Hierarchical energy management for FCHEV in car-following scenarios with speed prediction

The integration of connected and autonomous vehicle (CAV) technologies with advanced deep reinforcement learning (DRL) techniques has opened new frontiers in vehicular energy efficiency optimization. This study presents a novel hierarchical energy management strategy (EMS) framework designed to address the dual challenges of driving performance enhancement and fuel economy optimization for fuel cell hybrid electric vehicles (FCHEV) operating in car-following scenarios. The proposed framework comprises two sequential components. Initially, a hybrid CNN-BiLSTM-Attention neural network is developed to generate precise velocity predictions for the preceding vehicle. These prediction results are then utilized to establish a preliminary predictive cruise control (PCC) strategy for maintaining appropriate following speeds. Subsequently, the twin delayed deep deterministic policy gradient (TD3) algorithm is implemented to optimize the power distribution among the FCHEV’s multiple energy sources, utilizing the planned speed trajectory as input. Extensive experimental evaluations confirm the system’s effectiveness in maintaining battery state-of-charge (SOC) within stable ranges while simultaneously achieving superior fuel efficiency and demonstrating robust performance across diverse driving cycles.