基于SSA和注意力机制BiLSTM的燃气轮机传感器故障诊断方法研究  

Gas Turbine Sensor Fault Diagnosis Method Based on SSA and BiLSTM with Attention Mechanism

作　　者：程侃如 王玉璋[1] 杨志鹏 杨喜连 CHENG Kanru;WANG Yuzhang;YANG Zhipeng;YANG Xilian(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Shanghai Power Equipment Research Institute Co.,Ltd.,Shanghai 200240,China;School of Electronic Information and Electrical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240 [2]上海发电设备成套设计研究院有限责任公司,上海200240 [3]上海交通大学电子信息与电气工程学院,上海200240

出　　处：《动力工程学报》2023年第9期1181-1189,共9页Journal of Chinese Society of Power Engineering

基　　金：国家科技重大专项资助项目(No.2017-V-0011-0063)。

摘　　要：提出了一种结合奇异谱分析(SSA)和基于注意力机制的双向长短期记忆(BiLSTM)深度网络的传感器故障诊断方法。首先,利用SSA对传感器信号进行预处理,得到信号的趋势项和周期项;其次,将预处理得到的时间序列输入带有注意力机制的BiLSTM深度网络进行训练,得到分类器模型;最后,采用某9F级燃气轮机运行数据对该方法进行训练、验证和测试。结果表明:测试集的诊断准确率可达96.5%,该方法能有效解决故障信号稀疏性和幅度微弱性造成的传感器故障诊断精度低的问题。A sensor fault diagnosis method combining singular spectrum analysis(SSA)and Bi-direction long short-term memory(BiLSTM)deep network based on attention mechanism was proposed.Firstly,the sensor signal was preprocessed using SSA to obtain the trend term and period term of the signal.Secondly,the preprocessed time series were input into the BiLSTM deep network with attention mechanism for training to obtain the classifier model.Finally,the method was trained,verified and tested by using the operation data of a 9F gas turbine.Results show that the diagnostic accuracy of the test set can reach 96.5%,and the proposed method can effectively solve the sensor problem of low fault diagnosis accuracy caused by sparseness and weak amplitude of fault signals.

关 键 词：燃气轮机 传感器 故障诊断 深度学习 

分 类 号：TK221[动力工程及工程热物理—动力机械及工程]