Radial Basis Function Neural Network‐Based Observer Design for Fuel Cell Liquid Water Saturation

ABSTRACT

Water management remains a critical challenge in proton exchange membrane fuel cells (PEMFCs) owing to the need to balance flooding and membrane dry‐out. However, the lack of direct measurements of liquid water saturation in the catalyst layer hinders effective water‐state monitoring and control. To address this issue, a nonlinear dynamic observer is proposed for the online estimation of liquid water saturation based on a lumped‐parameter state‐space model with a triangular structure. A radial basis function neural network (RBFNN) is incorporated to approximate nonlinear terms associated with unmeasurable current derivatives and compensate for modeling uncertainties. Observer gain matrices and adaptive RBFNN weight update laws are systematically derived to guarantee estimation performance and stability. Furthermore, a computationally tractable observer design framework is established through the integration of linear matrix inequalities (LMIs) and algebraic constraint parameterization. Comparative simulations show that the proposed observer improves the estimation accuracy of stack temperature and liquid water saturation by 67.7 and 46, respectively, compared with a conventional high‐gain observer. These results demonstrate the effectiveness and potential of the proposed approach for real‐time PEMFC water‐state monitoring.