Prognostics of lithium-ion batteries health state: An ensemble method of one-dimensional convolutional neural network and Informer model

To meet the requirements for accurate and stable in lithium-ion battery state of health (SOH) prediction for electric vehicles, this paper proposes a lithium-ion battery SOH prediction method based on health features and ensemble models, which is named 1DCNN-Informer. Firstly, six sets of health indicators (HIs) with high correlation coefficients are extracted from the cyclic aging cycle curves of lithium-ion batteries using the traversal method. Secondly, the dependence of the target capacity on the six characteristic time sequence is established based on the powerful feature extraction capability of the Informer model, and the SOH prediction result is further obtained. Thirdly, a one-dimensional convolutional neural network (1DCNN) model is used to extract the temporal features of the decoder’s prior sequence; thereby the SOH prediction result at the current time step can be obtained. The final SOH prediction is determined by an adaptive weight fusion mechanism. The experimental validation conducted on NASA and CALCE datasets show that the prediction performance of lithium-ion battery SOH is superior than other mainstream methods.