Transformer Fault Warning Based on Spectral Clustering and Decision Tree

The insufficient amount of sample data and the uneven distribution collected across faults are key factors limiting application machine learning in power transformer fault warning, as demonstrated by poor adaptability established data-driven models under actual operating conditions. In this paper, an unsupervised supervised method is designed for early warning based on electrical quantities vibration signals. Fourier levels signals different conditions measured field, features clustered according to their intrinsic properties means a spectral clustering algorithm. A decision tree model characteristics each cluster then constructed calculate values spectrum conditions, enabling assessment production variability. above process, which field measurement mining analysis methods, cheaper than existing techniques at home abroad makes better use information training models.