基于动态阈值AdaBoost算法的风电机组发电机电气故障诊断研究  

Research on Electrical Fault Diagnosis of Wind Turbine Generator Based on Dynamic Threshold AdaBoost Algorithm

作　　者：彭艳来 樊永 杨晓峰 杨宏宇 王志新[3] PENG Yanlai;FAN Yong;YANG Xiaofeng;YANG Hongyu;WANG Zhixin(General Management Department,Longyuan Power Group(Shanghai)New Energy Co.,Ltd.,Shanghai 200000,China;Software Department,Shanghai Proinvent Information Tech Co.,Ltd.,Shanghai 200000,China;Department of Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200030,China)

机构地区：[1]龙源电力集团(上海)新能源有限公司综合管理部,上海200000 [2]上海博英信息科技有限公司软件部,上海200000 [3]上海交通大学电气工程系,上海200030

出　　处：《电气传动》2023年第6期91-96,共6页Electric Drive

基　　金：国家重点研发计划(2018YFB1503000)。

摘　　要：针对风场SCADA数据量庞大,难以实现风电机组发电机快速故障分类的问题,提出一种动态阈值AdaBoost算法(DTAdaBoost)。在该算法中引入动态阈值对样本集数据进行筛选,剔除训练模型贡献较小的数据,提出的算法不仅可以减小样本容量,还能使被错误分类的数据多次训练,最终实现发电机气隙不均、匝间短路、断条等电气故障的快速诊断。在与其他方法对比的实验结果表明:DTAdaBoost算法在电气故障诊断中比常规AdaBoost算法、SWTAdaBoost算法、GAAdaBoost算法运算时间快、准确率高,为风电机组的快速故障诊断提供理论基础。Aiming at the problems that the amount of data in wind farm SCADA data is huge and it is difficult to classify the fast fault of wind turbine generator,a dynamic threshold AdaBoost algorithm(DTAdaBoost)was proposed.Based on AdaBoost algorithm,the dynamic threshold was introduced to screen the sample set data,and eliminate the data that contribute little to the training model,reduce the sample size,train the incorrectly classified data for many times,at last the rapid diagnosis of electrical faults such as uneven generator air gap,inter turn short circuit and broken bar was realized.Compared with other classification methods,the experiments results show that DTAdaBoost algorithm has faster operation time and higher accuracy than conventional AdaBoost algorithm,SWTAdaBoost algorithm and GAAdaBoost algorithm in electrical fault diagnosis.It provides a theoretical basis for rapid fault diagnosis of wind turbine.

关 键 词：故障分类 SCADA数据 ADABOOST算法 数据处理 

分 类 号：TM614[电气工程—电力系统及自动化]