The Application Study Of Support Vector Regression Machine On Interharmonics Parameter Estimation

With the development of modern power electronics, a large number of nonlinearand impact load put into use, which making the electric system quality getting worse.The decline in power quality not only affects the using of user equipment, but alsothreatens the safe operation of the power system. Harmonic is one of the most commonproblems of power quality. However, there are not only integer times harmonic in thepower system, but also non-integer times harmonic ones which called interharmonics.Interharmonics cause more problem to the power system, not only may cause lightingequipment flicker, but also will influent the use of some equipment or destroy them. Thetreatment to harmonics and inter harmonics is based on knowing the parameters of them,so, it is important to study the interharmonics carefully.This paper first analyzes some traditional harmonic analysis method and theirshortcomings of dealing with the interharmonics, then will use the Support VectorMachine to do the harmonics and interharmonics parameters estimation which is a newhot spot in machine learning after the neural network.After describing the related knowledge of statistical learning theory and SVM, thispaper will use the kernel function base on Fourier spread to map the mapping the inputsamples into high dimension space, then use the function regression with SVM to getthe amplitude and phase information of interharmonics. The mapping is required toknow the number and frequency information of harmonic and interharmonic, and theaccuracy of the information of frequency have high demand, this paper adoptsRoot-MUSIC algorithm to estimate frequency information, which can guarantee theaccuracy of frequency estimation obtained in low SNR can be used by SVM.In the view of traditional SVM will use much more time on computing because ofthe growing number of samples, this paper adopts an iterative LS-SVM based on Renyientropy algorithm to improve the computing speed. As the parameters estimation errorwill be bigger under the condition of serious waveform distortion, this paper proposes anoise reduction method based on SVM and a noise reduction method based on statisticrandom error theory. The two kind of method both has advantages and disadvantages,and they are both verified to reduce the error by simulation.