Machine Learning Approach to Islanding Detection for Inverter-Based Distributed Generation
MetadataShow full item record
Despite a number of economic and environmental benefits that integration of renewable distributed generation (DG) into the distribution grid brings, there are many technical challenges that arise as well. One of the most important issues concerning DG integration is unintentional islanding. Islanding occurs when DG continues to energize portion of the system while being disconnected from the main grid. Since the island is unregulated, its behavior is unpredictable and voltage, frequency and other power system parameters may have unacceptable levels, which may cause hazardous effect on devices and public. According to the IEEE Standard 1547 DG shall detect any possible islanding conditions and cease to energize the area within 2 sec. In this dissertation work, a new islanding detection method for single phase inverter-based distributed generation is presented. In the first stage of the proposed method, parametric technique called Autoregressive (AR) signal modeling is utilized to extract signal features from voltage and current signals at the Point of Common Coupling (PCC) with the grid. In the second stage, advanced machine learning technique based on Support Vector Machine (SVM) which takes calculated features as inputs is utilized to predict islanding state. The extensive study is performed on the IEEE 13 bus system and feature vectors corresponding to various islanding and non-islanding conditions, such as external grid faults and power system components switching, are used for SVM classifier training and testing. Simulation results show that proposed method is robust to external grid transients and able to accurately discriminate islanding conditions 50ms after the event begins.
Subjectautoregressive signal modeling
inverter-based distributed generation
support vector machine
Matic Cuka, Biljana (2014). Machine Learning Approach to Islanding Detection for Inverter-Based Distributed Generation. Doctoral dissertation, Texas A & M University. Available electronically from