Accurate prognosis of the state of health(SOH) and remaining useful life(RUL) of lithium-ion battery is the key to ensure the safe use of lithium-ion battery. The traditional health feature extraction method is not suitable for the incomplete charge phenomenon in the actual use of the battery, which has weak anti-interference and low reliability. In this study, a light-weight automatic feature extractor based on temporal convolutional network is proposed for SOH online monitoring and RUL prediction. The depthwise convolution technique is used to enhance feature capture of original aging trend data of different channels. With the latest redundant feature processing technique in Ghost module, more " ghost " feature maps that can extract the required information from original features are generated through cheap convolution operation, which reduces about 40% of the calculation of the automatic feature extraction model in this study. In addition, The dynamic time warming barycenter average (DBA) algorithm is used to compress the redundancy in the original data in advance, focusing on providing new ideas for subsequent improvement points. Compared with the conventional baseline model on the accepted NASA data set, it is found that the accuracy of the proposed model RMSE is controlled within 2.5%, which has a higher prediction accuracy.
展开▼
