高斯混合模型风电机组异常数据识别方法研究  

Abnormal data identification for wind turbine based on Gaussian Mixture Model

作　　者：肖碧涛[1] 赖晓路[1] 郭鹏[2] 王桂松 Xiao Bitao;Lai Xiaolu;Guo Peng;Wang Guisong(Guodian Nanjing Automation Co.,Ltd.,Nanjing 211800,China;School of Control and Computer Engineering,North China Electric Power University,Beijing 102206,China)

机构地区：[1]国电南京自动化股份有限公司,江苏南京211800 [2]华北电力大学控制与计算机工程学院,北京102206

出　　处：《可再生能源》2023年第8期1051-1056,共6页Renewable Energy Resources

基　　金：国家自然科学基金面上项目(51677067)。

摘　　要：风电机组原始运行数据中存在大量异常数据,对异常数据进行识别和剔除是后续准确预测风电机组出力及评价发电性能的基础。文章分析了风电机组运行数据功率散点在风速-功率坐标系中的分布特征,提出了基于贝叶斯信息准则(BIC)的高斯混合模型异常数据识别方法。沿水平功率方向以一定间隔划分多个水平功率区间,采用高斯混合模型对落在每一水平功率区间内的功率散点进行聚类,并引入BIC准则自适应确定模型的高斯分量个数。结合功率散点分布特征先验经验,对每一水平功率区间内的多个高斯分量置信椭圆及其聚类功率散点进行异常标识。以风电机组实际运行数据为例,验证了高斯混合模型异常数据识别方法的有效性。There are a lot of abnormal data in the wind turbine SCADA data.Accurately identifying and removal these abnormal data are the basis for wind turbine power prediction and performance analysis.This paper detailed analyzes the scattering characteristics for the wind turbine data in wind speed-power coordinate and proposes an abnormal data identification method based on Bayesian Information Criterion(BIC)and Gaussian Mixture Model(GMM).The power dots(SCADA data)are allocated into different bins in the horizontally power direction.Gaussian Mixture Model is used to cluster the power dots in each horizontally power bin.BIC Criterion is introduced to determine the optimal number of Gaussian component.Based on the prior knowledge of the scattering characteristics of the SCADA data,the abnormal Gaussian component confidence ellipses and their clustering power dots are accurately identified.With the real wind turbine SCADA data,the method proposed in this paper proved to be effective.

关 键 词：风电机组 异常数据识别 高斯混合模型 BIC准则 

分 类 号：TK81[动力工程及工程热物理—流体机械及工程]