Dissolved Gas Analysis of Transformers Based on Kolmogorov-Arnold Network

Dissolved Gas Analysis (DGA) is a pivotal technique for diagnosing potential faults in power transformers by analyzing gases dissolved in transformer oil. Besides, Traditional machine learning methods, such as SVM and CNN, may not fully capture the complex relationships between gas concentrations and fault types and often struggle with precision. This paper presents a novel approach using the Kolmogorov-Arnold Network (KAN) for DGA, leveraging its capability to model complex, non-linear relationships within the data. The KAN-based model is trained with supervision on a comprehensive dataset of DGA data containing the concentration of five gases in transformer oil-Hz, CH4, CzHz, CzH4, CzH6 and the corresponding fault type. Experimental results demonstrate that the KAN-based DGA model not only improves diagnostic accuracy but also enhances the reliability of fault detection in transformers. This innovative approach holds promise for advancing the predictive maintenance and operational safety of power distribution systems.