Prediction of Battery Remaining Useful Life Based on Multi-dimensional Features and Machine Learning

With the high energy density, long cycle life and other advantages, lithium-ion batteries are widely used. However, working reliability has become one of their wide application restrictions. The key is to accurately predict the battery life and maintain or replace batteries in time. Considering the charging and discharging strategies of the real energy storage stations, based on the first 100 cycles of multi-stage quick charging data from the MIT-Stanford public datasets, various processing and measurement features are extracted; then, the coefficients of correlation between each feature and the battery life are compared; and next, four features more highly correlated with the battery life are selected and inputted into various machine learning models for training. With the proportion of training data and the model type changed, a prediction error of the optimal model and an average error in cell life prediction are gained separately, namely, 9.34% and 92.56 cycles.