基于分类算法的汽轮机故障诊断  

Fault diagnosis of steam turbine based on classification algorithm

作　　者：杨荣利 曹军[2] 张延启 周东阳[2] YANG Rongli;CAO Jun;ZHANG Yanqi;ZHOU Dongyang(Huadian Laizhou Power Generation Co.,Ltd.,Laizhou 261400,China;Xi’an Thermal Power Reserch Institute Co.,Ltd.,Xi’an 710054,China)

机构地区：[1]山东省莱州市华电莱州发电有限公司,山东莱州261400 [2]西安热工研究院有限公司,陕西西安710054

出　　处：《电子设计工程》2023年第9期68-72,共5页Electronic Design Engineering

基　　金：中国华电科技项目(CHDKJ20-04-110)。

摘　　要：汽轮机系统设备是火电厂的主力运行设备之一,对汽轮机系统进行有效的故障诊断及预测有助于保障火电机组的安全、稳定运行。基于随机森林算法对汽轮机数据进行处理,完成对汽轮机设备原始数据信息的降维与特征筛选。同时采用6种分类算法建立汽轮机正常和故障的算法模型,实现汽轮机设备状态正常、异常的故障诊断。在汽轮机含有故障的真实运行数据集上进行的实验结果表明,预测准确率较高的模型依次为梯度提升决策树、随机森林、决策树以及K近邻分类算法,且准确率最高可达99.98%;而预测用时较短的模型为K近邻和决策树分类算法,对20 000余条样本训练进行预测,最快可在0.034 s内完成。Steam turbine system equipment is one of the main operating equipment of thermal power plant.Effective fault diagnosis and prediction of steam turbine system is helpful to ensure the safe and stable operation of thermal power units.In this paper,the feature screening of steam turbine data is based on random forest algorithm,which realizes the dimensionality reduction and feature screening of the original data information of steam turbine equipment,and six classification algorithms are used to establish the classification algorithm model of normal and fault of steam turbine,so as to realize the fault diagnosis of normal and abnormal state of steam turbine equipment.The experimental results on the real operation data set containing faults of steam turbine show that the models with high prediction accuracy are gradient lifting decision tree,random forest,decision tree and K-nearest neighbor classification algorithm,and the highest accuracy can reach 99.98%;The model with short prediction time is K-nearest neighbor and decision tree classification algorithm.The training and prediction of more than 20000 samples can be completed in 0.034 s at the fastest.

关 键 词：汽轮机 故障诊断 分类算法 随机森林 K近邻 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]