基于改进DBSCAN算法的变压器不良漏抗参数辨识  

Bad Data Identification for Leakage Reactance Parameters of Transformer Based on Improved DBSCAN Algorithm

作　　者：孙小磊[1,2] 郑华[1] 李晖[1,3] 王智冬[3] 李隽[3] 王佳明[3] 

机构地区：[1]华北电力大学电气与电子工程学院,北京市102206 [2]国网江苏省电力公司,江苏省南京市210000 [3]国网北京经济技术研究院输电网规划中心,北京市102209

出　　处：《电力系统自动化》2017年第9期96-101,共6页Automation of Electric Power Systems

摘　　要：PSD-BPA在中国电力系统仿真计算中被广泛应用,但由于其数据格式的特殊性,往往容易出现许多人为原因的数据错误,这给仿真计算结果的准确性与可靠性带来了极大的隐患。首先,在给出变压器不良漏抗参数辨识步骤的基础上,结合PSD-BPA潮流数据中变压器参数数据的特点,提出了考虑特征相似度的具有噪声的基于密度的聚类(DBSCAN)改进算法。其次,基于各类参数向量簇的各属性最大相似系数,计算获得各类参数向量簇的典型特征向量。然后,基于各类的典型特征向量,针对聚类结果中的噪声簇,提出了基于离群系数的可疑不良数据分布模型;在此基础上,结合分布规律,提出了基于可疑度的不良参数判别方法。最后,通过实际算例验证了所述模型与方法的有效性。PSD-BPA is widely used in the simulation calculation of power systems. However, many man made data errors easily appear because of the special data form of PSD-BPA that brings great hidden danger to the accuracy and reliability of simulation results. First of all, on the basis of identification steps in bad data identification for leakage reactance parameters of transformer, an improved density based spatial clustering of application with noise （DBSCAN） algorithm is proposed based on feature similarity by combing the characteristics of parameter data in PSD-BPA power flow data. Secondly, a distributed model based on the outlier coefficient is proposed after extracting typical feature data in each classified data set. On this basis, a distinguishing method based on suspiciousness is proposed by referring to the law of distribution. Last but not least, the effectiveness of the model and method adopted in this paper is verified by actual calculation.

关 键 词：变压器 漏抗 不良参数 相关系数 具有噪声的基于密度的聚类(DBSCAN) 

分 类 号：TM41[电气工程—电器]