Evaluation of Cell Inconsistency in Lithium-Ion Battery Pack Using the Autoencoder Network Model
Shyr-Long Jeng, Wei-Hua Chieng
Abstract
Cell inconsistency is a common problem in the charging and discharging of lithium-ion battery (LIB) packs that degrades the battery life. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">In situ</i> , real-time data can be obtained from the battery energy storage system (BESS) of an electric boat through telemetry. This article examined the use of a 57-kWh BESS comprising six battery packs connected in series, each of which contained 16 LIB cells with a nominal capacity of 180 Ah. Because of cell inconsistency, the 96 cells had different voltages during the charging process. We compared the performance of four types of autoencoders (AEs): a fully connected (FC) model, convolutional neural network (CNN) model, long short-term memory (LSTM) model, and hybrid CNN–LSTM model. These AEs were employed to evaluate the cell inconsistency by minimizing the reconstruction error. The FC model exhibited the optimal performance among the four AEs in an abnormal battery condition. The LSTM model had the highest capability of distinguishing normal and abnormal cells. Nevertheless, the CNN–LSTM model, which combines the advantages of the CNN and LSTM models, is the most effective AE for the complex working conditions of electric vehicles.