Incorporation of data-mining in protection technology for high impedance fault detection

Modernizing the power distribution system implies improving the reliability and performance of protection devices. By incorporating data-mining in the process of designing protection functions, the limits of performance are extended. We propose a method that uses data-mining, able to detect high impedance faults (HIFs) in multi-grounded distribution networks when conventional devices are insufficient. HIFs are produced when overhead lines contact a quasi-isolated surface, such as a tree or the ground. The fault current can be lower than the residual current under normal conditions; hence overcurrent devices do not detect this fault. We describe a set of indicators that characterize HIFs and that can be used in data-mining to distinguish fault situations from other situations. The result is a HIF detection function whose development is based on pattern recognition analysis. The presented methodology can be applied to other fault detection problems to achieve more reliable protection devices.