Research On Problems Of S-transform Based Power Quality Disturbances Classification

With the large number of applications of non-linear devices in the power grid,especially the development of new energy power generation such as photovoltaic and wind power,and the high requirements of the power quality by users,more and more attention is paid to the power quality problem by the operators and consumers of the power grid,which becomes one of the hot issues in the field of power.As a prerequisite and preparation of power quality control and management,it is very important to detect and classify the power quality disturbance accurately.In this paper,we mainly study the classification of common PQDs(short for power quality disturbances),and make an in-depth discussion on the S-transform.Improved schemes are proposed for the shortcomings of S-transform.A new PQD classification algorithm based on decision tree and improved S-transform is designed,which can distinguish eight kinds of single PQD and four kinds of mixed PQD,and can maintain a high classification accuracy in different noise intensity.The research background of power quality problem is introduced in the first part of this paper.The research status of power quality detection and analysis is summarized,and the research results of power quality disturbance classification based on S transform in recent years are reviewed.Then,the principle of S transform is deduced and analyzed,and a new optimal parameter selection method is proposed for the improved S-transform based on the Gaussian window function.A new improved S-transform based on sparse spectrum is proposed for the shortcoming of heavy computation of S-transform.The improved effects of the two proposed methods is verified by Matlab simulations.Then,the time and domain characteristics of the single and mixed power quality perturbations are analyzed with S transform.According to the different characteristics of the various PQDs after S transformation,five feature values with high discrimination are selected and used.The feature values are calculated for a large number of instances of every PQDs,and threshold values are selected for each feature value for different PQDs using statistical principles Finally,a PQDs classification scheme based on decision tree is designed.The simulation results show that the scheme can classify 8 single PQDs and 4 complex PQDs,and can maintain a high classification accuracy in the case of noise.