Lithium-ion Battery Health Estimation Using DCNN Paralleled LSTM-Self Attention Networks

State of Health (SOH) is one of key indicators to characterize the aging degree of batteries. This paper proposed a SOH estimation model based on Deep Convolution Neural Network paralleled with LSTM and Multi-Self Attention Network (DCNN-LSTM-MSA Network, DLA-Net) combining battery charging partial curve. We defined two charging segments with different lengths and extracted relevant features, i.e., the time corresponding to the fixed voltage interval and its product with the voltage. The final input model features were selected through Pearson correlation coefficient testing. The results shows that the suggested model exhibits superior accuracy and improved estimation performance compared to conventional models.