Fuzzy Rule Value Reinforcement Learning based Energy Management Strategy for Fuel Cell Hybrid Electric Vehicles

Abstract

A novel Fuzzy rule value reinforcement learning based energy management strategy is proposed for the fuel cell hybrid electric vehicle. The optimization target of the proposed method is to reduce hydrogen consumption and maintain the continuous operation of the battery. Fuzzy rule value reinforcement learning uses a fuzzy inference system to approximate the state-action value function, which enables the possible implementation of continuous state and/or action space. It does not rely on data or experience to set rules, but learns and adjusts rules by itself through interacting with the environment. Therefore, the proposed method can adapt to the changes in models or operating conditions, such as fuel cell degradation and changes in driving conditions. Simulation tests verify the effectiveness of the proposed method in solving energy management problems. Faster and smoother convergence and powerful adaptability to the environment changes are also verified.