CNN-transformer based SOH prediction study for lithium batteries

State-of-health (SOH) prediction of lithium-ion batteries is a core task to optimize the battery management system (BMS) and extend the range. Aiming at the complex nonlinear and time-varying characteristics of the battery aging process, a SOH prediction method based on CNN-Transformer hybrid neural network is proposed, which serially fuses the local feature extraction capability of convolutional neural network (CNN) with the global time-series modeling advantage of Transformer. First, the voltage, current, and other time-series data are dimensionality reduced after the local as-health factors extracted from the battery charging and discharging curves by the CNN; and its regression prediction is achieved by using the Transformer encoder in conjunction with adaptive average pooling and fully connected layers to capture the long-term dependence of the capacity degradation. Experiments on NASA’s public dataset show the superior performance of the model, with significant improvement over single LSTM, CNN and Transformer models. The ability of the model to focus features on critical charge/discharge phases is verified. This study provides an efficient solution for battery health management and theoretical support for coupled modeling and dynamic optimization of charging and discharging strategies.