A novel framework with dynamic spatiotemporal graph modeling and adaptive optimization for lithium-ion battery remaining useful life prediction

Accurate prediction of remaining useful life (RUL) for lithium-ion batteries is critical for ensuring the reliability of electric vehicles. Conventional methods however fail to model nonlinear degradation dynamics and cannot efficiently capture charging features that are linked in space and time. To address these limitations, this work presents an integrated framework that integrates three methodological innovations: the Dimensionality Reduction for Ordered Points using Principal Component Analysis (DROPP-PCA) approach compresses sequential charging features by leveraging ordered sequences of feature vectors and constructing covariance structures that fit Gaussian processes. It extracts the principal components that have a correlation of more than 0.8 with the remaining capacity, which reduces the root mean square error by 75.0% on the MIT dataset and 31.0% on the XJTU dataset compared to conventional principal component analysis. The TimeFilter spatiotemporal graph topology model produces context-aware subgraphs (temporal, spatial, and spatiotemporal) in real time using a mixture-of-experts router. To do this, it breaks up multivariate sequences into smaller parts to look for electrochemical signatures. On the test sets of both the Massachusetts Institute of Technology and Xi'an Jiaotong University datasets the proposed framework achieves an R2 greater than 0.996 for remaining capacity prediction. The Status-Based Optimization (SBO) method adjusts hyperparameters using a performance-based learning procedure where each hyperparameter configuration is evaluated by validation error. This reduces bias in the remaining capacity estimation. The model predicts the remaining capacity for future cycles, from which the remaining useful life is derived based on the end-of-life threshold. This method blends bio-inspired optimization, topological dependency modeling, and mathematical dimensionality reduction to provide a new way to properly assess the health status of lithium-ion batteries. It makes real-world battery management systems far better at predicting future behavior and more efficient.