CTBANet: A new method for state of health estimation of lithium-ion batteries

The key index to characterize the lifespan of lithium-ion batteries is the state of health (SOH), and accurate SOH estimation is fundamental for the secure operation of batteries. Based on modal decomposition and deep learning, a novel SOH estimation method named CTBANet is proposed in this study, which is composed of a CEEMDAN (Complete Ensemble mode decomposition with Adaptive Noise) module, TCN (Temporary Convolutional Network), BiLSTM (Bi-directional Long Short-Term Memory) and Attention mechanism. During data preprocessing, the battery capacity is decomposed into various Intrinsic Mode Functions (IMFs) with diverse frequencies by the CEEMDAN algorithm. This can obtain the local regeneration characteristics of battery aging. The components with high correlation are selected by Pearson correlation analysis to reduce the calculation costs. Then, the deep learning part of the proposed method is used to estimate SOH. Among them, TCN uses causal dilated convolution to extract hidden information among variables in the feature matrix, which improves the ability of the model to extract time data features. Then BiLSTM combines bidirectional processing with longsequence modeling, which makes the model effectively predict the state of the next moment under the given long-time series. In addition, the Attention module emphasizes the important features by assigning weights to the BiLSTM output sequence to improve estimation accuracy. The effectiveness of the CTBANet method is verified by setting up multiple groups of experiments on the NASA dataset and the Oxford dataset. And MAE, MSE, and RMSE of each group of experimental results are within 0.6 %, 0.005 %, and 0.7 % respectively, which shows that the CTBANet method can precisely estimate the SOH of lithium-ion batteries.