基于超球面支持向量机的SF6数字化表计数据异常检测  

SF6 digital meter data anomaly detection based on hyperspherical support vector machine

作　　者：任友军 袁乐 印吉景 揣振国 REN Youjun;YUAN Le;YIN Jijing;CHUAI Zhenguo(Taizhou Power Supply Branch of State Grid Jiangsu Electric Power Co.,Ltd.,Taizhou 225300,Jiangsu China)

机构地区：[1]国网江苏省电力有限公司泰州供电分公司,江苏泰州225300

出　　处：《粘接》2025年第5期186-188,192,共4页Adhesion

基　　金：国网江苏省电力有限公司孵化项目(项目编号:JF2023003)。

摘　　要：为了弥补支持向量机SVM训练时间过长的问题,构建一种通过DBN实现的1/4超球面支持向量机QSSVM模型,实现SF6数字化表计在线测试功能的异常检测。先通过DBN对高维数据实施降维后再采用QSSVM和滑动窗口模型相融合的分析方法来实现对异常问题的高效测试。研究结果表明,当窗口扩大后,QSSVM准确度不断提高。在窗口为100情况下,QSSVM相对于SVM可以降低近一半的计算时间。当样本维度升高后,QSSVM依然具备优异检测性能,获得高达94.16%的检测率。该研究有助于提高SF6数字化表计数据异常检测能力,具有很好的实际推广价值。In order to make up for the problem of long SVM training time,a 1/4 hyperspherical SVM QSSVM model realized by DBN was constructed to realize the anomaly detection of the online test function of SF6 digital meter.Firstly,the DBN was used to reduce the dimensionality of high-dimensional data,and then the analysis method of QSSVM and sliding window model was used to achieve efficient testing of abnormal problems.The results showed that the accuracy of QSSVM continued to improve when the window was expanded.With a window of 100,QSSVM could reduce computation time by nearly half relative to SVM.When the sample dimension was increased,QSSVM still had excellent detection performance,and the detection rate was as high as 94.16%.This study is helpful to improve the anomaly detection ability of SF6 digital meter data,and has good practical promotion value.

关 键 词：无线传感器网络 气体异常检测 深度信念网络 支持向量机 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]