Topology Error Detection for Defensive Islanding of Power Grids

Resilience improvement through defensive islanding requires multiple simultaneous topological changes after an event to exclude damaged components and transform the grid into power balanced islands. Topology error identification is challenging under such conditions given the limited availability of measurements and uncertainty of events. This paper proposes a new approach for machine learning-aided topology error detection and identification for defensive islanding. The approach combines conventional data error detection with a data driven model to distinguish error types and identify the erroneous component. The topology check is done on two different levels. In the lower level (decoupled islands), a state estimation problem is solved for error detection in each of the smaller grids. Solving decoupled small systems requires less time and computation power, which is essential for the time sensitive post-disaster task at hand. The identification step, however, is at the high level (connected grid) due to the possibility of the topology error being in the links between the islands. Approach implementation is possible through numerous State Estimation (SE) and topology validation approaches. In this work, a DC SE is performed since the concern here is the connectivity rather than accurate power flow calculations. A neural network model is used to validate topology considering data and system uncertainties. A case study utilizing the proposed method is performed on different topological variations of the IEEE 118-Bus system. Scarcity and uncertainty of data are also considered. The results show promising potential of the proposed algorithm.