Anomaly Detection and Extra Tree Regression for Assessment of the Remaining Useful Life of Lithium-Ion Battery

Abstract

The knowledge of the Remaining Useful Life (RUL) of Lithium-ion (Li+) battery is significant in battery management and helps in designing numerous fit-for-purpose systems. We implemented a segmentation-type anomaly detection to establish the changing characteristics of Li+ battery by using the measured voltage and temperature at different timesteps. Hence extracting useful changepoint features of the voltage and temperature transitions such as mean, variance, skewness, kurtosis and voltage for predicting the RUL of the battery with Extra Tree Regression (ETR) algorithm. The model was predicted to an accuracy of 96.25%–97.86% when the Mean Absolute Percentage Error (MAPE) of three Li+ batteries were tested. The robustness of this study makes it a very useful technique for Li+ battery prognosis, design and uncertainty estimation of the performance of critical systems depending on the Li+ battery power.