基于LightGBM的风力发电机故障诊断  

FAULT DIAGNOSIS FOR WIND TURBINE BASED ON LIGHTGBM

作　　者：胡澜也 蒋文博 李艳婷[1] Hu Lanye;Jiang Wenbo;Li Yanting(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200241,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200241

出　　处：《太阳能学报》2021年第11期255-259,共5页Acta Energiae Solaris Sinica

基　　金：国家自然科学基金面上项目(71672109);国家自然科学基金重点项目(71531010)。

摘　　要：采用风力发电机数据采集与监视控制(SCADA)系统运行数据和状态数据,结合LightGBM方法构建风力发电机故障模式识别模型。首先依据风力发电机的运行状态数据,给无标签的SCADA数据添加状态标签。针对同时段出现的重叠故障,将其单独提取,定义为新类别的故障。最后采用集成学习中的LightGBM构建故障诊断模型,与Xgboost、梯度提升树、随机森林、支持向量机、多层感知器、深度信念网络进行对比,证实了LightGBM方法的优越性。In this study,wind turbine supervisory control and data acquisition(SCADA)system operation data and status data are used,and the wind turbine fault pattern recognition model is constructed by applying LightGBM method.Firstly,according to the operating state data of the wind turbine,the unlabeled SCADA data is tagged with various statues,and the overlapping faults appearing at the same time are separately extracted and defined as new types of faults.Finally,the fault diagnosis model is built by applying LightGBM.Compared with Xgboost,gradient boosting tree,random forest,support vector machine,multi-layer perceptron and deep belief network,the superiority of LightGBM method is confirmed.

关 键 词：风力发电机 SCADA系统 状态监测 模式识别 LightGBM 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TK83[自动化与计算机技术—控制科学与工程]