基于卷积孪生网络的风力机状态监测研究  

Research on Condition Monitoring for Wind Turbine Based on Convolutional Siamese Network

作　　者：杨万千 张明明 YANG Wanqian;ZHANG Mingming(School of Mechanical Engineering and Automation,Harbin Institute of Technology(Shenzhen),Shenzhen 518055,China;Institute of Green and Low-Carbon Energy Technology,Harbin Institute of Technology(Shenzhen),Shenzhen 518055,China)

机构地区：[1]哈尔滨工业大学(深圳)机电工程与自动化学院,深圳518055 [2]哈尔滨工业大学(深圳)绿色低碳能源创新技术研究所,深圳518055

出　　处：《工程热物理学报》2025年第4期1152-1161,共10页Journal of Engineering Thermophysics

基　　金：国家重点研发计划资助(No.2022YFB4201200);深圳市高等院校稳定支持计划重点项目(No.GXWD20220817140906007);哈尔滨工业大学(深圳)科研启动项目。

摘　　要：风力机运行故障具有突发性,几乎无预留时间做出反应。连带地,一些微小缺陷将引发一系列连锁故障和非必要损失,因此对其状态监测和提前预知机组状态是一个必要问题。本文基于卷积孪生网络,结合机组SCADA数据进行基于模型训练的状态监测。利用历史健康数据进行离线模型训练,经测试后将预训练模型加载到在线监测系统,基于状态指示来描述机组的运行状态,并结合统计过程控制技术定义机组状态异常的监测阈值,从而实现在线状态监测。通过对甘肃省某风电场运行SCADA数据及故障报告的分析对比,本文方法具有较好的机组状态监测和预知性能。The operation faults of wind turbines are sudden and there is almost no reserved time for people to respond.In turn,some minor defects will cause a series of chain faults and unnecessary losses.Therefore,monitoring the conditions and predicting the conditions in advance are necessary.In this article,based on convolutional siamese networks,SCADA data of wind turbines are combined for model training-based condition monitoring.Historical healthy data is used for offline model training,and the pre trained models are loaded into the online monitoring system after testing.Status-indication is defined to describe the operating conditions of the wind turbine.The monitoring threshold of abnormal wind turbines is proposed according to statistical process control during online monitoring.With the analysis and comparison of SCADA data and fault reports of a wind farm in Gansu Province,the method proposed in this paper is in good performance in monitoring and predicting wind turbine conditions.

关 键 词：风电机组 状态监测 卷积孪生网络 智能运维 SCADA 

分 类 号：TK83[动力工程及工程热物理—流体机械及工程]