传统时间序列预测多步风速时不能预测突变风速，使风电功率预测误差较大。采用基于数值天气预报（numerical weather prediction，NWP）风速及历史风速修正的卡尔曼滤波法对NWP风速进行多步修正，并通过修正后的NWP风速进行多步功率预测，第16步风速平均绝对误差降低了0.47m/s，将该修正NWP风速与支持向量回归相结合，构建风电功率预测模型。本文所构建的模型与ARIMA模型及NWP直接预测模型相比，误差分别降低了6.8%和8.4%。应用该模型对山东某地区风电场现场数据进行仿真测试，第16步预测准确率达到82.6%。

Traditional time series prediction of multi-step wind speed cannot predict abrupt wind speed, which makes the prediction error of wind power larger. The Kalman filter method based on numerical weather prediction(NWP) wind speed and historical wind speed correction is used to modify the NWP wind speed, which can be performed in the multi-step prediction The average absolute error of wind speed is reduced by 0.47m/s. Then the revised NWP wind speed and support vector regression are combined to construct the wind power prediction model. Compared with ARIMA model and NWP direct prediction, the error is reduced by 6.8% and 8.4% respectively. The model is applied to simulate the field data of a wind farm in Shandong Province, and the accuracy of 16th-step prediction reaches 82.6%.