Piecewise Iterative Algorithm Based Analysis Method for Power Hamonic Neural Network
Wu-bing CAO,Chun-mei ZHANG,Yong-sheng LIU,Zhong-wei NIE,Chao YANG,Zhi-hong FU
DOI: https://doi.org/10.3969/j.issn.1001-1390.2014.23.013
2014-01-01
Abstract:To improve the analytic speed, accuracy and anti-noise performance for harmonic and inter-harmonic a-nalysis, a piecewise iterative algorithm based enhanced Adaline neural network is adopted in this paper.Firstly, Han-ning-window double spectrum line interpolation FFT algorithm is used to make the sampling data divided according to the sampling time and to adjust the enhanced Adaline neural network parameters based on corresponding errors for each segment.Combing one-point and all-point iterative algorithm, the proposed method average the errors from all time frames to reduce the effect of noise on parameters adjustment, as well as preserve the harmonic and inter-harmonic in-formation included in errors, so as to increase the network accuracy.In addition, according to the relationship between parameter estimation errors and the first order partial derivation of error function with respect to its activation functions, a frequency adjustment method corresponding to the maximum component of correction signal amplitude is proposed to improve the real-time performance and accuracy.Simulation results verify the correctness of the analysis conclusions.