Differentiated warning rule of power transformer health status based on big data mining

Abstract

Accurate on-line monitoring of transformer operation status is essential to ensure the reliability of power system. The dissolved gas analysis (DGA) method, which compares the gas concentration to the pre-set threshold, is most widely used for transformer status warning. However, existing standards apply either single threshold value or rough classification to diversified transformers with different properties, which therefore affects the accuracy of status warning. This paper establishes a differentiated warning rule through big data analysis mining. The Fuzzy C -means method is applied to identify optimal transformer properties, which can reflect the individualized characteristics of transformer to the best extent. As verified by the probability plot, the full sets of dissolved gas data under the selected properties conform to the Weibull Model. Association analysis is then carried out between the dissolved gas distribution characteristics and defect/ fault rate, and the warning thresholds are accordingly calculated. Correlating the gas concentration and gas increase rates with established warning values, the transformer operation status can be identified. The verification test indicates that the differentiated warning rule shows better performance than conventional methods and demonstrates an accuracy as high as 98.21%.