Very Short-Term Wind Power Forecasting: State-of-the-Art.

摘要

Because of the high variability of the wind resource and the nonlinear relation between wind speed and power, wind power forecasting on the very short-term horizon is a complex task that is subject to the stochastic nature of the wind speed. At the same time, accurate forecasts are important for wind plant management and power system operations. This report presents a comprehensive state-of the-art review of very short-term wind power forecasting with a focus on forecasting horizons up to 6 hours ahead. The report describes both (1) numerical weather prediction (NWP)/physical, and (2) statistical/artificial intelligence (AI) forecasting techniques and models. We find that hybrid methods have shown to deliver better wind power predictions than persistence and most individual methods at the very short-term timescale. In fact, the hybrid approaches benefit both from the high level of accuracy in the physical models and of the computational learning capabilities of statistical/AI models. Therefore, the combination of both into a hybrid model is often the best approach. At the same time, the best forecasting methodology will also depend on the characteristics of the specific location as well as the intended use of the forecast. The development of NWP/physical models with higher accuracy and resolution, as well as new and more sophisticated statistical/artificial intelligence methods, can both contribute to improve the quality of very short-term forecasts. In addition, at these timescales, particular attention should also be addressed to improve wind power and wind speed ramp forecasting algorithms.