Study On Single-phase-earth Fault Line Selection Based On Improved PCA

The protection of small current grounding system is an important subject which needs to be urgently solved in our country.Due to the small residual fault and not obvious fault characteristics,there is quite low accuracy in existing protection method in the actual operation.At present,many protection methods use multiple fault features to carry out the selection of fault line.But,those methods ignore the association among the fault information and simply judge the fault by using widely training samples,which have some certain flaws.It is clear that the accuracy and robustness of the results have been improved.In this paper,a novel method of single-phase earth fault protection for distribution network is proposed.The proposed protection method normalizing the structure of various faulty information by using Copula function fully exploits the faulty information and considers the correlation among the fault information.In this paper,the maximum likelihood function and the marginal inference method IFM(Inference Functions for Margins)are used to study the parameter estimation of Copula function,and Akaike's Information Criterion,AIC is applied to find out the optimal fitting degree of Copula function.In addition,the best Copula function that is accord with electrical features over space-time is obtained giving detailed description concerning correlations between faulty characteristicsThe principal component analysis method that is in order to fully use the multiple faulty messages is used to complete the fusion of those fault characteristics which can reveal various characteristics indexes.This paper is aimed to improve the traditional methods' defect that can't accurately describe the non-normal distribution or non-elliptical distribution of fault data.The optimal correlation coefficient is constructed by using the optimal Copula function,and the linear correlation coefficient is replaced by the rank correlation coefficient,so as to propose the improved PCA that can process non-normal distribution of electrical characteristic variable data.In addition,this paper reduces the dimensions of original data and denoises along with the problem concerning the selection of primary component by accumulative contribution rate.The selection results are more reliable by upgrading the historical database associated with protection criteria building by Discriminant Analysis Method.In this paper,a 10kV power distribution system model is built and massive data has been obtained by changing the value of transition resistor and initial fault current angle.25 groups of untested sample data have been evaluated by the protection method proposed by this paper.The simulation results show that the protection method proposed by this paper can correctly find the faulty line under various fault conditions.