Neural network controller of a gas turbine plant low emission combustor

Abstract

One of the most important gas turbine engine components is the combustion chamber, the main source of harmful emissions. The study is devoted to the central issues of designing and testing of an automatic control system of harmful emissions and pressure pulsations in flame tubes of a gas turbine plant with a capacity of 16 MW GTP-16 based on a PI-controller with a built-in neural network mathematical model of a low-emission combustor (LEС). Algorithms for a neural network controller of emission of nitrogen oxides and carbon monoxide into the atmosphere, as well as pressure pulsations in the LEC’s flame tubes were developed. The algorithms are given in a graphical programming environment and integrated into the automatic control system of GTP-16, implemented on the PXI NI hardware and software platform. The performance of the emission controller was checked during bench tests on the GTP-16 simulator with LEС neural network model serving as a virtual emission sensor. The errors in estimating the emission of nitrogen and carbon oxides and pressure pulsations in the flame tubes were determined. The normality of the error distribution of the developed nitrogen oxide emission model was proven. A conclusion about the prospects of using neural networks for the development of an adaptive control system of emissions and flame tube pressure pulsations for LECs of the gas turbine plants was drawn.