AI-Driven Prediction of Spacecraft Battery Degradation Using Telemetry-Based Discharge-Curve Indicators

Spacecraft battery degradation critically constrains mission lifetime, yet in-orbit capacity measurements are unavailable and the operational environment introduces strong periodic stress patterns. This work proposes a feature-engineering framework to forecast key battery observables (charging/discharging currents and voltage) and predict the State of Health (SOH) from voltage-derived degradation. First, a sequence-to-sequence Multilayer Perceptron (MLP) is trained to predict future telemetry segments, enabling forward propagation of battery observables. Then, SOH is inferred using discharge-curve indicators derived from voltage, including a cumulative, history-aware extension of the Time Series Equal Discharging Voltage Difference indicator, referred to as CTSEDVD, to better capture progressive aging effects under partial-discharge operating conditions. The SOH model is developed and validated on laboratory datasets (NASA PCoE and XJTU) and subsequently applied to the GOCE mission, comparing SOH obtained from ground-truth voltage against SOH derived from forecasted voltage. Results show that the CTSEDVD-based pipeline markedly improves consistency between the two SOH trajectories, supporting the feasibility of in-orbit degradation monitoring and prediction using forecasted telemetry within a computationally efficient workflow.