Abstract
Abstract An emotion-inspired artificial neural network (ANN) was developed to predict proton exchange membrane fuel cell (PEMFC) performance from operating conditions, addressing the need for accurate, low-overhead models deployable in real time (e.g. hydrogen-powered electric vehicles). Novelty lies in emotion-modulated updates coupled with diversity-preserving evolutionary tuning, enabling adaptive learning and improved generalization under coupled temperature–pressure–humidity–load variations. Using a physics-based dataset and K-fold validation with Monte Carlo sensitivity, the model outperformed ANN while reducing computation. Voltage and current-density errors were low (root mean squared error, 0.03 V, 0.15 A/cm2), with higher fit (R2 ≈ 0.97) and ~22% lower cost; robustness was maintained across perturbations (R2 > 0.95, 200 runs).