Deep Reinforcement Learning DDPG Algorithm with AM based Transferable EMS for FCHEVs

Abstract

Hydrogen fuel cell is used to run fuel cell hybrid electrical vehicles (FCHEVs). These FCHEVs are more efficient than vehicles based on conventional internal combustion engines due to no tailpipe emissions. FCHEVs emit water vapor and warm air. FCHEVs are demanding fast dynamic responses during acceleration and braking. To balance dynamic responsiveness, develop hybrid electric cars with fuel cell (FC) and auxiliary energy storage source batteries. This research paper discusses the development of an energy management strategy (EMS) for power-split FC-based hybrid electric cars using an algorithm called deep deterministic policy gradient (DDPG) which is based on deep reinforcement learning (DRL). DRL-based energy management techniques lack constraint capacity, learning speed, and convergence stability. To address these limitations proposes an action masking (AM) technique to stop the DDPG-based approach from producing incorrect actions that go against the system's physical limits and prevent them from being generated. In addition, the transfer learning (TL) approach of the DDPG-based strategy was investigated in order to circumvent the need for repetitive neural network training throughout the various driving cycles. The findings demonstrated that the suggested DDPG-based approach in conjunction with the AM method and TL method overcomes the limitations of current DRL-based approaches, providing an effective energy management system for power-split FCHEVs with reduced agent training time.