电子测量与仪器学报

王朋飞,刘长良,徐健,刘卫亮.基于图注意力和时间卷积网络的风电齿轮箱故障预警方法[J].电子测量与仪器学报,2023,37(8):204-213

基于图注意力和时间卷积网络的风电齿轮箱故障预警方法

Wind turbine gearbox fault warning method based on graphattention and temporal convolutional network

DOI：

英文关键词:wind power gearbox fault warning graph attention network temporal convolutional network

基金项目:北京市自然科学基金(4182061)项目资助

作者	单位
王朋飞	1. 华北电力大学控制与计算机工程学院
刘长良	1. 华北电力大学控制与计算机工程学院，2. 保定市综合能源系统状态检测与优化调控重点实验室
徐健	1. 华北电力大学控制与计算机工程学院
刘卫亮	1. 华北电力大学控制与计算机工程学院，2. 保定市综合能源系统状态检测与优化调控重点实验室

Author	Institution
Wang Pengfei	1. School of Control and Computer Engineering, North China Electric Power University
Liu Changliang	1. School of Control and Computer Engineering, North China Electric Power University，2. Baoding Key Laboratory of State Detection and Optimization Regulation for Integrated Energy System
Xu Jian	1. School of Control and Computer Engineering, North China Electric Power University
Liu Weiliang	1. School of Control and Computer Engineering, North China Electric Power University，2. Baoding Key Laboratory of State Detection and Optimization Regulation for Integrated Energy System

摘要点击次数: 395

全文下载次数: 384

中文摘要:

针对风电齿轮箱故障预警中数据信息挖掘不充分问题,提出一种基于图注意力和时间卷积网络的风电齿轮箱故障预警方法。分别从时间与空间尺度建立各特征点的物理联系,拓宽特征维度以提升故障预警精度。图注意力网络构建不同数据测点间的空间拓扑结构,遍历每个节点的相邻节点进行加权求和达到聚合信息的目的;时间卷积网络使用特殊的因果膨胀卷积和残差网络,扩大感受野,提升时间特征捕捉能力。以华北某风电场实际数据为例进行验证,结果表明,提出方法能够在故障发生前 122 h 监测到风电齿轮箱的异常状态并发出预警信号;与其他方法进行对比,提出方法预警时间提前 52~ 63 h,模型预测误差减小 1. 05% ~ 3. 76%;使用 t-SNE 和概率密度曲线提升结果可解释性。

英文摘要:

In the context of wind turbine gearbox fault early warning, this paper proposes a method based on graph attention and temporal convolutional networks to address the issue of insufficient data information mining. By establishing physical connections for each feature point in both temporal and spatial scales, the method expands the feature dimension to enhance fault warning accuracy. Graph attention network captures spatial relationships, while temporal convolutional network improves temporal feature capturing. Experimental results using real data from a wind farm show that the proposed method can issue fault warnings 122 hours in advance, outperforming other methods by 52 to 63 hours with reduced prediction errors (1. 05% to 3. 76%). The approach also enhances result interpretability using t-SNE and probability density curve analysis.

查看全文查看/发表评论下载PDF阅读器