Probabilistic Wind Power Forecasts Considering Different NWP Models
Shenghong Wu, Yuan‐Kang Wu
Abstract
The intermittency of wind power generation is regarded as an obstacle to use wind energy for electricity. Accurate wind power forecasting is one of the direct methods to reduce the risk about wind intermittent. Compared with deterministic forecasting, probabilistic forecasting provides additional information about uncertainty, allowing power system operators to minimize operating costs for unit scheduling and power transactions. Numerical weather prediction (NWP) plays an important role on probabilistic wind power forecasting. Thus, this paper uses three different NWP models to generate the NWP wind speeds, and these NWP wind speeds and historical wind power measurements are used as the inputs for wind power generation. The used NWP models include decisive forecast (WRFD), ensemble forecast (WEPS), and real-time forecast (RWRF). These NWPs were obtained from the Taiwan Central Meteorological Bureau (CWB). Additionally, in this work, several deep learning models and traditional artificial neural networks were applied for wind power forecasting, in which the distribution of forecasting errors is used to construct a reliable prediction interval, and the lower limit upper limit estimation (LUBE) method is used. Based on the forecasting results, the use of NWP models has significantly improved the forecasting performance.