State of health estimation with attentional long short-term memory network for lithium-ion batteries

With the rapid growth of electric vehicle production, the market demand for lithium-ion batteries also shows a high growth trend. The state of health (SOH) estimation of lithium-ion batteries plays an important role in ensuring the safe and stable operation of electric vehicles. In this paper, we propose a novel SOH estimation method based on an attentional long short-term memory network (LSTM) with multi-source features, in which we consider eight health indicators extracted by analyzing the incremental capacity (IC), differential temperature, and differential thermal voltammetry curves. Specifically, to better complement the description of the IC curves, the Wasserstein distance is introduced as a health factor. Moreover, to improve the performance of our estimation model, an attention mechanism is embedded in the LSTM model to focus more on the critical information. The local attention mechanism uses a fixed window centered to calculate the weight coefficients of attention to address the drawbacks of global attention mechanisms. Finally, the model and feature validation experiments are conducted on two datasets. The experimental results demonstrate that the LSTM based on the local attention mechanism outperforms the traditional LSTM and LSTM based on the global attention mechanism in terms of accuracy for SOH estimation.