A vision transformer-based deep neural network for state of health estimation of lithium-ion batteries

Accurate state-of-health (SOH) estimation of lithium-ion batteries (LIBs) is crucial to ensure the safety and reliability of electric vehicles. Deep learning has become a popular method for SOH estimation. However, this has mostly relied on convolutional neural networks (CNNs) and recurrent neural networks (RNNs), not fully exploring the potentialities of the method. This paper proposes a new procedure for SOH estimation of LIBs based on vision transformer networks (VITs). By analyzing the training speed and accuracy for different sampling points, an adaptive algorithm is designed to choose the most appropriate sampled data for the VIT, guiding battery data collection in actual systems, and reducing manual work during neural network training. Moreover, the VIT is improved by adding a dimension transformation layer, a multilayer perceptron (MLP) and a trainable regression token. Experiments carried out on two different datasets revealed that the proposed framework is able to reach an accuracy better than 0.01, which is superior to that achieved with other available techniques. The new approach has high robustness, good accuracy and applicability, and the VIT has great potential in SOH estimation.