基于aFCM-KNN的风电功率缺失值填补  

Wind Power Data Imputation Based on aFCM-KNN

作　　者：李一凡 黄景涛 关海平 LI Yi-fan;HUANG Jing-tao;GUAN Hai-ping(College of Electrical Engineering,Henan University of Science and Technology,Luoyang 471023,China;Tongliao SPIC Power Generation Corporation Limited,Tongliao 028001,China)

机构地区：[1]河南科技大学电气工程学院,河南洛阳471023 [2]通辽电投发电有限责任公司,内蒙古自治区通辽028001

出　　处：《计算机仿真》2024年第8期52-57,共6页Computer Simulation

基　　金：国家自然科学基金(U1504617);教育部产学合作协同育人项目(202101222007)。

摘　　要：风电实时运行数据在采集、传输和存储过程中的缺失值问题,给基于运行数据的风电功率预测等应用带来困难。针对以上问题,提出一种基于自适应模糊聚类的近邻填补算法aFCM-KNN。鉴于风电数据自身具有的强随机性和波动性,基于FCM算法根据风速对风电数据进行工况聚类,为解决FCM需人为设定聚类个数受主观影响较大的问题,依据风电数据分布特征设计了一个自适应确定聚类个数的策略;考虑到聚类后直接填补容易受噪声的影响,基于KNN算法根据缺失值所在样本的近邻点对每个子簇内的缺失值进行填补,进一步提高了填补精度。在实际数据上的测试分析表明,与其它六种常用填补算法相比,该方法的填补准确率更高。The problem of missing values of wind power during the collection,transmission and storage brings difficulties to further applications such as wind power prediction based on operating data.Aiming at this problem,a near-neighbor imputation algorithm based on adaptive fuzzy clustering is proposed,named aFCM-KNN.In view of the strong randomness and volatility of the wind power data itself,the wind power data is clustered based on the FCM algorithm according to the wind speed.In order to solve the problem that the number of clusters needs to be manually set by FCM,an adaptive strategy for determining the number of clusters is designed according to the distribution characteristics of wind power data.Considering that the direct imputation after clustering is susceptible to noise,the KNN algorithm is used to impute the missing values in each subcluster according to the neighbor of the sample,which further improves the imputation accuracy.Test results on actual data show that the method has a higher accuracy than the other six commonly used imputation algorithms.

关 键 词：风电功率 缺失值填补 模糊均值聚类 近邻算法 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]