Gas turbine heat rate prediction in combined cycle power plant using artificial neural network

Abstract

Artificial neural network (ANN) models for predicting the heat rate (HR) of gas turbines in a combined cycle power plant (CCPP) were developed and compared in this study. The heat rate, a critical performance indicator, reflects the amount of fuel energy required to electricity generation. A lower heat rate indicates higher efficiency and reduced fuel consumption. The first model uses seven input variables, including fuel gas temperature (FT), ambient temperature (AT), relative humidity (RH), compressor outlet temperature (CT), compressor outlet pressure (CP), variable guide vane (VGV), and gas turbine heat input (HI). The second model includes an additional input variable, power output (PO), making it an eight-input model. Both models were performed in MATLAB using the Levenberg-Marquardt algorithm, with node variations from 1 to 20, to determine the optimal network architecture. The 8-input model demonstrated superior performance, with a higher prediction accuracy (R² = 0.986) and lower mean squared error (MSE = 518) compared to the 7-input model (MSE = 1,053). PO shows the strongest inverse relationship to HR (R = −0.898), which aligns with thermodynamic principles, where increased power output corresponds to improved energy conversion efficiency. CP, HI, and VGV also have significant negative relationships with HR. These findings indicate that incorporating power output as an additional input variable significantly improves the model’s ability to predict the heat rate. The ANN models offer a reliable and accurate tool for monitoring heat rates, optimizing energy efficiency, and supporting operational decision-making in gas turbines at combined-cycle power plants.