A Lithium-ion Battery Capacity Estimation Method Based on Multi-Source Weighted Domain Adaptation

Abstract

This paper presents a multi-source weighted domain adaptive lithium-ion battery capacity estimation method, including: acquiring capacity degradation data of the target domain and multiple source domains; extracting common features using adversarial networks; calculating the multi-order statistical moment total loss function between the source domain and the target domain based on the capacity degradation data of the target domain and multiple source domains; determining the weight values ​​corresponding to different source domains based on the distribution distance between the common features of the source domain and the target domain; training the sub-network corresponding to each source domain using the common features in each source domain as input and the capacity estimate as output, using the mean error loss function; training the initial lithium-ion battery capacity estimation model through the overall loss function to obtain the lithium-ion battery capacity estimation model; and inputting the capacity degradation data of the target domain into the lithium-ion battery capacity estimation model to obtain the capacity estimate. This method can accurately achieve unsupervised cross-domain remaining cycle capacity estimation of lithium batteries.

Images