Noise Data Removal Method of Frequency Response Curve Based on MNKriging Interpolation Algorithm

Abstract

The frequency response method is one of the most commonly used methods for determining the winding deformation of a power transformer. However, in the actual measurement, the frequency response curve of the test results will generate spiking data due to interference. If this impact cannot be accurately identified and reduced, the availability of the test results will be seriously affected, causing difficulties and even errors to the analysis and judgment of the test. A multidimensional nonuniform Kriging (MNKriging) interpolation algorithm is proposed in this paper to eliminate spiking data in the frequency response curve and improve the accuracy of the test results. The method constructs a nonuniform multidimensional deformation field model by using the optimal weight coefficient combination, and optimizes it with the particle swarm optimization algorithm to extend the Kriging interpolation to the multidimensional nonuniform field space. It has been applied to the interpolation of the transformer frequency response data. The results prove that the method reduces the noise interference to a certain extent, and therefore the points on the frequency response curve of the transformer winding deformation subject to noise interference are well recognized and repaired.