SF6 Decomposition Components Fault Diagnosis Based on Gaussian Process Classification and Decision Fusion

Abstract

The decomposition of SF6 is closely related to the internal insulation status of equipment. Using SF6 decomposition component information can effectively diagnose the internal insulation fault of gas-insulated equipment. This method has been widely concerned by the power industry. However, due to the limited amount of SF6 fault decomposition component data accumulated now and the vast majority of them are laboratory simulation data, the application of this method in engineering is limited. According to the characteristics of SF6 decomposition component data, this article proposes a fault diagnosis strategy based on Gaussian process classification (GPC) and decision fusion (DF). On the basis of a large amount of experimental data and on-site fault data accumulated in the early stage, the existing data are expanded by piecewise cubic spline interpolation. To improve the generalization ability of fault diagnosis, the expanded SF6 decomposition component data is applied to the ensemble learning model. The basic learner of ensemble learning will introduce the GPC model based on Markov chain Monte Carlo (MCMC) sampling. At the same time, an improved scheme of Bagging integrated learning is proposed, in which the voting rule is improved to the combination rule of evidence theory. The proposed scheme of “data driven + model driven” improves the diagnostic accuracy to 97%. It provides a set of efficient diagnosis strategies for the power industry.