Before lattice codes output few group nodal cross sections, they first have to generate multigroup cross section libraries from continuous energy cross section libraries for each material in each fuel cell. The computational cost of these cross section processing are large. Therefore, we propose the use of a representation model to estimate lattice code multigroup cross section libraries from nuclide concentration data and state parameters. The model combines Principal Component Analysis (PCA) and fully connected Neural Networks (NN). The model contains three sublayers in a PCA-NN-PCA configuration. The model is designed to handle large multi-group cross section data sets containing cross section data for several dozen nuclides and containing upwards of 50 energy groups. As proof of concept, our proposed method is trained on lattice code cross section data for the fuel pellet material in a typical light water reactor assembly. The tested 56 energy group cross section libraries contain microscopic cross sections for 37 nuclides and the fuel pellet macroscopic cross sections.