Improved Gaussian based rapid quantification of scheduling uncertainty considering source-load extreme scenario enhancement

Abstract

Extreme events are typically low-probability occurrences with limited historical data and a high degree of unpredictability. The inherent conflict between the dispatch rationality in extreme and conventional scenarios, makes it hard for traditional methods to consider both performances. This article introduces a rapid quantification method for evaluating dispatch uncertainty in extreme scenarios within integrated energy systems. The method enhances the speed and precision of energy dispatch predictions by establishing a direct correlation between meteorological data and energy dispatch. The process begins with the collection of extreme scenarios sets. The Maximal Information Coefficient (MIC) is then employed to identify distinctive meteorological characteristics across different sets of extreme scenarios. To compensate for the lack of historical data in these scenarios, the Synthetic Minority Over-Sampling Technique (SMOTE) is utilized to augment the scenario dataset. Subsequently, the outcomes of the integrated energy system (IES) are calculated as output. Finally, Gaussian Process Quantile Regression (GPR-Q) is applied to predict dispatch uncertainty in these extreme scenarios. After comparing with existing approaches, this method can innovatively avoid the prediction error of new energy to a certain extent and quickly provide the interval probability distribution of scheduling predictions with richer information. Such results better align with the needs of real dispatch scenarios.