Abstract

Forecasting the state of health and remaining useful life of Li-ion batteries is yet an unsolved challenge that limits technologies such as consumer electronics and electric vehicles. We build an accurate battery forecasting system by combining electrochemical impedance spectroscopy – a real-time, non-invasive and information-rich measurement that is hitherto underused in battery diagnosis – with Gaussian process machine learning. We collected over 20,000 EIS spectra of commercial Li-ion batteries at different states of health (SoH), states of charge (SoC) and temperatures – the largest dataset to our knowledge of its kind. Our Gaussian process model takes the entire spectrum as input, without manual feature engineering, and automatically determines which spectral features predict degradation. Our model significantly outperforms the state of the art, accurately predicting remaining useful life (RUL) even when the past operating conditions of the battery are unknown to the user. The model can be interpreted to shed light on the physical mechanisms of battery degradation. Our results are uniquely able to design the next-generation intelligent battery management systems which would enable a considerably safer operation of Li-ion batteries.