In order to reduce the impact of capacity recovery and noise on the remaining useful life(RUL)prediction of Li-ion battery,a joint prediction method of RUL based on the multi-scale temporal convolutional network(TCN)is proposed,which uses the timing information of operation data and capacity data.The original capacity data of the Li-ion battery are decomposed using variational mode decomposition(VMD)method,separating the nonlinear features and the main degradation trend into high-frequency and low-frequency components,respectively.The high-frequency components are predicted using a rolling iterative approach with a multi-scale TCN to capture short-term changes in capacity.Features extracted from the operation data are input into the multi-scale TCN to predict the low-frequency components,capturing the long-term trend of the capacity.The prediction results are restored to capacity prediction values.The National Aeronautics and Space Administration(NASA)dataset-based results indicate that the minimum root mean square error(RMSE)of the capacity prediction error is 0.011 1,the corresponding minimum mean absolute error(MAE)is 0.008 6,with the RUL prediction error mostly within two cycles by using this method.
维普
万方数据