V94.2 gas turbine identification using neural network

Abstract

This paper presents the identification of V94.2 gas turbine. This turbine is built by Siemens. It has 162.1 MW nominal power and 50 Hz nominal frequency and is located at Kermanshah power plant, Kermanshah city of Iran. The stored data from turbine include fuel pressure valve angle and IGV <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">1</sup> angle as inputs and compressor output pressure, compressor output temperature, fuel pressure, turbine output power and turbine output temperature as outputs. To simplify identification process, the system turns into MISO <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> systems to the number of outputs, and then correlation analysis is used to examine the dependence of the outputs to each input and other outputs. For turbine identification, dynamic linear models are estimated and then Feedforward neural network with one hidden layer is trained. The result shows dynamic linear models have poor performance in comparison with Feedforward neural network with one hidden layer. The neural network is able to identify a predictor model with fitness over 96% for outputs of V94.2 gas turbine.