基于多时间尺度的风电机组齿轮箱早期故障监测方法OA
A multi-timescale-based early fault monitoring method for wind turbine gearboxes
针对风电机组数据采集与监视控制系统(supervisory control and data acquisition,SCADA)数据中早期故障特征微弱且运行状态多变导致故障识别困难的问题,提出了一种基于并行门控循环单元(parallel gated recurrent unit,P-GRU)与多时间尺度指数加权移动平均(multi-time-scale exponentially weighted moving average,MTS-EWMA)的风电机组齿轮箱早期故障监测方法.首先,利用重叠滑动窗口采样(overlapping sliding window sampling,OSWS)策略对SCADA多变量时间序列进行重构,以充分保留时序信息.然后,构建P-GRU预测模型,通过并行时序特征提取结构增强模型对复杂运行工况下多变量动态关联特征的学习能力.然后,在此基础上构建MTS-EWMA监测框架,通过融合不同时间尺度的预测残差信息,实现对缓慢演化故障的敏感检测与误报抑制.实验结果表明,所提方法能够在风电机组实际故障发生前实现稳定预警,验证了该方法在风电机组早期故障监测中的有效性.
To address the challenges of fault identification in wind turbines caused by weak early fault features and variable operating conditions in SCADA data,this paper proposes an early fault monitoring method for wind turbine gearboxes based on parallel gated recurrent unit(P-GRU)and multi-timescale exponentially weighted moving average(MTS-EWMA).First,an overlapping sliding window sampling(OSWS)strategy is employed to reconstruct the SCADA multivariate time series to fully preserve temporal information.Subsequently,a P-GRU prediction model is constructed,and through its parallel temporal feature extraction architecture,the model's capability to learn multivariate dynamic correlation features under complex operating conditions is enhanced.Finally,an MTS-EWMA monitoring framework is constructed to fuse prediction residuals from different time scales,thereby enabling sensitive detection of slowly evolving faults while suppressing false alarms.Experimental results demonstrate that the proposed method can provide stable early warnings before actual wind turbine faults,thereby verifying its effectiveness for early fault monitoring of wind turbine gearboxes.
张程;雷步辰;余士霖;陆万林;陈昌亮
福建理工大学电子电气与物理学院,福建 福州 350118||智能电网仿真分析与综合控制福建省高校工程研究中心,福建 福州 350118福建理工大学电子电气与物理学院,福建 福州 350118福建理工大学电子电气与物理学院,福建 福州 350118福建理工大学电子电气与物理学院,福建 福州 350118国网福建省电力有限公司宁德供电公司,福建 宁德 352100
风电机组SCADA数据故障监测门控循环单元时间序列预测指数加权移动平均
wind turbineSCADA datafault monitoringgated recurrent unittime series forecastingexponentially weighted moving average
《电力系统保护与控制》 2026 (19)
73-84,12
This work is supported by the National Natural Science Foundation of China(No.52377088). 国家自然科学基金项目资助(52377088)福建省财政厅专项资助(GY-Z220230)福建省自然科学基金项目资助(2023J01951)
评论