Predictive control of PEMFC based on a combined empirical and mechanistic model

The modelling and control of proton exchange membrane fuel cell (PEMFC) possesses great challenges due to PEMFC system's inherent nonlinearities, time-varying characteristics and tight operating constraints. In this paper, we propose a constrained model predictive control (MPC) strategy based on a combined empirical and mechanistic model of PEMFC. First, we propose a hybrid modelling approach based on the combination of prior knowledge, under the form of mechanistic submodel, with empirical submodel devoted to the extraction of knowledge from operating data. The empirical submodel is a SVM model, which predicts the voltage at different stack currents and temperatures under the reference hydrogen and oxygen partial pressure. The mechanistic submodel calculates the correction voltage by taking account of hydrogen and oxygen partial pressure changes. Particle swarm optimization (PSO) algorithm and penalty function are then employed to solve the resulting nonlinear constrained predictive control problem. Simulation results demonstrate that the proposed method can deal with the constraints and achieve satisfactory performance.