A Measurement-Based Robust Non-Gaussian Process Emulator Applied to Data-Driven Stochastic Power Flow

In this paper, we propose a robust non-Gaussian process emulator based on the Schweppe-type generalized maxi-mum likelihood estimator, which is trained on metered time series of voltage phasors and power injections to perform stochastic power flow. Power system data are often corrupted with outliers caused by fault conditions, power outages, and extreme weather, to name a few. The proposed emulator bounds the influence of the outliers using weights calculated based on projection statistics, which are robust distances of the data points associated with the rows vectors of the factor space. Specifically, the developed estimator is robust to vertical outliers and bad leverage points while retaining good leverage points in the measurements of the training dataset. The proposed method is demonstrated on an unbalanced radial IEEE 33-Bus system heavily integrated with renewable energy sources.