163 / 2025-06-08 17:30:07

Transformer Intelligent Fault Diagnosis Based on Knowledge-Enhanced Data Generation and Kolmogorov-Arnold Networks

Transformer Fault Diagnosis,DGA,Knowledge-Constrained GAN,KAN

Special Sessions > SSL 08 New perspective of intelligent detection, diagnosis, prognosis, and maintenance: Generative AI and Industrial Large Model

Transformer fault diagnosis is vital for ensuring the reliability and safety of power systems. Dissolved Gas Analysis (DGA) serves as a key technique in this field. However, the scarcity and imbalance of DGA data significantly hinder the performance of conventional diagnostic models. To address this challenge, this paper proposes a novel diagnostic framework that integrates Knowledge-Enhanced Data Generation with Kolmogorov-Arnold Networks (KAN). By incorporating domain-specific knowledge into the Generative Adversarial Networks (GAN), the generator is guided to produce high-quality synthetic DGA samples, effectively mitigating data scarcity and imbalance. Simultaneously, KAN is employed for its superior nonlinear modeling capabilities to achieve high-precision fault type identification. Experimental results demonstrate that the proposed method significantly improves diagnostic accuracy and robustness on real-world DGA datasets, offering an efficient and reliable solution for intelligent transformer fault detection.