State of Health Estimation and Remaining Useful Life Prediction of The Lithium Battery for New Energy Vehicles with Long Short-Term Memory Neural Network

This paper introduces a model-based method to estimate the real-time State of Health (SoH) of the lithium battery of NEV (New Energy Vehicle) with machine learning algorithms upon the traditional ampere-hour integral method. The traditional methods for estimating the SoH (State of Health) of the lithium battery are ampere-hour integral, IC-curve, Big data, and Kalman filtering, but the problem of those methods is that it can only estimate the SoH in the past based on the historical battery data rather than the current SoH or the future life cycle. By combining machine learning algorithms and the ampere-hour method, we develop a way to estimate the real-time SoH, enabling the car manufacturer to understand better the current state of the lithium battery of NEV. Upon that, we also develop an algorithm to predict the future decay curve of SoH by using a deep neural network, the long short-term memory network, making the life cycle of the lithium battery more predictable. Our method hits 0.009 absolute mean error of real-time SoH prediction and 0.021 for future decay curve prediction from the real NEVs test by performing on the dataset based on actual real-time monitoring data provided by one OEM (Original Equipment Manufacturer).