Water Management in Proton Exchange Membrane Fuel Cell Based on Actor Critic Learning Control

Abstract

Due to the complexity of modeling water management system and difficulties of online measuring humidity inside the PEM fuel cell stack, the actor critic learning controller is proposed by using the available measurements which are stack voltage and the difference of stack voltage between current sample time and last sample time. In this method approximation of value function is based on least squares temporal-difference, and approximations of actor model and process model are based on local linear regression. Simulation results show that actor critic learning control can maintain water balance inside the fuel cell stack and achieve the maximum the stack voltage under the different operating conditions.