Dynamic modelling and simulation of a power gas turbine using artificial neural network: A comparative study
Date
2022-12
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Federal University of Technology, Owerri
Abstract
The dynamic modelling and simulation of a power gas turbine by comparing three forms of Artificial Neural Network was adopted given the complexities of the physics and
mathematical based turbine models. Layer Recurrent Neural Network (Layrecnet), Feed Forward Back Propagation (FF BPP) Network and Non Linear Autoregressive Network with
Exogenous Input (NARX) were selected for the dynamic modelling of the turbine. The start up data was trained with these networks and Multiple Input Multiple Output (MIMO) and
Multiple Input Single Output (MISO) models were developed for the machine. Furthermore; the selected models were validated with operational data from the turbine similar in
manouver to the data adopted for modelling. It is observed that “Layrecnet” has the least Mean Squared Error (MSE) of 1.12 and Mean of Absolute Percentage Error (MAPE) of
0.7310 in the MIMO model while “FF BPP” network comes a close second with MSE of 1.74 and MAPE of 1.4249. “LayRecNet” MIMO and MISO models were used to simulate the
start-up of the gas turbine because it ranked the highest among the three networks with the use of MSE and MAPE error performance metrics. However; the “FF BPP” network also
performed well as it had the best performance for the Turbine Outlet Temperature MISO model with MSE of 0.296 and MAPE of 0.495. The research showed that the “Layrecnet”
Network is a better tool for dynamic time series modelling as the network had the least MSE and MAPE with FF BPP coming a close second, while the much acclaimed NARX Network
is the least performing network. It was shown that neural networks can be considered a reliable alternative to conventional mathematical driven techniques. Therefore, by using the
developed tool, an optimization of the plant operation and maintenance is rendered possible.
Description
The thesis contains tables and figures
Keywords
Gas turbine, neural networks, blackbox models, start-up phase, load, simulation, plant, manouver, Department of Chemical Engineering
Citation
Attamah, C. S. ( 2022 ). Dynamic modelling and simulation of a power gas turbine using artificial neural network: A comparative study (Unpublished Master's Thesis). Federal University of Technology. Owerri, Nigeria