Two-stage prediction framework for wind power ramps considering probability distribution distance measurement

Abstract

The high volatility in wind power and uncertainty of ramp events has brought significant hidden hazards to maintain the stable and safe operation of the power system. In this paper, a two-stage prediction framework is proposed, taking the measurement of the probability density curve and the judgment of ramp-up mode. Aiming at the issue of limited prediction information and the sensitivity of prediction values at extreme points, an uncertainty prediction model for wind power based on a gated recurrent unit quantile regression network is proposed to realize the screening and judgment of ramps based on the prediction results and the divergence measurement of probability density distribution. Faced with notoriously unpredictable ramps and the small-samples poor learning performance, this paper proposes a ramp pattern discrimination model based on a gradient boosting decision tree, which describes the events in a future period from their magnitude and duration. Compared with state-of-the-art ramp forecasting methods, our proposed framework yields highly outstanding performances and realizes the adaptive detection for wind power ramps.