A novel SOC estimation method for lithium-ion batteries using the fusion of deep neural network and physical information model

Abstract

Neural network algorithms estimate the state of charge (SOC) of lithium-ion batteries by learning the mapping relationship between features and labels. However, these methods do not consider the influence of battery characteristics on SOC estimation, which makes the SOC estimation methods developed based on neural networks lack interpretability and suffer from output fluctuations. For these issues, this study introduces battery physical information to the neural network and proposes a SOC estimation algorithm based on the fusion of deep neural network and equivalent circuit model (ECM). Firstly, a neural network model integrating convolutional neural network (CNN) with bidirectional long short-term memory (BiLSTM), named CNN-BiLSTM, is designed. CNN-BiLSTM is able to extract important local features while capturing global dependencies between data, thus improving SOC estimation accuracy. Subsequently, a weight calculation algorithm based on terminal voltage residuals is proposed to fuse CNN-BiLSTM with ECM, thus obtaining a fusion model capable of expressing physical information and learning complex nonlinear variations. Finally, the relationship between fusion model and filter observation equation is leveraged to achieve SOC closed-loop estimation through the double extended Kalman filter (DEKF). The proposed SOC estimation algorithm is experimentally evaluated in a variety of scenarios. The results show that the proposed algorithm significantly improves the SOC estimation performance of baseline methods, with the error consistently maintained below 1.5 %. Meanwhile the method can effectively smooth the output fluctuation of neural network.