A Hybrid Physics-Data Driven Framework for Online Monitoring of Winding Insulation Degradation

Online assessment of winding insulation reliability is essential for ensuring safe and stable operation of electrical assets and preventing costly downtime. However, due to the weak and highly coupled characteristics of insulation degradation, accurately extracting relevant features from complex and multi-source monitoring signals remains a significant challenge. Traditional single modeling approaches, whether based solely on physical models or purely data-driven technique, often suffer from model inaccuracies and insufficient samples, which limit their applicability. This paper proposes a hybrids physics-data driven method for online insulation degradation assessment, which achieves feature alignment, complementary enhancement, and sample generation across simulation and experimental data sources, effectively alleviating the small-sample problem in insulation aging analysis. Experimental results demonstrate that the proposed hybrid method significantly improves sample diversity and preserves physics consistency, achieving an average insulation reliability assessment accuracy of 97.1%, with substantial improvements in both accuracy and stability over traditional methods. This work offers a novel approach for precise and efficient online monitoring of winding insulation.