基于HS-Clustering的风电场机组分组功率预测

摘要

为了寻求风电场功率预测精度和计算效率二者的平衡,提出了一种基于霍普金斯统计量与聚类算法(HS-Clustering)的风电场机组分组功率预测方法,该方法将霍普金斯统计量与聚类算法的优势有效结合,采用霍普金斯统计量确定场内机组分组个数,通过聚类算法识别不同机组的相似性将风电场分成不同的机组群,然后对每组机群分别建立功率预测模型,从而叠加得到整场输出功率;另外以实测风速、实测功率及二者组合作为机组分组模型输入,分析其对预测精度的影响程度.实例分析表明基于HS-Clustering的分组预测方法可以显著提高预测精度,同时保证较高的计算效率;风速是影响分组效果的主要因素,对于某些分组模型,功率又可以作为风速的重要补充.%In order to balance the forecast accuracy and computational efficiency, a wind power forecasting method for clustering wind turbines is proposed based on effective combination of Hopkins statistics (HS) and clustering methods, in which Hopkins Statistics is used to determine the clustering number of a wind farm, and wind turbines in a wind farm are clustered into several groups according to the identifying of similar characteristics by clustering method.Then power forecasting model of each clustering group is built separately, whose power output is added to obtain whole power output of the wind farm.In addition, the real-time monitoring wind speed, power output and their combination are taken as the inputs for clustered group model, and their influences on the accuracy of clustering forecast model are analyzed.The case analysis shows that the HS-Clustering based forecasting method can effectively forecast the output power of the whole wind farm with better accuracy and higher computational efficiency, wind speed is the main factor affecting clustering results, and wind power can be regarded as an important additional factor as to certain group models.