Deep Reinforcement Learning Energy Management Strategy of Hybrid Energy Storage System for Electric Vehicles Based on Adaptive Decomposition Level Wavelet Transform

The development of advanced energy management strategies (EMS) for hybrid energy storage system (HESS) has become an important approach for electric vehicle to reduce energy consumption and replacement costs of battery. Due to the large state–action space of HESS in the complex and dynamic environment, the computational efficiency of the deep reinforcement learning (DRL)‐based EMS is greatly reduced, and the peak and transient power of the battery cannot be effectively decreased. Therefore, a two‐layer EMS combining adaptive wavelet transform and DRL is proposed to achieve their complementary advantages, which can improve the computational efficiency of DRL while improving energy economy and extending battery lifespan. In the upper layer, an adaptive wavelet transform combining driving condition recognition and driving style recognition is designed to preprocess the load signal to separate high frequency and peak power, which can effectively reduce the impact of surge current on the battery. Subsequently, DRL takes the power that reduced action–state space as the input to make optimal decisions for redistribution in the lower layer, which can improve DRL computational efficiency while enhancing energy economy and extending battery lifespan. Finally, simulation results demonstrate the effectiveness of the proposed EMS compared to dynamic programming and existing DRL strategies under various driving cycles.