Rapid and flexible lithium-ion battery performance evaluation using random charging curve based on deep learning

Abstract

Accurate performance evaluation of lithium-ion battery is crucial for its detection, screening and echelon utilization. However, existing evaluation methods rely on specific or complex tests, leading to limited flexibility and high time costs. To address these challenges, a novel two-stage approach based on random charging curve is proposed to evaluate battery performance rapidly and flexibly. Firstly, a deep residual convolutional neural network (Res-CNN) is developed, requiring only a small portion of the charging curve as input to preliminarily estimate battery performance and screen. Secondly, multi-parameters are extracted from the batteries of the same level, and the comprehensive performance is evaluated based on an improved grey TOPSIS algorithm. Finally, two hundred and thirty LiFePO4 batteries are used to verify the effectiveness of this method. The results show that 94.55% accuracy at any initial voltage is achieved to quickly classify batteries into four levels according to battery performance. Furthermore, the consistency of the regrouped batteries has been greatly improved after comprehensive evaluation by comparative test. Obviously, this research has important guiding significance for residual value evaluation and rapid reorganization of battery echelon utilization in different scenarios.