一种改进LSOF的风电异常数据识别与清洗方法

An improved LSOF method for wind power anomaly data recognition and cleaning

针对风速-功率历史运行数据的识别和处理存在识别准确度低、分析过程复杂和异常数据清洗效率低的问题,提出了一种改进无监督学习的聚类局部结构离群因子识别方法(LSOF)。首先,通过最近邻域树法对邻域进行测量,旨在解决传统局部离群因子识别性能低,且对邻域大小敏感的问题;其次,利用改进无监督学习的聚类局部结构离群因子识别方法分别对每个局部结构进行计算评分,并将评分最高的局部结构报告为异常局部结构,在此基础上,利用最近邻域树特征区分异常值和异常值组;最后,通过某实际风电场数据进行验证。研究结果表明,该方法在邻域范围内对异常值识别具有较高的精度和鲁棒性。

Aiming at the problems of low recognition accuracy,complex analysis process and low efficiency of abnormal data cleaning in the recognition and processing of wind speed-power historical operation data,an improved unsupervised learning method for detecting cluster local structure outlier factors is proposed.Firstly,the neighborhood is measured by the nearest neighbor tree method,which aims to solve the problem that the traditional local outlier factor has low recognition performance and is sensitive to the neighborhood size.Secondly,the improved unsupervised learning clustering local structure outlier factor detection method is used to calculate and score each local structure respectively,and the local structure with the highest score is reported as the abnormal local structure.On this basis,the nearest neighbor domain tree feature is used to distinguish the abnormal value and the abnormal value group.Finally,the proposed method is verified by the data of a real wind farm.The results show that the method has high accuracy and robustness for outlier recognition in the neighborhood.