A Hybrid Drive Method for Capacity Prediction of Lithium-Ion Batteries

As one of the most attractive energy storage devices, capacity prediction of lithium-ion batteries is significant to improve the safe availability of new energy electronic devices. At present, methods based on neural network are widely used in battery capacity prediction. However, due to instability and incompleteness of the learning ability of a single neural network and limitations of health features (HFs), the stability and accuracy of capacity estimation results are directly affected. Therefore, a hybrid driven battery capacity prediction model is proposed in this article, which fully considers the local timing information and global degradation information during capacity degradation process. First, electrochemical impedance spectroscopy (EIS) in the complex frequency domain is combined with the characteristics extracted from the incremental capacity (IC) curve in the time domain to form multi-dimensional HFs. Then, Elman neural network (ENN) and support vector regression (SVR) are used to learn the local timing information and global degradation trend of capacity decay process, respectively. Finally, the information learned from the two parts is fused by the extreme learning machine (ELM) for weight allocation, so as to predict the battery capacity quickly and accurately. The experimental results show that the new method can estimate the capacity of lithium-ion batteries more accurately on different datasets.

Automated summary

In this article, a hybrid driven battery capacity prediction model is proposed, which considers both local timing information and global degradation trend by combining different domain features and using Elman neural network and support vector regression for learning. The results are then fused using extreme learning machine. The new method achieves fast and accurate prediction of lithium-ion battery capacity.