Analysis of Artificial Neural Network Methods with Application in Reactive Power Flow Control

Andressa Oliveira¹; Alessandra Picanco²

¹ Universidade Federal da Bahia; ² Instituto Federal de Educação, Ciência e Tecnologia da Bahia

doi:10.20906/CPS/CILAMCE2015-0068

Resumo

The Power Flow (PF) analysis allows planning, expansion and operation of the electrical system from the study of the permanent network system. In treatment of loads presupposes the availability of appropriate tools, especially when the system is large. The state of the system with violated voltage levels can be improved through the adjustment of voltage control devices on the buses and the optimal planning of reactive sources, considering the physical characteristics and operation. The PF is a nonlinear problem, in order to determine the magnitudes of voltage and angle, as well as the level of losses. So, there is a need for application of numerical techniques for optimizing the control of reactive to minimize losses. The Artificial Neural Networks (ANN) is a technique inspired in the human brain and there is the advantage of a reduced convergence time compared to other methods for the solution of real-time Power Flow. The systems used were IEEE 6-bus and IEEE 30-bus. The analysis consisted of training scenarios that it presents the combination of reactive sources that through the calculation of losses at each scenario, the results are compared with the MatPower© and PowerWorld© software using the Newton-Raphson method to solve the problem. Architectures studied by applying the PF are Backpropagation, Hopfield and Radial Basis Function (RBF). The scenarios were formed through the simultaneous variation of taps and reactive, that is related to each bus. And, therefore, it produces the values of the voltages and angles more precisely in order to reduce losses. The reduction of losses may be obtained by increasing the number of iterations. The convergence time and amount of epochs are variables in the Neural Network efficiency, for training in source code backpropagation architecture in IEEE 6-bus, convergence occurred in 22 times and 1 s. By using the architecture of Hopfield along with backpropagation and the RBF is possible to ensure that losses can be minimized. When compared the PowerWorld© with RBF, there i

Palavras-chave: Artificial Neural Networks; Power Flow; Losses