Data-driven reconstruction of interpretable model for air supply system of proton exchange membrane fuel cell

Abstract Appropriate air supply system controller is of great significance to improve the performance of proton exchange membrane fuel cell. Most of controllers rely on the high-precision, simple, and interpretable model. It is particularly important to establish the model for the fuel cell air supply system. Since the high-precision physically interpretable control-oriented model can provide an understanding of the underlying phenomena apart from computational tractability for aerodynamic problems. Data-driven sparse identification based on auto-encoder method is proposed to establish the model. It can be divided into the four steps. Firstly, collect data from a simulation model and the actual fuel cell system, and auto-encoder network is used to discover a coordinate transformation into a reduced space. Secondly, dictionary library is constructed from candidate terms based on system analysis. Thirdly, air supply model reconstruction problem is transformed into a sparse identification problem. Finally, the developed model is verified by two datasets. Compared with other methods, the results show that mean absolute error and root mean squared error of the three variables for proposed method are the smallest under both simulation data and real data. And the reconstruction results perfectly agree with the original simulation and the real data. Especially, the proposed method can be easily extended to other system modeling studies, such as the hydrogen supply system model and thermal management system model of the fuel cell system.