Joint prediction of the capacity and temperature of Li-ion batteries by using ConvLSTM Network

Abstract

Predicting the capacity and temperature of lithium-ion batteries is of critical significance to ensure their safety and stability, and consequently, extend the service life of battery systems. However, the degradation of capacity and thermal performance is typically regarded as independent processes, disregarding their coupling relationship. In response, this study constructs a combined model based on convolutional long short-term memory for the joint prediction of the reversible capacity and peak discharge temperature of batteries. The model’s feature extraction and pattern reconstruction capabilities are well-acknowledged. A variety of charging and discharging features (e.g., current, voltage, temperature, and incremental capacity) are analyzed and correlated with the evolution trends of battery capacity and temperature during long-term operation. Moreover, the evident phenomenon of capacity regeneration caused by intermittent rest is considered. Finally, the prediction results for different cells from public datasets show that the root-mean-square errors of capacity prediction vary from 0.01179 to 0.03304, and the mean absolute percentage error of peak discharge temperature prediction can be basically kept lower than 0.6%.