A Stacked BWO-NIGP Framework for Robust and Accurate SOH Estimation of Lithium-Ion Batteries under Noisy and Small-Sample Scenarios

Lithium-ion batteries (LIBs) have been widely used in mobile energy storage systems because of their high energy density, long life, and strong environmental adaptability. Accurately estimating the state of health (SOH) for LIBs is promising and has been extensively studied for many years. However, the current prediction methods are susceptible to noise interference, and the estimation accuracy has room for improvement. Motivated by this, this paper proposes a novel battery SOH estimation method, the Beluga Whale Optimization (BWO) and Noise-Input Gaussian Process (NIGP) Stacked Model (BGNSM). This method integrates the BWO-optimized Gaussian Process Regression (GPR) with the NIGP. It combines their predictions using a stacked GPR model which reduces the problem of large input data noise and improves the prediction accuracy. The experimental results show that the BGNSM method has good accuracy, generalization ability, and robustness, and performs well in small sample situations. The Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) are as low as 0.218% and 0.164%, respectively, which is close to 0. At the same time, R-Square (R2) is as high as 0.9948, which is close to 1, indicating that the estimated results in this paper are highly consistent with the actual results.