TDDAM: transformer based deep domain adaptation methodology for lithium-ion battery prognosis

Abstract

The status of health (SOH) is a vital indicator to characterize the remaining life of lithium-ion batteries (LIBs), and precise prognosis of the SOH is of great importance for battery management systems. In order to prognosis the SOH of LIBs, this paper proposed a Transformer based deep domain adaptation methodology (TDDAM). This paper applies the transformer model, which is widely used in natural language processing and other fields, to the prediction of LIBs. Meanwhile in order to solve the problem of model matching in different types of batteries or different environments, this paper combines domain adaptation method based on the maximum mean discrepancy. Firstly, we extract the data features of LIBs through position encoding and processing of the encoder structure with the multi-head self-attention mechanism as the core. Then, based on the maximum mean discrepancy index, the target domain data and the source domain data features are aligned, and the decoder part of the original transformer model is replaced with a fully connected layer for the prediction of SOH of LIBs in the target domain. This is the first time that a Transformer has been combined with the maximum mean discrepancy to be applied to LIBs prediction. Comprehensive experiments on two CALCE LIBs data showed that the TDDAM achieved smaller prognostic prediction errors over popular SOH diagnostic methods, indicating its great potential as a generic backbone for LIBs prognosis.