BiLSTM-ATTENTION-CNN predictive energy management based on ISSA optimization for intelligent fuel cell hybrid vehicle platoon

Predictive energy management strategies (PEMS) offer significant potential in enhancing the driving economy of fuel cell hybrid electric vehicle (FCHEV) platoons through the integration of speed prediction and advanced energy management strategies (EMS). This study develops and applies a speed prediction model and an energy management strategy to platoon driving scenarios using deep reinforcement learning, and designs a variable headway platoon following control strategy based on online real-time adjustment through Model Predictive Control (MPC) to determine the desired acceleration of the platoon in real time. For speed prediction, this study augments the traditional Bidirectional Long Short-Term Memory (BiLSTM) network with an attention (ATT) mechanism to filter out irrelevant traffic information and optimize network parameters using an improved sparrow search algorithm (ISSA) for heightened prediction accuracy. For energy management, this study proposes a Twin Delayed Deep Deterministic Policy Gradient (TD3)-based energy management strategy that incorporates the energy life decay factor, showcasing robust training performance. Simulation results demonstrate that the platoon PEMS exhibits exceptional adaptability across various operating conditions, achieving a 67.32% reduction in Root Mean Square Error (RMSE) compared to traditional LSTM networks and a 22.99% decrease compared to single vehicle prediction methods. Furthermore, it maintains a stable state of charge (SOC) level and minimizes overall costs.