A Hybrid Method of NARX and DS-Attention Applied for the State of Health Estimation of Lithium-ion Batteries

Due to the presence of capacity regeneration and random disturbances during the degradation process of lithium-ion batteries, the capacity decay curve of lithium-ion batteries exhibits a non-linear and unstable trend. Therefore, accurately estimating the SOH of lithium-ion batteries is critical to ensuring the safety and reliability of electrical device. This paper presents a hybrid method of nonlinear auto-regressive model with exogenous inputs neural network (NARX) and DS-Attention. The NARX model improves estimation performance by incorporating exogenous inputs as auxiliary factors, while the DS-Attention model is utilized to establish the mapping relationship between the input and output. The DS-attention model proposed in this paper is a specialized method of recurrent neural network (RNN), which improves the attention mechanism by adding division and self-adaptive functions. The division function improves the performance of NARX by effectively distinguishing external inputs and outputs. In addition, the self-adaptive function optimizes the query features in the attention mechanism. Experimental results demonstrate the exceptional performance of the proposed algorithm compared with similar algorithms by using different training set sizes and different numbers of neurons, and the advantage of DS-Attention is more obvious in the case of a smaller sample set.