ニューラルネットによる高調波負荷の分類

抄録

This paper presents an artificial neural-net based method for classifying harmonic loads in power distribution systems. The method is used to identify nonlinear relationship between harmonic loads and harmonic currents that vary from time to time. In recent years, nonlinear loads increase due to advanced technologies in power electronics applications. As a result, it is afraid that the harmonic distortion brings about several problems in power transmission and distribution systems. It is necessary to identify the harmonic loads and take an appropriate strategy so that the harmonic distortion is alleviated. However, this identification problem has not been studied so far due to the complex characteristics. The objective of this paper is to identify the nonlinear relationship between harmonic currents and types of harmonic loads as the first stage to detect harmonic sources. In this paper, a three-layered feedforward perceptron is utilized to classify harmonic loads. The neural network is effective for identifying nonlinear problems that have been hard to solve with the conventional methods. The weights between neurons are determined by the backpropagation algorithm. The proposed method has been successfully applied to several sample harmonic loads.