基于群体多维相似性的风机齿轮箱预警策略  

Fault warning strategy of wind turbines gearbox based on group multi-dimensional similarity

作　　者：刘帅[1,2] 刘长良[2] 甄成刚[1] 靳昊凡[1] Liu Shuai;Liu Changliang;Zhen Chenggang;Jin Haofan(School of Control and Computer Engineering, North China Electric Power University, Beijing 102206, China;State Key Laboratory of Alternate Electrical Power System with Renewable Energy Sources, North China Electric Power University, Beijing 102206, China)

机构地区：[1]华北电力大学控制与计算机工程学院,北京102206 [2]华北电力大学新能源电力系统国家重点实验室,北京102206

出　　处：《仪器仪表学报》2018年第1期180-189,共10页Chinese Journal of Scientific Instrument

基　　金：新能源电力系统国家重点实验室自主研究课题(LAPS2016-18);中央高校基本科研业务费专项资金(9163116001)项目资助

摘　　要：风场区域气候周期性变化及气象条件瞬时变化都会直接影响风电机组设备运行状况,已有齿轮箱故障预警策略较少与周边相似风机联动,偏向于＂单体预警＂。将高斯混合模型、动态时间规整及熵权值算法三者紧密结合,提出了一种基于群体多维特征相似性的故障预警策略：通过与周边风机的相似性联动,消除周期性及瞬时性环境因素对预警结果的影响;采用分级时间滚动窗口生成风机相似关系,保留数据的时间次序属性,展示研究对象的数据资源迁徙规律,判断潜在故障风机。最后,用福建沿海风场监控与数据采集系统（SCADA）数据验证了所提预警算法的有效性与实效性,至少可以提前26天预警潜在故障风机。The climate cyclical changes of the wind farm and the instantaneous change of the meteorological conditions directly affect the operation status of the wind turbines.The existing warning methods of the gearbox failure are less associated with the similar wind turbines,preferred to ＂unit warning＂.Combining Gaussian mixture model,dynamic time warping and entropy weighting algorithm,a fault early warning strategy based on group multi-dimensional feature similarity is proposed.The effect of cyclical and instantaneous environmental factors on early warning results is eliminated by comparing the similarity with the surrounding wind turbines.The wind turbines similarity relationship is generated using the graded time sliding window,and the time order attribute of raw data is retained.The data resource migration pattern of research object can be shown to determine the potential fault.Finally,the effectiveness of the proposed algorithm is verified by SCADA data of Fujian coastal wind farm.The results show that the potential failure of the wind turbine can be predicted at least 26 days ahead.

关 键 词：齿轮箱故障预警 群体评价 高斯混合模型 动态时间规整 熵权值 

分 类 号：TH17[机械工程—机械制造及自动化]