基于CKF-LSTM的燃气轮机气路故障诊断研究
Study on Gas Path Fault Diagnosticfor Gas Turbine based on CKF-LSTM
提出了基于容积卡尔曼滤波-长短时记忆网络（Cubature Kalman Filters - Long Short-term MemoryCKF-LSTM）的多模型燃气轮机气路故障诊断方法，该方法集成了基于模型和数据驱动的优点。使用容积卡尔曼滤波器建立先验故障状态估计模型，提取运行状态残差特征；采用LSTM神经网络同时对多个先验状态故居模型残差特征进行识别，实现气路故障的诊断。利用重型燃气轮机典型故障仿真数据对所提出的方法进行测试，验证结果表明：基于CKF-LSTM的燃气轮机气路故障诊断方法具有良好的诊断准确性，在重型燃气轮机3个工况点的故障诊断准确率均高于95.6%。
A multi-model gas turbine fault diagnosis method based on Cubature Kalman Filters-Long Short-term Memory CKF-LSTM is proposed, which integrates the advantages of model-based and data-driven. A priori fault state estimation model is established with volumetric Kalman filter to extract the residual feature of operating state. LSTM neural network is used to identify residuals of former residence models with multiple prior states at the same time to realize gas path fault diagnosis. The proposed method is tested by using typical fault simulation data of heavy-duty gas turbine. The verification results show that the gas path fault diagnosis method based on CKF-LSTM has good diagnostic accuracy, and the fault diagnosis accuracy is higher than 95.6% at the three working conditions of heavy-duty gas turbine.
康宇航;曹云鹏;李淑英;
哈尔滨工程大学动力与能源工程学院,哈尔滨150001;哈尔滨工程大学动力与能源工程学院,哈尔滨150001;哈尔滨工程大学动力与能源工程学院,哈尔滨150001;
TK478
heavy-duty gas turbine fault diagnosis kalman filter neural network multi-model
1-7 / 7
评论