A data mining approach for transformer failure rate modeling based on daily oil chromatographic data

Abstract

Evaluating the real-time failure rate of transformers can effectively guide the planning of maintenance and reduce their failure risk. This paper proposed a novel transformer failure rate model that considers the impact of maintenance based on daily oil chromatographic monitoring data mining. Firstly, to ensure the quality of the modeling data, an improved k-nearest neighbor (KNN) algorithm based on genetic algorithm (GA) is proposed to repair the missing monitoring data. The repaired data is then mapped to the equivalent state duration (ESD) by the M-BPNN proposed, which is used to modify the multistate Markov process of transformers so as to quantify the impact of maintenance on failure rate. Considering the changing characteristics of the dissolved gases' content in the short period, the ESD is further merged in sequential periods to obtain the merged equivalent state duration (MESD). Finally, an analytical function of the transformer failure rate with respect to the MESD is obtained. Case studies on a typical substation demonstrate that the proposed approach has the ability to characterize the impact of maintenance and the actual failure rate, thereby improving the accuracy of the substation reliability assessment.