基于深度置信网络的燃气轮机气路故障诊断方法  

A Gas Path Fault Diagnostic Method of Gas Turbine Based on Deep Belief Network

作　　者：闫丽萍 董学智[1,2] 张永军 王涛[1,2] 高庆 谭春青[1,2] YAN Li-Ping;DONG Xue-Zhi;ZHANG Yong-Jun;WANG Tao;GAO Qing;TAN Chun-Qing(Institute of Engineering Thermophysics,Chinese Academy of Sciences,Beijing 100190,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院工程热物理研究所,北京100190 [2]中国科学院大学,北京100049

出　　处：《工程热物理学报》2020年第4期840-844,共5页Journal of Engineering Thermophysics

基　　金：国家重点研发项目(No.2018YFB0905101)。

摘　　要：针对燃气轮机气路部件性能衰退故障,引入在深度学习领域取得重大突破的深度置信网络DBN。然而目前DBN中初始参数的选择大多依赖经验,没有形成完备的参数选择机制。针对以上问题,本文提出一种基于遗传算法自适应调整参数的深度置信网络优化算法。以三轴式燃气轮机为对象,将DBN优化算法与其他神经网络方法比较,结果表明DBN优化算法故障诊断平均精度可达88.3%,明显优于BP、RBF、ELM和SVM方法。增设高压压气机出口温度或压力传感器可显著提高燃气轮机故障诊断精度,在DBN优化算法下,故障诊断平均精度提升至98.4%。Focus on the performance degradation of gas path component in gas turbine,deep belief network(DBN)which has made a breakthrough in deep learning filed,is introduced.However,the selection of initial parameters in DBN mostly depends on experience at present,and there isn’t a complete parameter selection mechanism.In order to solve the above problems,a deep belief network optimization method which adaptively adjusts parameters based on genetic algorithm is proposed in this paper.Comparing DBN optimization method with other neural network methods for three-axis gas turbine,the results shows the average fault diagnosis accuracy of DBN optimization method can reaches 88.3%,which is obviously better than BP、RBF、ELM and SVM.Adding high pressure compressor outlet temperature and pressure sensors can significantly improve diagnosis accuracy.The average accuracy of the proposed method can be improved to 98.4%.

关 键 词：深度置信网络 遗传算法 监测参数 气路部件故障 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TK478[自动化与计算机技术—控制科学与工程]