Fault location in power distribution systems using a learning approach based on decision trees

This paper presents a learning-based strategy that uses decision trees for locating faults in radial power systems, which is aimed to improve the power quality as demanded by the deregulated electrical markets. The proposed method first subdivides the power system into several regions, and then, a classification technique based on decision trees is trained using a fault database. The obtained decision trees are used to assign a faulted zone to a new fault event, which reduces the restoration time and as a consequence helps to maintain good quality indices. The proposed method is validated in the IEEE 34-node test feeder considering several operating conditions, such as variations in load, substation voltage, line length and fault resistances. The obtained results prove the good performance of the proposed fault location method. Finally, the implementation of this method on real power distribution systems helps to maintain good power continuity indices, with a low investment.