Svm-based Distribution System Fault Line Selection And Ranging

The overhead line, cable and overhead line-cable mixed line are used in the power distribution network of our county. With the development of the city power system, power cables are widely used and the distributed capacitance also increased, so the Capacitive current dose not follow the running order. In order to extinguish instantly the grand arc the power distribution network use the neutral grounding coil operation mode, the resonant grounding system. When the single-phase ground fault occurred in a relatively low voltage power distribution network, the weak signal and the complexity structure of the power distribution network make the fault line selection and determining the fault location difficulty.In response to this situation, this paper analysis the resonant grounding system characteristics about the zero sequence current firstly when the single phase grounding fault occurred, then using S transform feature extraction of line energy as input samples. Based on support vector machine (SVM) classification and fitting function, this paper proposed a fault line selection and a fault distance location methods.The major work as follows:(1) Using the MATLAB simulation model, we establish a resonant grounding system, which includes overhead lines, cables and overhead-cable lines. And this system of fault simulation conducted a comprehensive comparative analysis of different fault conditions about zero sequence current waveform. To the proposed fault line detection and distance location methods provides an experimental basis.(2) Through simulating the resonant grounding system and ground fault we got transient zero-sequence voltage and current values and recorded a quarter of the frequency of fault lines zero sequence current. Using the good time-frequency analysis and feature extraction of S transform, on the recording zero-sequence current of S transform, we take the characteristic frequency points of the square matrix element as a characteristic energy spectrum. We make S-transform of each line of energy as the sample properties, and the corresponding fault line as the label. In order to get better generalization ability properties of the samples are normalized, then input to support vector machines for model training. With optimizing the parameters we obtain the support vector machine model distribution network fault line selection.(3) Simulate the resonant grounding system on different lines, different fault locations, different resistance and different initial transition phase in case of single-phase ground fault, and extract the energy spectrum of fault lines, then analyze it. The simulation analysis shows that the fault lines in the first four largest energy frequencies are distributed in lower frequencies department and the energy decay slowest, fault characteristics are best reflected. Make the sum of the square of the first four energy as a characteristic sample then input it to the fault location support vector regression machine. At last get the fault location model.