Research on classification diagnosis of DC magnetic bias degree of power transformers based on probabilistic neural...

As the most important equipment in the power system, the operation state of the transformer directly affects the stability and safety of the power grid. Once the transformer has a DC magnetic bias fault, it will lead to problems such as excessive temperature, increased vibration, and excitation current distortion. To quantitatively classify the DC magnetic bias degree of power transformers, this paper proposes a classification method based on a probabilistic neural network optimized by the Mirage Search Optimization algorithm. First, a three-phase two-winding transformer simulation model is established in PSCAD to generate samples under different DC magnetic bias conditions, and the bias degree is divided into four categories: normal, slight, middle, and heavy. Neutral-point DC current and excitation-current total harmonic distortion are selected as excitation-response input features, and their sufficiency is verified through feature relevance analysis and ablation experiments. The MSO algorithm is used to optimize the smoothing factor of PNN, thereby improving the classification boundary and generalization ability of the model. Experimental results show that the proposed MSO-PNN model achieves a mean accuracy of 98.15% over 30 independent runs, with the best accuracy reaching 99.01%, outperforming other optimized PNN models. Under simulated Gaussian-noise conditions, the model maintains an accuracy of 96.4% at SNR = 15 dB and 91.0% at SNR = 10 dB. These results indicate that the proposed method provides an accurate and robust diagnostic framework for transformer DC magnetic bias degree classification.