Vol.:(0123456789) 1 3 Electrical Engineering https://doi.org/10.1007/s00202-020-01074-8 ORIGINAL PAPER A random forest‑based approach for fault location detection in distribution systems Hatice Okumus 1  · Fatih M. Nuroglu 1 Received: 8 January 2020 / Accepted: 22 July 2020 © Springer-Verlag GmbH Germany, part of Springer Nature 2020 Abstract Finding the fault location in distribution network is difcult in comparison to transmission network based upon high branching and high impedances composed of the contact with environmental factors when a fault occurs. For this matter, the possible faulty section is identifed and then the fault location is determined with the proposed algorithms in this study. Wavelet transform is applied to the current–voltage data along with feature extraction methods, and then, Random Forest is used for identifying the faulty section and estimating the fault location. To approve the validation, the algorithms are applied to IEEE-34 node test feeder. Single-phase to ground faults regarding diferent impedance values at diferent points of the feeder have been efectively detected by using single measurement point data from the sending end of the feeder. Keywords Distribution networks · Fault location · Fault section identifcation · Wavelet transform · Random forest 1 Introduction It is essential for electrical energy, which has become indis- pensable in every aspect of our lives, to be supplied to the customer in a quality manner. The quality of electrical energy depends on its continuity and the reliability of the system. This can be slightly possible by fnding the fault location in the network, making quick restoration and reduc- ing the downtime. In the literature, various studies have been found using diferent methods for the detection of fault location in the transmission network. However, in distribution network, the branched structure and environmental factors (weather conditions, animals, etc.) make it difcult to detect the fault location. This has led researchers to focus on the problem of locating faults in the distribution network in recent years. Methods of fault location in distribution network can be summarized in three main topics as impedance-based, trav- elling wave-based and knowledge-based. Impedance-based methods fnd the distance of the fault by calculating the impedance during the fault using current and voltage data. Many studies have been carried out using both one and two end measurements [1–6]. Although it is a low-cost method, it is highly afected by load currents, fault impedances and the topology of the distribution network. In travelling wave- based methods, fault distance is found based on the veloc- ity of the travelling wave, the time for the wave refected from fault point to reach the measurement point and the time for travelling wave to reach measuring point. In literature, many researchers have studied this method [7–11] but this method needs high costing devices and is highly afected by noise. Knowledge-based methods are the least cost-efective compared to other methods. Fuzzy logic, artifcial neural networks (ANN), support vector machines (SVM), pattern recognition methods can be used under this title. Gururajapathy et al. [12] used voltage sag data obtained from a single measuring point and classifed fault types with SVM. They also tried to fnd the possible faulty sections with support vector regression (SVR) by estimating the fault impedances using voltage sags. Using these imped- ance and voltage sags, they calculated the fault location. The method was applied to a 40-node distribution feeder modelled in PSCAD, and a maximum error rate of 35% was found. Pessoa et al. tried to obtain the faulty zone along with the fault location using the efective value, amplitude value and energy-coefcient values of 3-phase current–volt- age data. The algorithm was tested on the IEEE-34 node test feeder modelled in the ATP, and data were recorded from PQ meters located at 3 points on the model. The faulty zone was * Hatice Okumus haticeokumus@ktu.edu.tr 1 Department of Electrical and Electronics Engineering, Karadeniz Technical University, Trabzon, Turkey