Power Management Strategy of Hybrid Vehicles Using Sarsa Method

An exclusive trend toward smart grids have been started and hybrid electric vehicles development as one of the necessities for smart grids in many countries throughout the world has been put on the agenda. Hybrid electric vehicles consist of two or more energy sources. Fuel cell/battery hybrid electric vehicles are among suitable substitutions for today's gasoline vehicles. In this paper, based on SARSA algorithm, which is one of the reinforcement learning methods, a power management strategy is represented. The hybrid electric vehicle controller is considered as a learning agent in reinforcement learning method and through its interaction with the environment, by trial and error, learns to propose a power management strategy between fuel cell and the battery, such that the best possible return is obtained in a long time. The best possible return in this problem is battery state of charge sustenance while achieving the optimal fuel consumption. Finally, the results of both Q-learning and SARSA algorithms are compared.