Performance-oriented model learning and model predictive control for PEMFC air supply system

As an efficient, clean and pollution -free power generation device, proton exchange membrane fuel cell (PEMFC) has been widely applied in transportation, distributed power generation and other fields. However, the performance of PEMFC cannot be separated from appropriate cathode gas supply, for example, it is prone to oxygen starvation under frequent variable load conditions. This is due to the mismatch between the air supply rate and the electrochemical reaction rate, making it difficult for the system to meet the air flow requirement instantaneously when subject to the load changes abruptly. Thus, oxygen starvation control has become a particularly challenging nonlinear control problem because of the great difficulty in achieving an accurate system identification model and an efficient controller. To this end, a long short-term memory (LSTM) neural network -based model predictive control (MPC) is developed to model and control the PEMFC air supply system, which combines the advantages of LSTM and MPC. Firstly, LSTM is utilized to train an online control -oriented model from the measured dataset. Secondly, a sparrow search algorithm (SSA) is utilized to update the hyper -parameters of LSTM model, which can obtain more accurate predictive model in MPC. Thirdly, the MPC with LSTM-SSA model is solved online using different high efficiency solvers. Fourthly, the proposed LSTM-SSA based on MPC is adopted to model and control a PEMFC air supply system. Finally, the stability proof of the proposed method is illustrated in the Appendix. The simulation results reveal that the data -driven learning method and MPC method have significant advantages in modeling and improving the system performance.