Comparison between three back-propagation algorithms for power quality monitoring

In this paper Levenberg-Marquardt, Conjugate gradient, Resilient back-propagation algorithms are compared for power quality monitoring. Three Networks are trained in MATLAB. Each network is trained with the single algorithm mentioned above. Data for training is generated with the help of numerical model of power quality events in MATLAB. Voltage sag and swell is taken into consideration. The networks are so trained that it should detect and classify the voltage sag/swell accurately. Training performance of each network is presented with the help of performance and validation graph. Trained networks are tested with the help of simulation model. Simulation model is made in MATLAB which can generate sag/swell of any magnitude for any time period. Algorithms are compared on the basis of the ability of trained network to detect as well as classify the sag/swell and the performance of training.

I. Introduction

The power quality monitoring includes the detection and classification of power quality disturbances such asvoltage sag, swell, harmonics, transients, flicker etc. Each disturbance is characterized by magnitude, shape and frequency of the voltage signal. The effects of these power quality events are hazardous for the domestic as well as industrial customer [1]. To avoid loss of money and time, it is necessary to mitigate these disturbances. To mitigate the disturbance it is very necessary to find the right cause of the problem and what the problem is. Hence to find out the problem, monitoring of the power quality is very necessary. There are several methods are proposed like Fourier transform, Short time Fourier transform, wavelet transform, Stockwell transform, artificial neural network, fuzzy logic [2].