Avaliação de estabilidade de tensão através das técnicas de inteligência artificial.

Abstract

In this Thesis the results obtained in the evaluation of the security of power systems in relation to the voltage instability using the neural network RPROP and the neuro-fuzzy system ANFIS (Adaptive Neuro-Fuzzy Inference System), combined in a hybrid arrangement are presented. This hybrid arrangement has its performance compared with the results presented by the neural network Multilayer Perceptron (MLP), RPROP, kNN (k - Nearest Neighbour) and ANFIS. It is also presented the results using a combination of neural networks made by two classifiers (RBF-DDA and kNN). These classifiers were combined using a method denominated Modified Weighted Maxim Sum (MWMS). These results are compared with the results presented by the neural networks MLP and RBF-DDA acting alone, and for the kNN. The classifiers were trained to give as exit, of a group of input variables (operation point of electric system), the condition of security of the system among one of two: secure or alert. These conditions had been obtained by the Margin Index calculated through the Continuation Method. It is important to standout the right formation of the sets of training and test data that had been presented to the Classifiers, and also the importance of the strategy of the division of the power system in groups of bars (areas) to increase the performance of the security assessment. The efficiency of the approach was evaluated through the results gotten with two real systems: the South-Southeast Brazilian system of 77 bus; and North-Northeast system of 482 bus.