[Objective] Numerical simulation of a gas turbine combustor is an important step in its design process. Due to the complexity of the physical and chemical processes, the calculation cost is high. The calculation cost can be reduced by constructing surrogate model of combustor. This paper focuses on the key steps in the construction of a surrogate model suitable for cold gas flow and the mixing process of combustor. Furthermore, this paper proposes a surrogate model for the central nozzle and the subsequent combustor space of a heavy-duty gas turbine. [Methods] The construction of the surrogate model includes several key steps: design of experiments (DOE), numerical simulation, dimensionality reduction, and an interpolation process. Two parameters are selected as the input parameters for the surrogate model: The fuel mass flow rate Gf and the combustor inlet air pressure p2. Latin hypercube sampling is used in the DOE to determine 12 operating conditions for computational fluid dynamics (CFD) simulations, and the results are used to build the surrogate model. Proper orthogonal decomposition is used for dimensionality reduction, wherein a set of basis functions and corresponding coefficients are extracted. The basis functions reflects the main characteristics of the combustor flow field. Moreover, the data dimensionality is reduced from the number of grid nodes of the combustor to the number of basis functions, which do not exceed the number of operating conditions. The Kriging model is used to interpolate the coefficients of the basis function with the input parameters of the surrogate model. Four verification conditions are set up to determine the accuracy of the surrogate model through a comparison of the surrogate model results with the CFD simulation results. The outlet cross section of the central nozzle and the longitudinal section of the combustor are selected to compare multiple key parameter distributions, including axial velocity, radial velocity, tangential velocity, CH4 concentration, turbulent kinetic energy, and pressure. The vector operations are used to compare the distributions of various parameters, which can simultaneously reflect the differences in the numerical and spatial distributions of various parameters. [Results] The results showed that the error in most parameters was ~1%. The results also revealed that the construction method of the surrogate model could be applied to cold high-speed and high-turbulence strong rotational flow and fuel/air mixing. The accuracy was higher than the international average level, and the application criteria of the construction method were proposed. The influences of interpolation methods, sample numbers, and basis function numbers on the accuracy of the surrogate model were analyzed. The accuracy of SM was higher than extrapolation. Increasing the number of sample operations and basis functions could improve the accuracy of the surrogate model, but also increased the computational cost. [Conclusions] The SM construction method (POD & Kriging) is suitable for the cold gas flow and mixing process in the combustor. The paper lays the foundation for subsequent research on the construction method of combustor SM, which includes combustion reactions and geometric structure changes. © 2023 Press of Tsinghua University. All rights reserved.
