Partial-Range SOC-Insensitive Model With EIS Change Pattern Recognition Model for Battery Aging Estimation

Electrochemical impedance spectroscopy (EIS) holds significant potential for evaluating battery degradation. However, EIS readings are not only affected by battery degradation but also by the state of charge (SOC). Traditional models for estimating battery capacity rely on impedances measured at specific SOC points, and thus can suffer from substantial inaccuracies when SOC estimation errors occur. To tackle this challenge, we propose a novel partial-range SOC-insensitive model for precise battery capacity estimation using transformer neural networks complemented by an EIS change pattern model based on the k-nearest neighbors (KNN) algorithm. To the best of our knowledge, this is the first study to develop an EIS-based battery capacity model that considers incorrect SOC scenarios. Test results show that our partial-range SOC-insensitive model can estimate battery capacity with a root-mean-square percentage error of 2.69%, even with a 30% SOC error, within the SOC range of 20% to 50%. Adding the EIS change pattern recognition model further improves the performance of the partial-range SOC-insensitive model, reducing the maximum absolute percentage error from 19% to less than 3% in scenarios involving 50% to 70% SOC error during battery cell testing.