Application analysis of PNN method in fault diagnosis

This study conducts an in-depth exploration of the innovative application of probabilistic neural networks (PNN) in the field of power transformer fault diagnosis. By combining the improved three-ratio method and selecting five gases—methane, ethane ethylene, acetylene, and hydrogen—three groups of key ratios are constructed as input feature vectors for the PNN, significantly enhancing the fault characterization capability. Meanwhile, experiments based on 33 sets of industrial on-site transformer fault data verify that the diagnostic accuracy of this method exceeds 93%, providing a reliable technical path for preventive maintenance in power systems.