Deep reinforcement learning-based hierarchical control strategy for energy management of intelligent fuel cell hybrid electric vehicles

Speed optimization and energy management strategies for intelligent fuel cell hybrid electric vehicles (IFCHEVs) can significantly enhance energy utilization efficacy in dynamic driving environments. This study proposes a hierarchical cooperative optimization strategy for IFCHEVs operating in environments with dynamically varying preceding vehicles. The upper-layer speed optimization integrates a hybrid Particle Swarm Optimization-Gaussian Process Regression model algorithm (PSO-GP) to predict the speed of the preceding vehicle. This algorithm synergizes the global search capability of PSO with the probabilistic modeling advantages of GP. Then, a Model Predictive Control-based Adaptive Cruise Control (MPC-ACC) framework is used to dynamically optimize the host vehicle speed using real-time preceding vehicle state information while ensuring a safe distance. At the lower layer, a Twin Delayed Deep Deterministic Policy Gradient (TD3) algorithm allocates energy between the fuel cell and battery, explicitly minimizing hydrogen consumption and mitigating power source degradation. Simulation results demonstrate that the proposed strategy synergistically incorporates preceding vehicle speed predictions and real-time road conditions into IFCHEV speed optimization, achieving a 4.06 % reduction in hydrogen consumption and a 3.50 % decrease in global cost while enhancing traffic flow stability.