Font Size: 

The parametrized reduced order modeling (PROM) technique is an important way to improve the efficiency of numerical simulation for complex unsteady flows. In this paper, a non-intrusive PROM is proposed based on the notion of data driven. A set of flow parameters samples selected by Latin hypercube sampling (LHS) are used to provide effective high-fidelity flows data. Spatial features of flows are extracted by proper orthogonal decomposition (POD) technique, and a novel temporal features extraction method is proposed to improve the efficiency of reduction for unsteady flow with multiple parameters. The reduced coefficients obtained by projecting high-fidelity solutions onto reduced space are mapped with flow parameters by Gaussian process regression (GPR). This approach provides stable and efficient foundations for predicting the reduced coefficients for complex unsteady flows with multiple parameters. The accuracy and effectiveness of PROM are demonstrated by two numerical examples, namely flow past a cylinder and turbine flow with multiple flow parameters.