基于机器学习的风电机组变桨系统故障研究  

FAULT ANALYSIS OF WIND TURBINE PITCH SYSTEM BASED ON MACHINE LEARNING

作　　者：熊中杰 邱颖宁[1] 冯延晖[1] 程强 Xiong Zhongjie;Qiu Yingning;Feng Yanhui;Cheng Qiang(School of Energy and Power Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)

机构地区：[1]南京理工大学能源与动力工程学院,南京210094

出　　处：《太阳能学报》2020年第5期85-90,共6页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金(51505225);江苏省六大人才高峰(ZBZZ-045);江苏省自然科学基金(BK20131350);中央高校基本科研业务费专项(30915011324)。

摘　　要：针对变桨系统的齿形带断裂故障问题,首先通过分析变桨系统的工作原理,基于多维SCADA信号进行特征数据挖掘,再利用主成分分析法对数据进行预处理,并保留时序信息重构数组,最后利用高斯核支持向量机进行机器学习,实现对齿形带断裂故障的智能检测。结果表明该方法可准确诊断齿形带断裂故障,并已通过多台风电机组监测数据进行验证,准确性可达到98.8%,证明该文所用方法和模型的广泛适用性。研究结果可对未来智慧风电场的开发管理提供有利用价值的工具。To solve the problem of timing belt fracture in pitch system,feature signals are mined from large amounts of SCADA data by the analysis on the working principles. The signal data are processed by principal component analysis method next,retaining the timing information. Finally,the Gaussian kernel support vector machine is used for machine learning,which achieves the intelligent detection of the fault. The results show that the method can diagnose the fracture fault of timing belt accurately,and the accuracy of the fault detection algorithm can reach 98.8%,which proves the extensive applicability by checking the data from multi wind turbines. Furthermore,the study provides a useful tool for the development and management of the future wisdom wind field.

关 键 词：风电机组 故障诊断 主成分分析 核支持向量机 SCADA数据 变桨系统 

分 类 号：TM315[电气工程—电机]