To address the lack of physical interpretability and weak model generalization in purely data-driven methods for estimating the State of Health (SOH) of LIBs, this study proposes a lithium battery SOH estimation method based on Differential Temperature- Voltammetry (DTV) characteristics and Particle Swarm Optimization-Gated Recurrent Unit (PSO-GRU). To capture thermodynamic characteristics during battery aging, First, this paper computes DTV curves based on thermo-electrical coupling responses during charge-discharge cycles and extracts a 6-dimensional physical feature vector to quantify Multiphysics evolution patterns in battery aging. Second, a PSO-GRU prediction model is established, employing the particle swarm optimization (PSO) algorithm to adaptively optimize the hyperparameters of the gated recurrent unit (GRU). Additionally, to address the issue of significant data fluctuations in the early stages of battery aging that interfere with long-term trend prediction, a training set optimization method based on aging stage segmentation is proposed. Simulation experiments demonstrate that this method achieves significantly higher SOH estimation accuracy on the NASA battery dataset compared to standard GRU and long short-term memory (LSTM) models. After applying the optimization strategy, the model’s mean absolute error (MAE) on the test set decreased from 1.45% to 0.75%, and the root mean square error (RMSE) decreased from 1.86% to 0.97%, demonstrating enhanced generalization capability and robustness. The experimental results validate the necessity of excluding non-stationary data from the formation period for constructing high-accuracy, long-term prediction models, providing new insights for the engineering application of data-driven methods.
Han et al. (Tue,) studied this question.