The Neural Network Analysis Method Based On Piecewise Iteration For Electrical Harmonics And Interharmonics

The extensive applications of non-linear loads and power electronic devices resultin the electric network voltage and current waveform distortion is more and moreserious. There is not only high integer harmonics content but also many non-integerharmonics which are interharmonics, threatening the safety and economic operation ofthe power system and electrical equipment. The accurate estimation of harmonics andinterharmonics parameters is the premise for their governance. Therefore, the study ofharmonics and inter-harmonics analysis methods is important.This paper comprehensively compared Fourier transform, wavelet transform,modern spectral estimation and artificial intelligence in accuracy, timeliness, scope,stability, adaptability and other aspects. Updated Adaline neural network belonging tothe artificial intelligence was used in harmonics and interharmonics analysis because ofits adaptive learning, adjusting the network activation function vector and weight vector,high accuracy, but still in its infancy.Hanning window interpolated FFT algorithm and updated Adaline neural networkwere combined to overcome the disadvantage of updated Adaline neural network that itscalculation accuracy depended on the initial value of the network.FFT transformestimating the frequency, amplitude and phase is the most elementary for the analysis ofharmonics and inter-harmonics, but there are spectral leakage and fence effect. Thispaper used theoretical analysis, formula derivation and Matlab simulation to verify thewindowed interpolation algorithm can effectively inhibit the FFT spectrum leakage andfence effect.In order to improve the accuracy and anti-noise performance of updated Adalineneural network, piecewise iteration algorithm was proposed in this paper which dividedsampled data according to the sampling time, and adjusted the variable parameters ofthe network with the error function of each section. This algorithm combined theadvantages of one point iteration algorithm and all points iteration algorithm, bothaveraging the first-order partial derivatives of error function with respect to the adjustedparameters, reducing the impact of noise on the network training, and fully retained theharmonics and interharmonics information included in the error function, improving theparameter estimation accuracy of the network. In addition, according to the relationshipbetween the parameter estimation errors and the first partial derivative of error function with respect to frequency, the amount of frequency adjustment of the largest amplitudecomponent was corrected to improve the network’s accuracy. Simulation results showedthat the harmonic and interhaimonic analysis method proposed in this paper achievedhigh precision, strong resistance to noise, and real-time.Finally, the actual voltage and current measurement data of computer, table lamp,washing machine and forging factory was analyzed by updated Adaline neural networkbased on the Hanning windowed interpolation FFT algorithm. The analysis resultsdemonstrated that the proposed neural network analysis method based on piecewiseiteration for electrical harmonics and inter-harmonics had a high accuracy and stability.