SOC Estimation of Li-ion Battery using on Variational Mode Decomposition and Transformer-Generative Adversarial Network

Abstract

Accurate estimation of the state of charge (SOC) of lithium battery is crucial to improve the dynamic performance and energy utilization of batteries. The method, the existing neural network are used to estimate SOC, has the problems of low accuracy and poor stability under complex working conditions. A new algorithm are proposed to estimate the SOC, which combines Transformer and Generative Adversarial Network (GAN), and the Variational Modal Decomposition (VMD). Firstly, as the excellent prediction ability of Transformer, Transformer is used as the generative network of GAN. Secondly, VMD is used to decompose the SOC historical data into six subsets to increase the input features. Finally, DST work data from the University of Maryland CALCE dataset is used for model training, and the VMD-Transformer-GAN algorithm is compared with LSTM, GRU, and BiLSTM algorithms for experiments. The experimental results show that the VMD-Transformer-GAN algorithm algorithmic estimation model has high stability and accuracy, which verifies the feasibility of the improved scheme.