基于预测信息二维坐标动态划分的风电集群功率超短期预测

Ultra-short Term Wind Power Prediction Based on Two-dimensional Coordinate Dynamic Division of Prediction Information

风电集群的大规模并入电网对功率预测的准确度提出了更高的要求。为能充分利用预测功率信息和数值天气预报(numerical weather prediction,NWP)信息,该文提出一种基于功率变化趋势和风速变化波动的二维坐标的风电场动态分群方法。将4h时间尺度的预测过程分成4个等长时间尺度的循环过程,在每次1h的循环过程中应用平衡迭代规约和聚类(balanced iterative reducing and clustering using hierarchies,BIRCH)对各场站的二维坐标聚类,完成对集群的划分,根据划分结果构建训练集,通过门控循环单元(gate recurrent unit,GRU)模型完成各子集群的功率预测,重复这一过程直至完成4h的超短期功率预测。算例结果表明,所提方法的预测精度相比静态划分提升1.8%,相比统计升尺度提升4.31%,可有效提高风电集群的功率超短期预测准确度。

The large-scale integration of wind power clusters into the power grid puts forward higher requirements for the accuracy of power prediction.In order to make full use of the predicted power information and numerical weather prediction(NWP)information,a wind farm dynamic clustering method based on two-dimensional coordinates of power change trend and wind speed change fluctuation was proposed in this paper.The 4h time scale prediction process was divided into four equal time scale cycle processes.In each 1h cycle process,the balanced iterative reduction and clustering using hierarchies(Birch)was applied to cluster the two-dimensional coordinates of each station to complete the division of clusters.According to the result of partition,the training set was constructed,the power prediction of each sub-cluster was completed by gate recurrent unit(GRU),and the process was repeated until the ultra-short-term power prediction was completed for 4 hours.The example shows that the prediction accuracy of the proposed method is improved by 1.8%compared with static partition and 4.31%compared with statistical lifting scale,which can effectively improve the power prediction accuracy of ultra-short term power prediction of wind power clusters.