Research On Fault Line Selection Of Ineffectively Grounding Power System Based On Support Vector Machine

As an important part of life, power system running must always be safe and reliable.Since the characteristics of ineffectively grounding power system and unobvious faultfeatures, the problem of fault line selection has been studied for years and still has not beensolved perfectly.Many kinds of fault line selection methods are gradually proposed, but each of themethods has some limitations and shortcomings. In order to solve the problem, thisdissertation focuses on the study about how to select the fault line with Support VectorMachine (SVM). The main contents are as follows:(1) Single-phase grounding fault steady state feature and transient state feature aresystematically analyzed. According to the actual structure of ineffectively grounding powersystem and based on electromagnetic simulation software, the simulation model of groundingfault in ineffectively grounding power system is set up.(2) SVM and Least Square Support Vector Machine (LSSVM) specific ideas andclassification principles are analyzed. On the one hand, a fault line selection method based onwavelet packet transform and LSSVM is presented. Through calculating the energy withcoefficients of each band, the maximum energy value as the feature of LSSVM. Simulationresult proves that the method is valid. On the other hand, a fault line selection method basedon wavelet packet transform improved by LSSVM is presented. The method aims at increasethe accuracy of fault line selection.(3) In order to improve the method applicability of fault line selection, a fault lineselection method based on information fusion by LSSVM is proposed. Steady-statefundamental component, steady-state fifth harmonic component, and transient energycomponent extracted from zero-sequence current separately by means of Fast FourierTransform (FFT) and wavelet packet transform. Various features are combined by LSSVM toimprove the accuracy of fault line selection.(4) In order to resolve the problem of select LSSVM model parameters which has greatinfluence on classification accuracy and generalization ability, the swarm intelligenceoptimization algorithm is proposed. By means of Quantum Particle Swarm Optimization(QPSO) to optimize the model parameters, then fault line selection model training and testingby using the optimized parameters. Comparing with Particle Swarm Optimization (PSO) andgird search method, the QPSO method has higher reliability and accuracy.