A method for predicting winding insulation deterioration of power transformers considering fragmentation and censoring defects of monitoring data

Aiming at the problem that the accurate prediction of winding insulation deterioration trajectory is hindered due to the fragmentation and double-censoring defects in transformer deterioration monitoring data, which in turn impedes the condition assessment and fault warning of transformers, this paper proposes a method for predicting the winding insulation deterioration of power transformers, taking into account the fragmentation and censoring defects of monitoring data. First, the fragmentation and random censoring characteristics of transformer deterioration data are analyzed, and the defective data are processed by piecewise weighted ensemble imputation to achieve highly reliable data reconstruction. Second, Gaussian process regression is used to smooth the interpolated data, which can suppress the interference of measurement errors and reduce the uncertainty in the interpolation process. Third, based on the theory of functional principal component analysis, the common deterioration trends and individual difference characteristics are extracted from the interpolated deterioration data to construct a deterioration estimation model. Finally, the Bayesian optimization algorithm is introduced to dynamically update the model, thereby realizing the accurate prediction of the transformer deterioration trend. The results show that the proposed method can accurately predict the transformer deterioration trend when the deterioration data present different degrees of fragmentation and censoring defects.