Cross-Domain Prognostic Method of Lithium-Ion Battery in New Energy Electric Aircraft With Domain Adaptation

Abstract

The prognosis of lithium-ion batteries is exceptionally critical in numerous industrial applications, and the precise estimation of battery health indicator is crucial for health management. In practical application, due to the differences in working conditions and service conditions of lithium-ion batteries applied in new energy aircraft, there exists certain discrepancy in the changes of indicators characterizing the health state during the charging and discharging process. Therefore, State of Health (SOH) prognosis models trained by one type of battery typically do not adapt well to other batteries. In this paper, the source domain data with labels and the target domain data without labels are utilized for training the prognosis model. The maximum mean discrepancy (MMD) is adopted to measure the distinction between the source and target domain samples. The proposed model relies on the convolutional neural network architecture for training. The fusion of estimation loss and inter-domain difference loss makes the model pay attention to the inter-domain gaps simultaneously during training and gradually reduces the domain gap. Experiments are carried out for validations, proving the fusion effects of two domains under different transfer degrees by setting the MMD loss weights. The proposed method guides the health index prediction of lithium-ion batteries for new energy electric aircraft. In general, the effectiveness of the proposed method has been well confirmed.