Predicting the Remaining Life of Lithium-ion Batteries Using a CNN-LSTM Model
Alireza Rastegarpanah, Yuan Wang, Rustam Stolkin
Abstract
Accurate predicting the remaining useful life of lithium-ion batteries is essential for the market of Electrical Vehicles (EVs) and the battery industry. However, diverse ageing processes, substantial battery variability, and dynamic operating circumstances are identified as main challenges for predicting the remaining useful life (RUL) of lithium-ion batteries (LIBs). This study proposes a machine learning solution for estimating the RUL of LIBs by using a Convolutional neural network (CNN) model with an extra Long Short-term memory (LSTM) layer. The developed CNN-LSTM model is trained by a dataset containing data extracted from 124 commercial lithium-ion batteries cycled under fast-charging conditions. In this study, we use only 100 cycles to predict the remaining cycles. The developed model achieved a competitive loss value of 0.0206 and the mean absolute error value was 0.1099 for the current cycle of the battery and 0.0741 for the remaining ones.