Fault Line Detection For Indirectly Grounding Power System Based On Genetic Algorithm Optimizing Neural Network

Non-effectively grounding neutral system is widely used in medium voltage distribution network at home and abroad. The fault current is small when single phase to ground fault is occurred, therefore the power system can continue maintain 1~2h power. It should to elect and remove the fault line quickly because the non-fault phase voltage is increased, the insulation may be damaged as the time increasing. It is very significant to research and design a reliable method of fault line selection to improve the safe operation of power system.Firstly, the fault characteristics for single phase to ground fault in non-effectively grounding neutral system are studied. The fundamental characteristics, transient characteristics and harmonic characteristics of zero sequence current under fault conditions are analyzed.Secondly, theories of neural network are introduced, though analyzing the fault characteristics for single phase to ground fault in non-effectively grounding neutral system, a 3-8-3-1 structure of BP neural network model is established in this dissertation. Three kinds of fault features are defined and extraction methods are proposed, BP neural network method to fault line selection is designed by the model already proposed. The defect of this method is analyzed, and solutions are proposed.Thirdly, an improved adaptive genetic algorithm is proposed based on the analysis the genetic algorithm factors such as fitness function, crossover rate, mutation rate and other factors. In order to improve learning efficiency and avoid to local minima, the dissertation use improved adaptive genetic algorithm optimization neural network weights and thresholds. Then, the line selection model obtains an excellent convergence and extensive adaptability.Finally, using EMTP and MATLAB, the fault characteristics for single phase to ground fault are carried out. Neural network is optimized for improved adaptive genetic algorithm, the optimal solution of improved adaptive genetic algorithm is assigned to neural network. The test samples that obtained by simulation of the neural network model is tested. The line selection model has an ideal line selection effect though comparing to the BP neural network model with random weights and thresholds or simple genetic algorithm optimization BP network model.