DEEP LEARNING BASED HIERARCHICAL PREDICTIVE CONTROL FOR OXYGEN STOICHIOMETRY OF PROTON EXCHANGE MEMBRANE FUEL CELL ENGINE

A deep learning based hierarchical predictive control is developed for regulating the oxygen stoichiometry of proton exchange membrane fuel cell (PEMFC) engine in this study. Firstly, a hierarchical predictive control scheme is proposed by designing the first-level predictor to determine the operation current of PEMFC engine, and then the second-level model predictive control (MPC) generating robust control input. BP neural network is selected to formulate the first-level prediction model and airflow model is linearized to design MPC with suitable prediction horizon and control horizon. A simulation test is carried out through operating in a mixed driving cycle MANHATTAN + (a part of) UDDS to verify the efficacy of the proposed method. The results indicate that the oxygen stoichiometry tracks the reference value well avoiding the starvation of the PEMFC engine.