Informativeness-aware moving horizon estimation for temperature-only monitoring of an open-cathode PEM fuel cell

This work addresses temperature-only monitoring of an open-cathode proton exchange membrane fuel cell (PEMFC) under weakly informative operating conditions. A reduced-order thermo–hydration model is embedded in a constrained single-shooting moving horizon estimator to reconstruct a liquid-water saturation proxy while adapting dominant lumped parameters. To prevent parameter drift during poorly informative windows, the estimator combines bounded disturbance regularization with an informativeness-aware admission mechanism that merges discounted sensitivity-based indicators with PEMFC-specific physical visibility checks. Unsupported parameter directions are frozen through an accept–freeze update. The framework is evaluated through simulations under low- and high-excitation conditions, including outlier-corrupted temperature measurements, and through experiments on an open-cathode PEMFC test bench. Relative to an extended Kalman filter under identical sensing assumptions, the proposed method yields more stable unmeasured-state and parameter trajectories while preserving accurate temperature reconstruction. Under outlier-contaminated low-excitation operation, the method preserves unmeasured-state consistency; a separate noise-free parameter-drift test further demonstrates responsiveness to a degradation-like slow parameter drift.