PCFE‐Net: An Advanced Interpretable Model Fusing Feature Enhancement and Physical Monotonic Constraints for Predicting Corrosion Rates of Galvanized Steel in Extreme Typhoon and Non‐Typhoon Weather Conditions

ABSTRACT

Typhoons, as frequent extreme convective weather events, significantly affect regional environmental conditions and thus influence the atmospheric corrosion behavior of metallic materials. Yet, few studies examine corrosion mechanisms under typhoon conditions. This study focuses on the typhoon‐prone southeastern coast of China, using minute‐level Atmospheric Corrosion Monitor (ACM) sensor data to build a unified prediction framework for galvanized steel corrosion rates under both typhoon and normal conditions. We introduce five typhoon‐related variables: typhoon translation speed (Moving_ty), maximum wind speed near the typhoon center (Wind_ty), typhoon occurrence (Occur_ty), relative bearing between the typhoon direction and the corrosion site (Bearing_ty), and distance between the typhoon center and corrosion location (Distance_ty). We developed PCFE‐Net, a neural network combining feature enhancement and physical monotonic constraints, which significantly outperformed traditional machine learning methods. Predicted corrosion maps matched observed spatial patterns, and interpretability analyses confirmed the model aligns with established physical principles.