State of charge estimation of lithium-ion battery based on adaptive boosting algorithm optimised extreme learning machine

Abstract

State of Charge (SOC) is critical for accurately determining the remaining capacity of lithium-ion batteries in electric vehicles. This study employs an improved adaptive boosting algorithm (AdaBoost) to estimate the SOC of these batteries. To enhance SOC estimation accuracy, we develop a novel approach integrating an improved AdaBoost algorithm with a Sparrow Search algorithm-optimised Extreme Learning Machine (ELM-SSA) as the weak learner. The implicit layer bias of the Extreme Learning Machine (ELM) is traditionally selected through empirical methods, which can compromise its estimation performance. To address this limitation, this study employs the Sparrow Search Algorithm (SSA) for global parameter optimisation of the ELM. The experimental data underwent a two-stage denoising and smoothing process using variational modal decomposition (VMD) combined with the Savitzky-Golay filter method (VS). The preprocessed data were subsequently utilised as input for our proposed SOC estimation methodology. The method’s effectiveness and robustness were validated using a dynamic operating condition dataset from Panasonic 18,650 battery cells. Through comparative analysis with LSTM, CNN-LSTM, ELM-SSA, and VSELM-SSA, the proposed method demonstrates favourable predictive performance. Specifically, it achieves maximum mean squared error (MSE) and variance of 0.000757 and 0.000756 respectively, and maximum mean absolute error (MAE) of 0.0203.