A Machine Learning Approach forecasting Li-ion Battery State-of-Health and Discharge Behavior

Lithium-ion batteries power countless modern technologies, making accurate remaining useful life (RUL) prediction essential for safe and reliable operations. This study proposes a novel machine learning approach for Li-ion battery health monitoring using an accelerated life testing dataset from NASA Ames and UCF. Two complementary methodologies will be developed in this paper: (1) a state-of-health (SOH) forecasting model that predicts battery degradation from a single discharge cycle using a modified exponential decay function, and (2) a discharge curve forecasting framework that accurately forecasts voltage as a function of state-of-charge (SOC) and time using deep learning architectures. Unlike conventional approaches that require multiple cycles for reliable predictions, the developed method enables RUL forecasting using data from only one discharge cycle. Furthermore, the proposed approach operates with any discharge current profile, eliminating the need for the standardized $1 C$ rating typically required in battery testing protocols. The SOH forecasting is accomplished through a computationally lightweight Stacking Regressor ensemble that delivers strong explanatory power, while discharge curve prediction relies on a Hybrid (CNN+LSTM) architecture that effectively captures temporal dependencies in battery voltage dynamics during discharge cycles.