Data-Driven Islanding Detection Using a Principal Subspace of Voltage Angle Differences

The likelihood of an unintentional power system islanding is increased in systems with significant penetration of distributed generation. To mitigate the adverse effects of islanding, a quick and reliable islanding detection method is needed. This paper first analyzes covariance matrices of a linearized power system model, and relates them to the principal component analysis of experimentally obtained covariance matrices. Additionally, a new model-independent islanding detection method is proposed that uses measurements of voltage angle differences between multiple locations in the system. The angle differences are first preprocessed to remove the effects of nonstationarity. Thereafter, a probabilistic model of principal component analysis is trained using the acquired measurements. The principal and residual spaces extracted from the measurements are used to discriminate between islanding and other events in the system. The applicability of the proposed method is demonstrated by using real measurements gathered from several locations in a transmission grid.