Proposta de Algoritmos Inteligentes para Localizar Faltas e Monitorar a Qualidade da Energia em Redes de Distribuição de Energia Elétrica

This work presents a methodology for automatic diagnosis and fault location on electric power distribution systems through the combined use of modern techniques of network analysis, signal processing and, intelligent systems. Some algorithms were developed aiming to detect, classify and locate in time and space the power quality disturbances. The continuous process of detecting these disturbances is accomplished through statistical analysis and direct current multilevel signal analysis in wavelet domain, recorded in the substation. The main characteristics, or behavioral indices, of the current and voltage signals are extracted employing the discrete wavelet transform, multiresolution analysis and the concept of signal energy. These indexes correspond to the input vectors of a number of independent Fuzzy ARTMAP neural networks, which aim to estimate the fault type and the power quality events that are present in the sampled signal. The input patterns of these networks are composed by energy coefficients of the first four levels of details and approximations of current and voltage signals and, the output network presents the disorder type code and the phases involved in the problem. The fault location is performed after the classification process, and the proposed technique applies to three-phase radial or weakly meshed feeders with unbalanced loads, lateral extensions triphasic, biphasic or, monophasic and presenting distributed generation. The system diagnosis and fault location proposed in this paper are able to assist the utilities in their daily works, providing important information for the operation and maintenance of power distribution systems, minimizing the repairing time and, consequently, obtaining less financial losses due to rapid fault diagnosis and restoration of the normal system operation. A real life three-phase distribution system with 134 nodes, 13.8 kV and 7.065 MVA was used to test the proposed algorithms, which showed satisfactory results, attesting that the proposed algorithms are efficient, fast and, above all, intelligent.