Study Of Single Phase-to-ground Fault For Neutral Un-effectual Grounded System Based On Support Vector Machine

Neutral un-effectual grounded system occupies a very important position because of its reliability of power supply in the medium and lower voltage distribution networks in our country. Moreover, single phase-to-ground fault accounts for a large proportion in distribution networks. When single phase-to-ground fault in neutral un-effectual grounded system occurs, it has an important practical significance to find out the fault line and the fault point as soon as possible for stable operation in distribution networks. Therefore, fault line selection and location for single phase-to-ground fault in neutral un-effectual grounded system have been always a very popular research subject.Firstly, the thesis reviews and simply analyzes existing methods for single phase-to-ground fault line selection and location in neutral un-effectual grounded system, summarizes the advantages and defects of these methods, and points out the confronting problems.Secondly, the thesis simply introduces the concepts of neutral ungrounded system and neutral extinction-coil grounded system. And then it respectively analyzes the features of steady state and transient state, takes neutral ungrounded system for example, and verifies the correctness of the theory by building model under Matlab/Simulink platform.Thirdly, the thesis introduces the related concepts of statistical learning theory as a foundation, and then elaborates Support Vector Machine algorithm with the derivation of formulas. It builds a support vector machine model for single phase-to-ground fault in neutral ungrounded system. Before using support vector machine, it needs to denoise the original data by the method of wavelet analysis, and data preprocessing needs to determine the appropriate wavelet function, layer number of decomposition and threshold processing method. After denoising, to extract the features of new data as sample properties, the thesis proposes a method by using the cumulative sum of the absolute values of zero-sequence current to highlight fault feature. And then, it sets the labels corresponding samples for support vector machine, trains the training data and optimizes the parameters during the training process, then it obtains the fault line selection results with the test samples inputting the trained model. The results show that it has a higher accuracy of line selection by using this method and it has a certain feasibility.Finally, it introduces the related concepts of Support Vector Regression and makes the derivation of formulas in detail. Based on data preprocessing, the absolute values of zero-sequence current are used to extract feature. It finds out that the quantities that can reflect the characteristics are concentrated in a just short period of time after fault occurs through observation. Therefore, it uses the absolute values of zero-sequence current in this period of time as the independent variables of support vector regression. Train the support vector regression with independent variables and the corresponding dependent variables, optimize the parameters during the training process, and then obtain the fault location results with the test samples inputting the trained model. The results show that in most cases this method there are the smaller error between the predicted values and actual values, and fault location is basically completed.