大型风力机异常功率数据清洗方法

Abnormal power data cleaning method of the wind turbine based on improved DBSCAN

针对风力机异常功率数据难以有效清洗的问题,提出改进的DBSCAN方法.首先将数据集离散分割,然后在各离散区间内自适应粗估DBSCNA算法参数并聚类,最后计算统计特征相似性修正聚类结果.以某风场2.5 MW风力机SCADA系统实测数据验证所提方法,结果表明:改进方法的召回率为97.97%,准确率为97.97%,F_(1)值可达97.85%,可有效清洗风力机功率数据集,且变更数据集时改进方法结果更稳定.

The anomalous power data of wind turbine are difficult to clean effectively,an improved DBSCAN method was proposed.Firstly,the data set was segmented discretely.Then,the parameters of DBSCNA algorithm were estimated adaptively and clustered in each discrete interval.Ultimately,the statistical feature similarity was calculated to correct the clustering results.The proposed method is validated with the measured data of the wind turbine SCADA system in a wind field of 2.5 MW.The findings indicate that the recall rate,accuracy rate and F1 value of the improved method are 97.97%,97.97%and 97.85%,respectively.The improved method can effectively clean the wind turbine power data set,and the improved method is more stable when the data set is changed.