Volume 11: Sustainable Energy Solutions for Changing the World: Part III

Remaining useful life prediction of lithium-ion batteries with limited historical data Chang Wnag, Weiling Luan

Abstract

The lithium-ion battery system has strong coupling and nonlinear characteristics, bringing great challenges to its online failure prediction and life estimation. Several studies have shown the potential of deep learning methods on remaining useful life (RUL) prediction of lithium-ion batteries, these methods mostly use historical cycling data from beginning to the prediction point. However, long-term cycling data can be costly and difficult to obtain in practical applications. This article presents a data-driven algorithm using a combination of deep convolutional neural network (DCNN) and long short-term memory (LSTM) to predict the RUL of lithium batteries based on the data of the past 10 continuous cycles. Here the DCNN processes time-series data including capacity, temperature, and capacity difference at the same voltage during discharging, the LSTM is used to process the scalar data of each cycle including internal resistance, discharge time, and discharge capacity. The proposed network uses an open dataset with 124 batteries for training and validation. The generalization ability of the model for batteries under different charging/discharging strategies is also validated. The proposed DCNN-LSTM network demonstrates well performance on capturing the life changes of batteries with different life lengths and working conditions using limited historical cycling data with an average root mean square error of 5%.

Keywords remaining useful life prediction, lithium-ion battery, convolutional neural network, long short-term memory, deep learning