Voltage Prediction and Fault Detection by using Model Fusion Strategy for Lithium-ion Batteries

Yijing Zhou, Mengdan Wu, Qi Yao, Zhiyu Duan, Shunkun Yang

Abstract: Detecting voltage faults in lithium-ion batteries is crucial in reducing potential property damage and injury risks. Current research focuses more on using measured values for voltage prediction, but many valuable operating condition information is ignored due to its difficulty in measurement. Therefore, a novel method using model fusion for voltage prediction and fault detection has been proposed in this paper, which especially focus on extracting implicit operating condition information for real-time prediction. It includes three modules: prediction module, similarity module, and fusion module, with CNN and LSTM as the basic unit. The incremental training scheme is adopted in the fusion module to ensure real-time performance. At the same time, the detrending optimization before incremental training has improved prediction accuracy. The predicted results can be used for threshold-based fault detection and differential-based fault detection methods. The experimental results demonstrate the satisfactory performance of the model in terms of predict accuracy, training efficiency and fault detection.

External IDs:dblp:conf/qrs/ZhouWYDY23