Research On Detection Of Harmonics In Power Systems Based On Neural Network And Wavelet Analysis

With the continuous development of modern industry,various electronic devices have been used in power systems.Consequently,nonlinear loads in power systems greatly increase,resulting in a lot of harmonic pollutions.Furthermore,this kind of harmonic pollution along with the development and advance of industry is becoming more and more serious,thus,it is particularly important to accurately detect these harmonics so as to suppress or remove them in power systems.This paper mainly introduces and compares several harmonic analysis methods in power systems: ones based on analog filters,based on Fourier transform,based on instantaneous reactive power and based on neural networks.Then,this paper shows the shortcomings of several analysis methods,and main methods utilized in current industry.By analyzing various orders of harmonic signals with wavelet transform,one can avoid deficiency of Fourier transform in analysis of non-stationary harmonics.Next,the dissertation analyzes the effect on harmonic analysis of different mother wavelets,and compares their advantages and disadvantages by numerical simulations,reaching a conclusion that the db4 wavelet function is the best for processing of non-stationary signals.Meanwhile,this paper completes a preprocessing of noisy power quality signals and finds that wavelet method has a good performance.On this basis,we choose a relatively new harmonic detection method—a combined method of wavelet and neural network.First,this method extracts features components with the help of wavelet transform,separates signals in different frequency bands,which are then fed into a neural network for the purpose of recognition,finally the parameters such as types,amplitudes and phases are achieved.This kind of method by combining wavelet and neural network takes the advantages of both the time-frequency locality of wavelet transform and adaptability of neural network,well implementing analysis and simulation of non-stationary harmonic signals.Simulation results verify the feasibility and effectiveness of the proposed method in this dissertation.