A Quantile Regression-Based Approach for Online Probabilistic Prediction of Unstable Groups of Coherent Generators in Power Systems

This paper proposes a novel framework for probabilistic data-driven prediction of unstable groups of coherent generators in interconnected power systems. In contrast to existing techniques in which deterministic classification or forecasting approaches are applied to an offline database, the current study relies on a prediction interval (PI)-based method to tackle prediction uncertainties. First, similarity coefficients (SCs) are considered as internal outputs and calculated for all offline cases. Then, at some generator terminals as selected via a feature selection process, voltage values are measured and used as the input features of the prediction tool. Quantile regression forest is conducted to generate PIs, in which several intervals with certain probabilities are predicted for SCs between any pair of generators. Thereafter, the obtained PIs are used to shape an empirical cumulative distribution function of SCs; a Monte Carlo simulation is then conducted to find a reliable estimate of possible grouping patterns. Finally, a decision-making phase is employed to draw clear distinctions among various parts of the most plausible grouping pattern with respect to a reliability index. This approach can offer power system operators wider flexibility to select a corrective control strategy. The effective performance of the developed approach is demonstrated on several IEEE test systems, followed by a discussion of results.