基于多域特征提取的电力数据离群点检测研究

Research on outlier detection of power data based on multi domain feature extraction

为解决传统方法存在随机检测误差导致检测结果不精准的问题,提出基于多域特征提取的电力数据离群点检测研究。该方法从波动性、趋势性和变动性三方面分析6维数据特征,通过多域特征训练对数据进行降维,有效剔除冗余特征。使用K-means算法任意选择对象作为初始聚类中心,根据聚类对象计算各个对象与聚类中心的距离,以此划分离群点区域。计算误差函数,避免离群点分布位置误差影响检测结果,结合网格算法确定6维区间数目和数据分布密度,汇集每一维区间正常点,由此完成离群点检测。实验结果表明,该方法检测到的离群点位置与实际分布位置一致,且只出现2个离群点丢失的情况,其余离群点均能被检测出来,说明应用该方法检测结果精准。

To solve the problem of inaccurate detection results caused by random detection errors in traditional methods,a research on outlier detection in power data based on multi domain feature extraction is proposed.This method analyzes the characteristics of 6-dimensional data from three aspects:volatility,trend,and variability.Through multi domain feature training,the data is dimensionally reduced,effectively eliminating redundant features.The K-means algorithm is used to arbitrarily select objects as the initial clustering center,and the distance between each object and the clustering center is calculated based on the clustering objects,in order to divide the outlier area.The error function is calculated to avoid the influence of the distribution position error of outliers on the detection results.The number of six dimensional intervals and the data distribution density are determined by combining the grid algorithm,and the normal points in each one-dimensional interval are collected to complete outlier detection.The experimental results show that the location of outliers detected by this method is consistent with the actual distribution position,and only two outliers are lost.The remaining outliers can be detected,indicating that the detection results of this method are accurate.