Detection of Impending Ramp for Improved Wind Farm Power Forecasting

Abstract

Detection of impending front-induced ramp events is studied as a new class of change detection problem - change detection for multiple time series with spatial dependency. A critical step to ramp event detection is to capture the spatial dependency between neighbor turbines’ power output. To this end, a graphical model is utilized to model the dependency of turbine-level ramp events. Then, change point detection is carried out for the time series of individual turbines’ power output, by using the belief from neighbor turbines in the dependency graph. Once an impending ramp is detected, the magnitude of ramp is then forecasted by using current measurement data. A key observation is that due to the movement of front, the best predictors for individual turbines’ power output vary across three different regions of the wind farm. With this insight, different predictive models are adopted for forecasting power output from each region. Through numerical experiments, the proposed detection-based wind power forecasting method is proven to outperform conventional methods for wind power ramps.