风电场两种超短期风速预报方法的效果对比

Comparison of two wind speed prediction methods in ultra short-term at wind farm

针对风电场所需的时效在0～4h，且时间分辨率不低于15min的超短期风速预报。根据测风塔实时发回的实测风速序列，建立了基于BP神经网络的风电场风速时间序列外延预报模型。另一方面，建立MM5模式预报风速与实测风速的误差序列，并利用BP神经网络作误差序列的外延预报，从而利用误差的预报值对MM5风速预报值进行订正，获得新的预报值。综合对两种方法的预报效果指标分析以及拟合曲线的比较结果表明：使用BP神经网络对MM5风速预报值进行修订的方法在总体上效果较优，特别是当影响风电场的天气系统变化明显，近地层风速变率较大时，该方法的预报效果更具有明显的优势。

Based on the observed wind speed sequences from the wind tower, a wind speed-time se- ries epitaxial forecast model according to BP neural network was established for ultra short-term wind speed forecasting with 0-4 h effectiveness and ≥ 15 minutes time resolution. On the other hand, the error sequence of wind speed between MM5 model forecasting and the measurement was found, and the BP neural network was used as the extension of simulation error sequence to correct the MM5 wind speed forecasts, so the new forecasting value was developed. Comparison of the forecasting effect and the fitting curve between the above two methods shows that BP neural network plays an important role in correcting the MM5 wind speed forecasts, especially when weather system changes significantly and the ground wind speed changes very rapidly.