基于数据标签的智能电网监控与异常检测

Outlier Data Detection and Monitoring of Smart Electric Grid Based on Data-tag

智能电网是电网的智能化系统,是以输电网、各级电网协调发展为基础的通信信息支撑平台,是包括输变电、配电与电力调度的各电压等级的信息化、自动化和互动化等为特征的高度一体化系统。智能电网在电网监控的数据传输协议、计算与处理效率、各种信息与网络攻击和数据异常检测等方面仍存在不足。因此,本文利用物联网和数据标签技术,提出一种基于数据标签的智能电网监控架构和异常数据检测算法。首先,针对智能电网数据标签监控与异常检测的框架,设计了智能电网的监控数据标签与异常检测框架、数据标签化方法和监控大数据任务划分方法;然后,对异常检测流程、稀疏化与精简算法和检测算法进行设计,提出一种基于数据标签的数据精简和异常检测算法;最后,对实验和仿真进行设置,对时序数据维度数、异常数据量的算法准确率和召回率与参比算法进行了仿真与对比实验,并对不同测试数据量的本文算法与参比算法运行时间进行了对比实验。结果表明:本文设计的数据标签智能电网监控与异常数据检测算法与参比算法相比,当时序数据维度数递增时,其异常数据检测的准确率大于80%,召回率高于82%;当异常数据量增加时,本文算法的异常数据检测准确率和召回率较优越;比较不同测试数据量的运行时间发现,本文算法比参比算法的运行时间少2.0~3.0 s。

Smart grid,the intelligent system for the power grid,can be defined as a communication and information support platform based on the coordinated development of transmission networks and distributed power grid.It is also a highly integrated system characterized by informatization,automation,and interaction of all voltage levels,including power transmission,transformation,distribution,and dispatch.However,the smart grid is facing the following technical challenges,including the data transmission protocol for power grid monitoring,computing/processing efficiency,as well as detections for information,network attacks,and data anomalies.To this end,a novel data tag-based intelligent power grid monitoring architecture and anomaly detection algorithm is proposed using the Internet of Things and data-tagging technologies.The monitoring framework and anomaly detection are firstly designed based on data tagging for the intelligent power grid,including data labeling method,and task categories for monitoring-related big data.Subsequently,the anomaly detection paradigm is designed to support the monitoring framework,in which the sparsification/simplification and detection algorithms are proposed based on data labels.To validate the proposed framework,the experimental and simulation configurations are provided to conduct the result comparison over selective baselines,in terms of the accuracy and recall under different sequential data dimensions,anomaly data amounts,as well as running time for different test data amounts.Experimental results demonstrate that,compared to the baselines,the proposed approach achieves over 80%accuracy and 82%recall with the increase of the sequential data dimensions.In addition,a similar performance improvement in terms of accuracy and recall can also be obtained for different anomaly data amounts.As to the running time,the proposed approach also harvests a reduction of 2.0~3.0 seconds for different test data amounts.