Abstract: Proton exchange membrane fuel cell (PEMFC) is an important modern sustainable energy power generation device, and accurate estimation of its performance degradation is crucial for practical applications. Conventional data-driven approaches lack consideration of aging mechanisms, leading to unsatisfying handling of voltage recovery during performance degradation. By analyzing various characteristic information in PEMFC operational data, this paper proposes a dynamic operational condition-based prediction method for PEMFC performance degradation. First, the random forest algorithm is employed for data feature analysis to determine the features used for model training. Subsequently, a bidirectional gated recurrent units (BiGRU) model is built, and parameter optimization is performed using the whale optimization algorithm (WOA). Finally, the optimized BiGRU model is utilized for predicting PEMFC performance degradation, and the prediction effectiveness is evaluated. Experimental results on the PEMFC dynamic data set show the prediction accuracy is improved by 28.1%～33.7% compared with that of GRU and increased by 25.5%～31.2% compared with that of BiGRU.