1 Fault Classification in Power Distribution Systems using PMU Data and Machine Learning Fl´ avio Lori Grando, Andr´ e Eugenio Lazzaretti, Miguel Moreto, and Heitor Silv´ erio Lopes Technological University of Paran´ a (UTFPR) – Curitiba – PR – Brazil flaviogrando@alunos.utfpr.edu.br, lazzaretti@utfpr.edu.br, miguel.moreto@ufsc.br, hslopes@utfpr.edu.br Abstract—This work presents the analysis of machine learning methods for fault (short-circuit) classification in electrical distri- bution networks using data from PMUs (Phasor Measurement Units) installed along the network. The Alternative Transient Program (ATP) was used to simulate 26,928 different instances distributed into 33 types of faults – single and multi-phase, including or not the ground and different wire breakages – and one normal condition of the system. The IEEE 123-bus distribution system was used as the test system. We compared five machine learning methods for classification: Linear Discriminant Analysis (LDA), k-Nearest Neighbors (kNN), Support Vector Machines (SVM), Artificial Neural Networks (ANN), and Decision Trees (DTs). The best result was achieved by the SVM with Gaussian kernel and ANN. The input data (feature extraction) was also varied, testing data from one or several PMUs, ABC se- quence phasors and symmetrical sequence phasors. We obtained slightly better results for symmetrical components and multiple PMUs in the network. Finally, classes of the same short-circuit with different wire breakages were grouped, raising the overall classification accuracy. Overall conclusion is that the proposed approach is feasible for fault classification using PMU-data in a distribution network. Keywords—Distribution Systems, Fault Classification, Machine Learning, Phasors, PMU. I. I NTRODUCTION The synchronized phasor measurement technology has opened a new paradigm in the observability of electrical systems, allowing to trace in real-time the dynamics of the system through synchronized data with high precision and resolution [1]. The synchronization of the measurements is obtained through Global Positioning System (GPS). Thus, Phasor Measurement Units (PMUs) extract measurements of synchronous phasors (synchrophasors) and frequency of si- nusoidal signals at different points along an electric power system. Then, the information is sent out to a Phasor Data Center (PDC), as shown in Fig. 1. Conventional synchronized phasor measurement systems are based on PMUs that extract measurements of potential and current transformers at power substations. Because they are based on measurements at substations, they are unable to record local dynamics along the distribution system [2]. On the other hand, previous works suggest several applications for synchrophasorial measurement at distribution level [3], [4], [5]. Due to the observability that the PMU can provide and the large amount of information generated, the synchrophasor technology has been largely applied with machine learning- PMU PMU PMU PMU PDC GPS Satellites Figure 1. Representation of a Synchronized Phasor Measurement System. based approaches to detect, classify, and locate events [6], [7], [8]. Several common events of a distribution system can be observed by PMUs, such as the load and equipment switching- transients, as well as short-circuits (faults) [9]. These two group of events are the major interest, since they usually result in blackouts and high costs for power utilities. In this sense, several works have been conducted, mainly motivated by the expansion of the system and the growth of smart grids [10], [11]. In [12], for instance, a PMU data-driven framework was pro- posed to distinguish a malfunctioned capacitor bank switching and a malfunctioned regulator on-load tap changer switching from two normal operating events, using the IEEE 123-bus. For different noise levels and number of PMUs, the authors showed the feasibility of using PMU data to satisfactory classify those events. A similar data-driven approach was proposed in [13]. Au- thors classified power quality events using real-data collected during 15 days from two micro-PMUs installed on a real distri- bution feeder. The power quality events include the detection of internal phase imbalance in a 900 kVAR capacitor bank as well as a potential malfunction in its Volt/VAR controller. On the other hand, [14] presented the application of a micro-phasor measurement unit for power distribution network monitoring. Particularly, the authors discuss the detection of abnormal events, that is, transients in voltage and current waveforms that may be caused by faults, topology changes, load behavior, and source dynamics, without however, discriminate among types of faults. However, in the particular case of faults along the dis-