Ultra-Short-Term Wind Power Interval Prediction Based on Fluctuating Process Partitioning and Quantile Regression Forest

Abstract

As errors in point forecasts of wind power are unavoidable, interval forecasts can adequately describe the uncertainty in wind power and thus provide further guidance to dispatchers in their decision making. Current interval prediction methods are still incomplete in terms of tapping into the physical variability of wind power, especially for the specific time scale of the ultra-short term. This paper therefore proposes a new framework for interval forecasting of ultra-short-term wind power that incorporates the power fluctuation process. Firstly, a fluctuating process of wind power series is defined and a Kalman-SOM method for clustering the fluctuating processes of wind power is constructed. Secondly, a quantile regression forest interval prediction model is constructed for multiple fluctuation processes for ultra-short-term time scales. Finally, the effectiveness of the overall framework is validated at a wind farm in Jilin Province, China. Compared with the traditional interval prediction method. The interval bandwidth is reduced by 0.86% on average, and the interval coverage is increased by 1.4% on average. The results demonstrate the effectiveness and feasibility of the method in this paper.