Reduced‐order modeling framework using two‐level neural networks

Abstract Established reduced‐order modeling (ROM) methods, for instance, Galerkin‐projection, approximate the solution by linearly projecting high‐dimensional spaces to a lower‐dimensional space spanned reduced basis. However, accuracy of these methods may be insufficient complex and multiscale simulations due restriction linear space. Alternatively, autoencoders (AEs) can used nonlinear dimensionality reduction. We combine reduction techniques with time series prediction build data‐driven ROMs. The presented framework consists two‐level neural networks. is nonlinearly compressed in first level using encoder function AE. Subsequently, temporal evolution latent vector predicted second level. original easily reconstructed decoder In comparison projection‐based ROMs, example, proper orthogonal decomposition (POD) Galerkin projection, this allows naturally include parameters without interpolation demonstrate on two‐dimensional flow field simulation around circular bodies parameterized inlet fluid velocity.