基于LOF-GMM方法的电网异常数据动态辨识及分析  

Dynamic Identification and Analysis of Abnormal Data in Power Grid Based on LOF-GMM Method

作　　者：张懿操 陆俊 洪德华 吴禹 Zhang Yicao;Lu Jun;Hong Dehua;Wu Yu(Information and Communication Branch,State Grid Anhui Electric Power Co.,Ltd.,Hefei Anhui 230061,China;Information System Integration Branch,State Grid Institute of Electrical Science and Technology Co.,Ltd.,Nanjing Jiangsu 211106,China)

机构地区：[1]国网安徽省电力有限公司信息通信分公司,安徽合肥230061 [2]国网电科院有限公司信息系统集成分公司,江苏南京211106

出　　处：《电气自动化》2024年第4期66-68,共3页Electrical Automation

摘　　要：为了进一步提高电网异常数据动态辨识精度,结合高斯混合模型,并利用邻域局部异常因子来确定此点是否属于异常数据,设计了一种基于邻域局部异常因子-高斯混合模型聚类方法的电网异常数据动态辨识方法。研究结果表明:设计的邻域局部异常因子-高斯混合模型聚类算法满足了配电网大数据一体化动态清洗过程需要,获得更高精度的负荷预测结果,有助于大幅增强配电网的响应能力。设计的方法实现了缺失数据填补精度与速度平衡,具有较好工程应用价值。In order to further improve the dynamic identification accuracy of power grid abnormal data,a dynamic identification method of power grid abnormal data based on the clustering method of neighborhood local abnormal factors-Gaussian mixture model was designed by combining the Gaussian mixture model and using the neighborhood local abnormal factors to determine whether the point belongs to abnormal data.The research results show that the designed neighborhood local anomaly factor-Gaussian mixture model clustering algorithm effectively meets the needs of the dynamic cleaning process of distribution network big data integration,obtains more accurate load forecasting results,and helps to significantly enhance the responsiveness of the distribution network.The designed method achieves a balance between accuracy and speed in filling missing data,and has good engineering application value.

关 键 词：配电网 数据清洗 异常数据辨识 高斯混合模型 随机森林 

分 类 号：TM16[电气工程—电工理论与新技术]