基于SVM的AMI环境下用电异常检测研究  

SVM Based Energy Consumption Abnormality Detection in AMI System

作　　者：简富俊[1,2] 曹敏 王磊 孙中伟[2] 张建伟 王洪亮[4] 

机构地区：[1]华北电力大学云南电网公司研究生工作站,昆明650217 [2]华北电力大学电气与电子工程学院,北京102206 [3]云南电力试验研究院(集团)有限公司电力研究院,昆明650217 [4]云南电网公司博士后工作站,昆明650217

出　　处：《电测与仪表》2014年第6期64-69,共6页Electrical Measurement & Instrumentation

摘　　要：高级测量体系的建设在传统电力系统中引入了许多新技术,对电力系统安全提出了新的考验。网络的开放性和安全性之间的矛盾加大,使得非法电力用户窃电的手段增多,如何有效检测窃电成为电网信息化的一个新问题。根据高级测量体系系统架构的特点,使用One-class SVM无监督机器学习架构对电力用户负荷异常进行检测,可以在小样本、样本分类不均衡环境下提高检测的准确性。使用对检测结果过滤的方法对检测结果进行分类处理,降低系统的虚警率。系统能提高用电稽查效率,降低电力系统的非技术性损失。最后对系统进行架构搭建实现,使用真实算例验证了算法的执行效率和检测效率。Electrical power system is facing serious security problems due to Advanced Metering Infrastructure（AMI） system which introduces a lot of new technologies in traditional electrical power system. As a result of smart grid, the contradiction of openness and security is increased which will give rise to the increase of electricity fraud. How to detect electricity fraud has become a new issue of grid informatization. On the basis of the AMI’s architecture, the paper adopts One-class SVM technique to detect the abnormal behavior of electricity users which works at a non-supervision Machine learning mode and can get a high accuracy of detection in small sample or unbalanced classification environment. In order to reduce the false alarm rate of the system, the system uses filtering method to handle the test results of SVM classification processing. System can improve the efficiency of electrical inspection and reduce the Non-Technical Losses（NTL） of power system. The paper also gives an implementation of the system which verifies execution efficiency and detection efficiency of the algorithm by real example.

关 键 词：高级测量体系 用电异常 机器学习 非技术性损失 

分 类 号：TM714.3[电气工程—电力系统及自动化]