基于双通道特征融合并行优化的燃气轮机气路故障诊断方法  

A gas path fault diagnosis method for gas turbine based on parallel optimization of dual-channel feature fusion

作　　者：张菲菲 应雨龙 李靖超[2] ZHANG Feifei;YING Yulong;LI Jingchao(School of Energy and Mechanical Engineering,Shanghai University of Electric Power,Shanghai 200090,China;School of Electronic and Information,Shanghai Dianji University,Shanghai 201306,China)

机构地区：[1]上海电力大学能源与机械工程学院,上海200090 [2]上海电机学院电子信息学院,上海201306

出　　处：《热力发电》2022年第12期30-38,共9页Thermal Power Generation

基　　金：国家自然科学基金面上项目(62076160);上海市自然科学基金面上项目(21ZR1424700)。

摘　　要：为解决当前燃气轮机基于数据驱动的气路故障诊断精度提升的问题,提出了一种基于双通道特征融合并行优化的气路故障诊断方法。该方法首先通过燃气轮机热力模型构建气路故障数据样本集;然后使用卷积神经网络及长短时记忆神经网络双通道并行挖掘数据的空间特征和时序特征,并在两通道中分别引入首层大卷积核及挤压激励网络的优化方法;最后将两通道提取的特征融合为一维张量,输入到全连接层进行气路故障类型识别。实验结果表明,相较于传统机器学习及深度学习的气路故障诊断方法,所提方法具有更优的辨识精度,平均诊断准确率达到了98.24%,具有实用性及可行性。In order to solve the problem that the diagnostic accuracy of current gas turbine gas path faults based on data-driven needs to be improved, a gas path fault diagnosis method based on parallel optimization of dual-channel feature fusion is proposed. Firstly, a gas turbine thermodynamic model is used to construct a sample set of gas path fault data. Secondly, convolutional neural network and long and short term memory neural network are used to mine the spatial and temporal features of the data in parallel, and the optimization methods of the first layer large convolutional kernel and squeeze excitation network are introduced in the two channels respectively. Finally, the features extracted from the two channels are fused into a one-dimensional tensor and then reconstructed into the fully connected layer for gas path fault type identification. The experimental results show that, the proposed method has better recognition accuracy than the conventional machine learning and deep learning gas path fault diagnosis methods, with an average diagnosis accuracy of 98.24%, which is practical and feasible.

关 键 词：燃气轮机 气路故障诊断 卷积神经网络 长短时记忆神经网络 特征融合 并行优化 

分 类 号：TK47[动力工程及工程热物理—动力机械及工程]