Adaptive islanding detection and diagnosis using wide area monitoring

Abstract

This paper proposes a method for the detection and diagnosis of power system islanding events using phase-angle monitoring. The method is based on multiblock principal component analysis (MB-PCA), an extension of PCA that allows the underlying relationship between data sets to be identified. MB-PCA divides the large-scale system data into several blocks thus enhancing the model's ability to explain and diagnose power system events. Also incorporated into the method is a moving window technique. This allows the method to update over time enhancing power system situational awareness, by using data that best represents the current condition of the time-varying power system. Using Hotelling's T2 and Q statistics allows the method to detect specific power system events with the main emphasis placed on the detection of power system islands. Further diagnostic approaches, such as contribution plots, are used to locate and isolate the detected events. The reliability of the proposed method is demonstrated using simulated case studies showing the method's ability to detect a genuine islanding event and its avoidance of nuisance tripping.