基于神经网络的风电场超短期风速数值预报的动态修订

Dynamic Modification of Super Short Term Numerical Wind Forecast Based on Neural Networks at Wind Farm

针对风电场风功率预测所需的离地70m、0～4h的超短期风速预报。本文利用中央气象台发布的MM5格点输出的数值预报风速及测风塔实时发回的气象资料，探讨了利用神经网络将前期误差观测值和测风塔湍流指标等因子对MM5数值预报风速进行动态修订的方法。建立动态修订超短期预报模型，为满足风电场超短期风功率预报的工程应用提供一定的参考。结果表明，修订后的预报风速平均绝对误差等指标大幅降低。有效地提高了预报精度。

For effective planning and scheduling and for Wind Power Prediction （WPP） at 70 meters above the ground and 0-4h super short term wind speed forecasting, this paper uses the NWP wind speed of MM5 grids from the National Meteorological Center to analyze the prediction error at the wind tower height in a wind farm which is located off the coast. Based on the meteorological data from the wind tower and after data statistical analysis, it is found that the numerical forecast wind speed errors have correlations with themselves and the prediction errors are caused by the elements of sustainability. A method using earlier observation errors and turbulent index to revise the wind speed forecasting of MM5 is discussed and an ANN dynamic modification model for super short term forecasting is set up. The results show that after correction of the forecast wind speed, the mean absolute error is reduced and the prediction accuracy is improved effectively. It is also shown that the error index decreases about 40%, and the prediction curve can better reflect the high frequency of wind speed fluctuations, which better agrees with the measured wind speed curve. Update can be done once every four hours, satisfying the requirements of power grid dispatching. The method is simple and economic and can be used widely in small and medium-sized wind farms. It will help effective use of wind power as well as safe operation of power companies.