Árvores de decisão aplicadas à redução de múltipla estimação de faltas em sistemas de distribuição de média tensão

Abstract

Fault location is not a trivial task in distribution systems, since the radial characteristic of the feeders makes fault locators present multiple estimations (especially those based on impedance). In this sense, there is difficult to accurately locate the fault. Therefore, this work presents a decision tree-based approach to reduce the multiple fault estimation issue. For this purpose, the feeder was divided into sectors, which do not overlap. Next, five types of faults were considered in the simulations, which were performed by using the PSCAD™/EMTDC™software. It is worth mentioning that only one meter was used, allocated in the substation. Thus, the voltage and current signals from this meter were subjected to a feature extraction process that, in turn, were used as inputs to decision trees that could be trained and validated to locate the faulted sector. Based on the results, the proposed approach was able to identify above 96.3% for sectors under single-phase faults, 84.3% for two-phase faults and 73.0% for three-phase faults.