Dynamic Data Reconciliation of Gas Turbine Based on PCA-LSTM

Due to the state deviation and degradation of the sensor, the quality of the measurement data will be reduced, which will affect the reliability of the fault diagnosis results and the stability of the control. In this paper, taking GE 9 FA heavy duty gas turbine as the research object, a data reconciliation model based on principal components analysis and the long-short term memory neural network (PCA-LSTM) is proposed to eliminate the influence of data quality reduction. By using the PCA principal component extraction method, the LSTM network input dimensions is reduced. When the sensor failed, the proposed dynamic data reconciliation method based on PCA-LSTM can isolate the sensor fault well and can reconstruct the fault data of the measurement parameters. The reconstruction error of the sensor fault signal does not exceed 1.5%.