Fast Prediction Method for Early-Stage Remaining Useful Life of Lithium Batteries Based on Similar Samples

With the widespread use of lithium-ion batteries, accurate and fast Remaining Useful Life (RUL) prediction during early degradation is crucial. Current methods prioritize accuracy over speed and perform poorly for batteries with vastly different lifespans under varying operating conditions, this paper proposes a fast prediction method based on two-stage similar sample selection and Gaussian process regression (GPR). The approach first identifies the top five candidate samples most similar to the test sample based on early-stage battery data, then applies Kmeans clustering to analyze the later-stage features of these candidates, selecting the largest cluster as the final similar samples. Finally, GPR is employed for efficient training and prediction using the selected samples. Experimental results show the proposed method achieves 2.3% prediction error with 0.024s runtime on public datasets, significantly improving prediction speed while maintaining high accuracy.