Research On Harmonic Detection Algorithm Of Power System Based On FFT,Wavelet Transform And Neural Network

With the development of modern power industry,a large number of non-linear power electronic devices are widely used in power system.Harmonic pollution of power quality is becoming more and more serious,which threatens the normal operation of power users and the safety of distribution network and equipment.Therefore,how to detect and suppress harmonics accurately and quickly is of great significance to improve the power quality.This paper focuses on four aspects:1.This paper describes the principle,harm and suppression of harmonic generation,discusses the advantages and disadvantages of several common harmonic detection algorithms at home and abroad,introduces the basic theories of Fourier transform,wavelet transform and neural network,and chooses the simulation examples respectively.2.A harmonic detection algorithm based on wavelet and double FFT is proposed,which is called WT-DFFT algorithm.The traditional FFT has some advantages in analyzing the steady-state harmonic,but it cannot detect the unsteady harmonic.The wavelet transform is suitable for detecting the unsteady signal,but it is not suitable for the detection of steady state signal.FFT and wavelet combined algorithm combine the advantages of both,and can detect both steady and unsteady signals simultaneously,but the practicability is not strong and the accuracy is not high,WT-DFFT algorithm is proposed in this paper,using the traditional FFT to identify harmonic distribution,determine the layer number of wavelet decomposition and frequency band,on the basis of the wavelet transform has been used to separate harmonic signals,to pay attention to the high frequency of unsteady signal,the wavelet threshold denoising refactored analysis its features,and to pay attention to the steady signal in low frequency band,add window interpolation FFT detected all of the steady state between integer harmonic and harmonic frequency,amplitude and phase.The algorithm has higher precision,practicability and adaptability.3.To improve the traditional BP neural network,the BP Neural Network with Adjustable Learning Rate,Momentum Factor and Activation Function Parameter is constructed.Because WT-DFFT algorithm to steady state signal detection accuracy there is still room for improvement,but the traditional BP neural network,by the strong learning ability and adaptive ability,for the steady state of integer harmonics detection precision is high,so to some extent,can make up for the inadequacy of the former.However,the learning rate and momentum factor of the traditional BP neural network are fixed,and it is easy for the training to show the phenomenon of slow convergence or oscillation near the convergence point.Its excitation function is a fixed function,so the detection of interharmonic wave cannot be completed.Vector of BP neural network,the momentum factor and variable parameters,the parameters of the excitation function into vector is constructed,the momentum factor and parameters of excitation function in the adjustment of the BP neural network,compared with the traditional BP neural network,the improved BP neural network can be optimized harmonic frequency,harmonics detection components,and the computing time greatly decreases,and the convergence is enhanced.4.In this paper,FFT,wavelet transform and neural network are proposed,which is called WT-DFFT-BP algorithm.Due to the improved BP algorithm still has the following drawbacks:(1)the detection of harmonics for,because it usually provides the weak signal,the network often in order to achieve the global minimum error and "sacrifice" the detection accuracy,and the closer and integer harmonic frequency span,the greater the precision is affected by the;In the noise environment,FFT preprocessing often fails to obtain the accurate spectrum,and it is difficult to determine the number of neurons.Without precise iteration initial value,it increases the difficulty of convergence and operation time.In the strong noise environment,the training is easy to fall into local minima,it is difficult to reach the predetermined precision,and the detection capability is almost invalid.(4)cannot be detected transient harmonic and so on,so this article integrated the WT-DFFT algorithm with the advantage of the improved BP neural network respectively,on the basis of accurate detection of transient harmonic,the steady-state signal detection accuracy for further optimization.By WT-DFFT algorithm to detect the total number of steady-state harmonic set for improving the number of neurons in BP network,the precision of harmonic coarse frequency as the basis to set up improved BP network harmonic frequency iterative initial value,adjusting the optimization of every steady-state harmonic frequency,amplitude and phase,and get all the steady-state harmonic frequency,amplitude and phase of high precision revised.Harmonics for the weak component detection problem,because the WT-DFFT algorithm to get the initial value of already has a better precision,thus setting frequency,amplitude and phase to iterate within a given threshold range,so that reduces the search range,avoid the integer harmonics and between the aliasing between harmonic component.The algorithm is of higher precision and stronger noise resistance,practical,real-time and adaptive ability,can accurately detect by a variety of harmonic signal constitute the complex power system harmonic signal.In view of the traditional FFT and window interpolation algorithm,wavelet algorithm,the proposed WT-DFFT algorithm,BP algorithm and improved BP algorithm is proposed in this paper,as well as the joint WT-DFFT-BP algorithm,this paper constructed a number of signal model,write a program in the Matlab software environment has carried on the simulation research.The simulation results show that the proposed algorithm is feasible.