Anomaly Detection in Data-Driven Coherency Identification Using Cumulant Tensor

As a model reduction tool, coherency identification has been extensively investigated by power researchers using various model-driven and data-driven approaches. Model-driven approaches typically lose their accuracy due to linear assumptions and parameter uncertainties, while data-driven approaches inevitably suffer frombad data issues. To overcome these weaknesses, we propose a data-driven cumulant tensor-based approach that can identify coherent generators and detect anomalies simultaneously. More specifically, it converts the angular velocities of generators into a fourth-order cumulant tensor that can be decomposed to reflect the coherent generators. Also, using co-kurtosis in the cumulant tensor, anomalies can be detected as well. The simulations reveal its excellent performance.