Dynamic double similarity fusion based on ΔQ power law for early-cycle RUL prediction of lithium-ion batteries

Accurately predicting the remaining useful life (RUL) of lithium-ion batteries is essential for ensuring system safety and minimizing economic losses. However, existing methods generally rely on numerous feature parameters while underutilizing the potential of historical sample data. Additionally, in the early stages of RUL, the weak correlation between capacity and historical samples often compromises prediction accuracy. To address these challenges, this paper proposes a novel prediction framework, DDSF-QD (Dynamic Double Similarity Fusion model incorporating ΔQ Power Law and Dynamic Time Warping). First, the original data are denoised using the CFK method (CEEMDAN-Fuzzy Entropy-K-means) to improve data quality. Then, the proposed ΔQ(V) power law is constructed to evaluate lifetime similarity, and combined with the DTW (Dynamic Time Warping) algorithm to calculate the initial spatial similarity. High-quality historical samples are used to train the network, and then their corresponding weights are dynamically fused within the DDSF (Dynamic Double Similarity Fusion model) framework to update sample weights. This enables the prediction network to estimate future capacity degradation trends and predict RUL based on a predefined failure threshold. Experimental results on classical lithium-ion battery aging datasets demonstrate the superior performance of the proposed method, achieving a coefficient of determination (R²) of 99.85%. To the best of our knowledge, both the ΔQ(V) power law and the DDSF framework are proposed for the first time, highlighting their potential in battery health management. • The ΔQ power law model is established to obtain high-quality lifetime similar samples. • The DDSF model is developed to dynamically adjust the fusion coefficients. • The CFK model is employed to optimize the denoising process.