Presented at Powertech 2015, Eindhoven A Comparative Analysis of Intelligent Classifiers for Passive Islanding Detection in Microgrids Riyasat Azim, Kai Sun, Fangxing Li, Yongli Zhu, Hira Amna Saleem Dept. of Electrical Engineering and Computer Science University of Tennessee Knoxville, TN, USA mazim@vols.utk.edu, kaisun@utk.edu, fli6@utk.edu, yzhu16@vols.utk.edu, hiraamna@hotmail.com Di Shi, Ratnesh Sharma NEC Laboratories America Energy Management Department Cupertino, CA, USA dshi@nec-labs.com ratnesh@nec-labs.com Abstract—This paper proposes a passive islanding detection technique for distributed generations in grid-connected microgrids and presents a comprehensive comparative analysis of intelligent classifiers for passive islanding detection application. The proposed method utilizes pattern recognition techniques in classification of underlying signatures of wide variety of system events on critical system parameters for islanding detection. Case study on a grid-connected microgrid model with different types of distributed generations is performed to evaluate the proposed method and compare the classifier performances. Test results demonstrate the effectiveness of the proposed method in detection of islanding events. Index Terms--Decision trees, islanding detection, microgrids, naïve-Bayes, neural networks, support vector machines. I. INTRODUCTION Integration of distributed generation (DG) resources with electric power systems (EPS) offers potential solution to energy security and reliability with minimum environmental impacts. However, several technical considerations are required in system planning and operation processes for DG integration. Inadvertent islanding is one of the major issues associated with DG integrations. IEEE Std. 1547 defines islanding as “A condition in which a portion of area electric power system (EPS) is energized solely by one or more local EPSs through associated points of common couplings (PCC) while the portion of area EPS is electrically isolated from rest of the area EPS”. IEEE Std. 1547 recommends isolation of DGs within a maximum of 2 seconds in events of island formation [1]. Although islanded operations may be able to enhance reliability by supplying local loads and reducing downtime when supply from area EPS is unavailable, but several operational and safety considerations including power quality standards, voltage and frequency controls, and safety hazards are required before such operations can be realized in practice. The concept of microgrids allows such self-governing system operations. A microgrid is essentially a distribution network consisting of a cluster of DG resources and loads with advanced controls, protections and energy management system to operate in grid-connected mode, islanded (autonomous) mode and ride through between the two modes. Transition from grid-connected mode to islanded mode requires fast and accurate islanding detection as the primary step. Islanding detection techniques are generally divided into three main categories, namely active, passive and communication based techniques. Communication based islanding detection methods mainly use “transfer trip” or “power line signaling” in order to detect islanding conditions. These methods require extensive communication infrastructures and hence expensive. Active techniques rely on perturb and observe methods. Although active methods have smaller non-detection zones (NDZ) compared to passive techniques, they cause degradation of power quality and require complex control for perturbation injections [2-3]. Passive islanding detection methods rely on local measurements of system parameters (such as- voltage and current) and detects islanding events by locating abnormalities in those system parameters. Several passive islanding detection methods have been proposed in literature [4-12]. Passive islanding detection methods do not degrade power quality, but these methods suffer from a larger non-detection zone (NDZ). Especially, in presence of power balance in the island (i.e. generation and load are approximately balanced in the islanded section of the system). This paper investigates passive islanding detection based on pattern recognition techniques and presents a comprehensive comparative analysis of pattern recognition techniques based classifiers for passive islanding detection in distributed generation (DG) from microgrid standpoint. The passive islanding detection method is based on extraction of a unique set of critical system features from voltage and current measurements at target DG locations, and utilization of intelligent classifiers for detection of islanding events. The set of system features were selected to enhance islanding detection accuracy in presence of multiple types of DG units, under different system operating and loading conditions. A detailed case study on grid connected microgrid model implemented with IEEE 13 node distribution feeder system is performed to validate the effectiveness of the proposed islanding detection This work was supported by NEC Laboratories America, Inc.