Abstract—In this paper, we analyze and test a scheme for the estimation of electrical fundamental frequency signals from the harmonic load current and voltage signals. The scheme was based on using two different Multi Layer Artificial Neural Networks (ML-ANN) one for the current and the other for the voltage. This study also analyzes and tests the effect of choosing the optimum artificial neural networks’ sizes which determine the quality and accuracy of the estimation of electrical fundamental frequency signals. The simulink tool box of the Matlab program for the simulation of the test system and the test of the neural networks has been used. Keywords—Harmonics, Neural Networks, Modeling, Simulation, Active filters, electric Networks. I. INTRODUCTION OR the last few years, many different topologies have been developed for harmonic currents and voltages extraction from the AC line. The quality, speed and accuracy of these extracted signals are very important in active harmonic filter control. Some topologies for the extraction are based on the classical fast Fourier transform theory, Instantaneous Power Theory (IPT), ADALINE Artificial Neural Network (ANN) and Particle Swarm Optimization (PSO). The classical fast Fourier transform theory is the most intuitional and basic method that can virtually solve any composition and decomposition problems but it takes a long time to solve the equations, which is not suitable for the online power filtering and instantaneously varied signals. Others propose [1] to use four ADALINEs as an alternative for online extracting of the direct, inverse, and homopolar voltage components from a composite voltage. The first two ADALINE (the Current ADALINE) extracts the harmonic components of the distorted line current signal Wael M. El-Mamlouk is Senior Electrical Engineer, Cairo, Egypt (e-mail: wmamamlouk@yahoo.com). Metwally A. El-Sharkawy is working with Electrical Power and Machines Department, Ain Shams University, Cairo, Egypt (e-mail: masharkawy@yahoo.com). Hossam.E. Mostafa is working with Electrical Department, Jazan University, Jazan, Kingdom of Saudi Arabia (e-mail: hossam65eg@yahoo.com). and the second two ADALINE (the Voltage ADALINE) estimates the fundamental component of the line voltage signal. Reference [3] presents an algorithm for harmonic estimation. It utilizes the particle swarm optimizer with passive congregation (PSOPC) to estimate the phases of the harmonics, alongside a least-square (LS) method that is used to estimate the amplitudes. The estimation accuracy is greatly improved in comparison with that of the conventional discrete Fourier transform. Some controls of the active filter’s current are performed by means of the dead-beat control technique which calculates the phase voltage; so as to make the phase current reaches its reference by the end of the following modulation period. A serious drawback of this control technique is an inherent delay due to the calculation time [4]. The goal of the study is to analyze and test a scheme based on using two different Multi Layer Artificial Neural Network (ML-ANN) with shift method for input samples[5] using a sample by sample investigation of the input signal. The scheme was based on using two different (ML-ANN), one for the current and the other for the voltage. These tests are conducted using three different architectures employing ML-ANN to compare the THD of the estimated fundamental signals. We used the current harmonic modeling contents of an adjustable speed drive (ASD) as an example of a harmonic produced load connected at different locations of the test system consisting of 13 buses Balanced Industrial Distribution System extracted from the whole system presented at [6] and with different loading conditions and at different locations for the harmonic loads. We test the accuracy of the estimated fundamental frequency components by decomposing the output signal using the classical fast Fourier transform theory for the twelve cases under the test stage of the ANN. II. GENERAL DESIGN OF THE ML-ANN The sizes of networks depend on the number of layers and the number of hidden-units per layer. By varying the number of hidden layers and the number of simulated neurons within each layer the performance of a ANN can be improved or degraded. Testing the Accuracy of ML-ANN for Harmonic Estimation in Balanced Industrial Distribution Power System Wael M. El-Mamlouk, Metwally A. El-Sharkawy, and Hossam. E. Mostafa F World Academy of Science, Engineering and Technology International Journal of Electrical and Computer Engineering Vol:3, No:9, 2009 1721 International Scholarly and Scientific Research & Innovation 3(9) 2009 ISNI:0000000091950263 Open Science Index, Electrical and Computer Engineering Vol:3, No:9, 2009 publications.waset.org/9322/pdf