基于DBSCAN-ML的液压风力发电机故障诊断研究  

Research on Fault Diagnosis of Hydraulic Wind Turbine Based on DBSCAN-ML

作　　者：宾世杨 李利强 程乐 陈浩武 BIN Shiyang;LI Liqiang;CHENG Le;CHEN Haowu(State Power Investment Group Guangxi Xingan Wind Power Co.,Ltd.,Guilin Guangxi 541300,China;College of Urban and Environmental Scienses,Northwestern University,Xi’an Shaanxi 710127,China)

机构地区：[1]国家电投集团广西兴安风电有限公司,广西桂林541300 [2]西北大学城市与环境学院,陕西西安710127

出　　处：《机床与液压》2024年第14期227-235,共9页Machine Tool & Hydraulics

基　　金：国家电投集团科技项目(KY-TC-2023-SK2433);陕西省科学技术研究计划项目(2022KXJ-887)。

摘　　要：传统风力发电机对于系统故障的解决方案是有限和预先确定的,而具有大量传感器数据的故障预测诊断可以有效预防可能发生的系统故障,从而降低设备维护成本。为此,提出一种基于DBSCAN-ML的风力发电机故障诊断策略。基于密度的应用噪声算法空间聚类(DBSCAN)从正常状态数据中分类出异常状态的风力机数据,然后采用决策树和随机森林算法2种机器学习(ML)算法构建预测模型,最后使用K折交叉验证进行测试。通过广西31台风力发电机组数据对此故障诊断方案进行案例验证。结果表明:DBSCAN算法可以有效分离异常状态数据,且决策树预测模型和随机森林模型可以分别获得92.7%和92.1%的准确率,通过数据挖掘和建模可以检测风力发电机组的故障,并可以预测部件的维护需求。Traditional wind turbines are limited and predetermined for system solutions of faults,but fault prediction diagnosis with a large amount of sensor data can effectively prevent possible system faults,thus reducing equipment maintenance costs.Therefore,a fault diagnosis strategy for wind turbines based on DBSCAN-ML was proposed.The density-based spatial clustering of applications with noise(DBSCAN)was applied to classify wind turbine data in abnormal state from normal state data,and then two machine learning(ML)algorithms,namely decision tree and random forest algorithm,were used to construct prediction models.Finally,K-fold cross validation was used to test.With the data of 31 wind turbines in Guangxi,this fault diagnosis scheme was verified by case.The results show that DBSCAN algorithm can effectively separate abnormal state data,and the accuracy of decision tree prediction model and random forest model can obtain 92.7%and 92.1%,respectively.Through data mining and modeling,wind turbine faults can be detected.The maintenance needs of the parts can be predicted.

关 键 词：风力发电机 基于密度的应用噪声算法空间聚类(DBSCAN) 机器学习(ML) 决策树 随机森林 K折交叉验证 故障诊断 

分 类 号：TM133.33[电气工程—电工理论与新技术]