基于Bagging算法构造强分类器的one class SVM导线舞动预测应用  

Prediction of transmission line galloping by one class SVM based on the Bagging algorithm for constructing a strong classifier

作　　者：程永锋 汉京善 刘彬 李鹏 姬昆鹏 CHENG Yongfeng;HAN Jingshan;LIU Bin;LI Peng;JI Kunpeng(China Electric Power Research Institute,Beijing 102401,China)

机构地区：[1]中国电力科学研究院有限公司,北京102401

出　　处：《振动与冲击》2020年第9期152-158,共7页Journal of Vibration and Shock

基　　金：国家电网有限公司科技项目资助(18-GW-07/SGJLDK00KJJS1800056)。

摘　　要：考虑到传统物理分析方法无法解决导线舞动的预测问题,综合运用机器学习算法,对已有的舞动历史数据进行筛选和预处理,并挖掘有效信息,利用one class SVM算法解决舞动数据中负样本缺失问题,采用集成学习算法中Bagging算法建立分类器学习方法,实现了数据的随机抽样,分成不同组数据集进行相互独立的训练,避免对舞动数据过拟合,提升机器学习算法的抗噪声能力以及泛化能力,采用k折交叉验证算法进行模型的验证,并利用F1-score描述导线舞动预警模型的性能,验证了该方法在舞动预测方面的有效性。Considering that the traditional physical analysis method cannot solve the prediction of conductor galloping,a machine learning algorithm was synthetically used to screen and pre-process the existing dance history data.Mining effective information from the selected and preprocessed data,the one class SVM algorithm was used to carry out an unsupervised learning of the galloping history data.In the training of the galloping prediction model,the Bagging algorithm was used in the ensemble learning algorithm to train the classifier.The monitoring data was randomly sampled into different sets of data sets for independent training,which avoids over-fitting and improves the anti-noise ability and generalization ability of the machine learning algorithm.Using the k-fold cross validation algorithm to verify the accuracy of the prediction model and using the F1-score to describe the performance of the traverse galloping early warning model,the effectiveness of the machine learning method in the aspect of galloping prediction was confirmed.

关 键 词：导线舞动 机器学习 ONE CLASS SVM 集成学习 BAGGING算法 F1-score 

分 类 号：TM751[电气工程—电力系统及自动化]