A DFP-Neural Networks Algorithm for Analysis of Power System Harmonics

This paper presents an algorithm of neural networks based on DFP to analyze the harmonics of power system. The main idea is to use the zero-crossing point method to calculate exact frequency, and then use DFP algorithm to train the weight of neural networks, and obtain the harmonics parameters at last. The algorithm can avoid the local least problem on grads descent method and the value problem of the learning rate. Besides, it does not involve the operation of the complex number and has a high convergence speed. The algorithm can show great advantage, when there is high harmonics. The simulating results show that the exact amplitudes and phases of high harmonics can be obtained very fast by using the algorithm.