training to obtain optimal PMS. However, improper interaction by mechanism of exploration and exploitation of RL agent can result in constrains violation of system state. Therefore, in this paper, a PMS based on Q learning integrating state of charge (SOC) constrains (QL-SOC) is proposed to avoid violating limit constrains of SOC of HEPS. Comparison results indicate that QL-SOC approach can ensure RL agent to complete training with no violating limit constrains, and its convergence speed is 3.5 times as much as that without QL-SOC approach. Simulation results based on air-land driving condition prove that proposed RL-based PMS can keep SOC stabilizing around preset value well.