Ultra-high-accuracy state-of-charge fusion estimation of lithium-ion batteries using variational mode decomposition

Accurate estimation of state-of-charge (SOC) is crucial for improving the performance and lifespan of power batteries. Current research shows that applying data-driven techniques to compensate for SOC estimation errors from model-based methods often leads to overfitting and significant error fluctuations. Moreover, using battery model parameters as input features for data-driven methods raises concerns about interpretability. To address these challenges, this paper proposes an ultra-high-accuracy SOC estimation method that integrates the Extended Kalman Filter (EKF) based on the dual polarization model with an enhanced least-squares boosting regression tree (LSBoost) algorithm. Initially, the maximal information coefficient is used to evaluate and select model parameters highly correlated with SOC as additional input features for LSBoost. Subsequently, the variational mode decomposition technique is utilized to optimize the estimation results of the feature-enhanced LSBoost. An adaptive fusion mechanism is then designed to combine the advantages of the EKF and the enhanced LSBoost for SOC estimation. Finally, the proposed method is validated under five driving cycle tests, demonstrating that the mean absolute error and root mean squared error of SOC estimation results do not exceed 0.0673 % and 0.0993 %, respectively, thus achieving ultra-high accuracy in SOC estimation. This is significant for optimizing energy management in electric vehicles.