Neural Network Driven by Electrochemical Performance Data for Predicting the Discharge Termination Time of Seawater Electrolyte-Based Metal-Air Batteries

Abstract

Seawater electrolyte-based metal-air batteries exhibit great promise for marine energy supply systems. However, conventional statistical analysis methods, though applicable to seawater metal-air battery lifetime prediction, have inherent limitations of insufficient prediction accuracy and large error. Herein, a deep time-series regression framework based on InceptionTime and incorporating prior-biased attention pooling is proposed to construct a nonlinear mapping between electrochemical performance sequences and the discharge termination time of catalysts. Specifically, chronoamperometric profiles are employed to extract long-term stability features, while prior knowledge derived from linear sweep voltammetry is introduced to strengthen the attention weighting over critical potential regions. Under a nested leave-one-catalyst-out cross-validation framework with the one-standard-error rule for epoch selection, the model demonstrates high consistency across various aggregation strategies on a small-sample test set containing different air cathode catalysts, yielding a coefficient of determination over 0.90. These findings suggest that integrating multiscale temporal features with electrochemical prior knowledge through an attention-driven regression framework can improve the prediction accuracy and reduce the prediction error of discharge termination time in the current dataset, thereby providing a preliminary data-driven approach for catalyst evaluation in seawater metal-air batteries.