Title :
Identification of heavy duty gas turbine startup mode by neural networks
Author :
Refan, Mohammad Hossein ; Taghavi, S.H. ; Afshar, Ahmad
Abstract :
In this study, measured data is used for identification of stationary gas turbine in startup stage. To choose persistently excited signal, appropriate input/output selection techniques are utilized, in addition to the application of regressor addition for performance improvement. Also effect of sampling interval and data pretreatments considered. Peak shaving, scaling, offset correction applied to practical data to change raw data into proper material for identification. Preprocessed data applied to two configurations of neural networks which are MLP and RBF for prediction model application with optimal parameters and different arrangements. Finally, attained MLP and RBF networks compared with each other, from performance and convergence speed point of view. Practical data collected during startup from v94.2 v5 heavy duty gas turbine located in an actual power plant.
Keywords :
gas turbine power stations; multilayer perceptrons; power engineering computing; radial basis function networks; MLP networks; RBF networks; data pretreatments; excited signal; heavy duty gas turbine startup mode identification; input-output selection techniques; neural network configurations; offset correction; peak shaving; power plant; prediction model application; preprocessed data; raw data; sampling interval effect; stationary gas turbine identification; System identification; gas turbine startup; neural networks; nonlinear modeling;
Conference_Titel :
Thermal Power Plants (CTPP), 2012 4th Conference on
Conference_Location :
Tehran
Print_ISBN :
978-1-4673-4844-7
