SOH Prediction of Battery Packs Using Dynamic Graph Convolution Combined with KAN-Driven Methods

Accurate prediction of battery pack State of Health (SOH) is crucial for ensuring the safety, reliability, and efficient operation of large-scale energy storage systems, such as energy storage power stations. To address the challenges posed by the complex spatiotemporal interactions and nonlinear dynamic characteristics among battery modules, this paper proposes a novel predictive framework that integrates a dynamic graph generator with a Kolmogorov–Arnold Network (KAN). The dynamic graph generator models the time-varying interactions among battery modules through a dynamic adjacency matrix, effectively capturing intricate spatiotemporal dependencies. Meanwhile, the KAN leverages nonlinear feature decomposition to facilitate efficient extraction and fusion of high-dimensional features, further capturing the temporal dependencies of battery degradation, thereby enhancing the predictive accuracy and robustness of the model. Experimental results demonstrate that the proposed framework exhibits superior performance across multi-scale prediction tasks, accurately capturing the SOH evolution trends of battery packs. This study provides an efficient and reliable technical solution for SOH prediction, meeting the intelligent health monitoring requirements of energy storage power stations, renewable energy grid integration, and power system optimization, thereby offering significant potential for broad engineering applications..