Sensitivity analysis of a CNN–LSTM prognostic framework for proton exchange membrane fuel cells: Effects of sliding window size and training data allocation

• The hybrid CNN–LSTM reduces short-term voltage prediction RMSE by 49% compared to a standalone LSTM baseline. • A sliding window of 40 hours is optimal for recursive long-horizon lifetime forecasting, keeping RUL error within ±5 hours. • Model robustness is strong: with >500 hours of training data, RMSE variation remains below 0.001 V. • The methodological framework provides practical, data-driven guidance for configuring CNN–LSTM PEMFC prognostics and can extend to dynamic-load scenarios.