Empirical reduction of dynamical reactor models via chaos sampling: comparison with classic reduction methods

An empirical model reduction technique, yielding low-dimensional approximations of the full high- or infinite-dimensional dynamical reactive systems while retaining their essential features, is presented in this work. The first step of the procedure i.e., the sampling of the representative set of simulation data to be used for the construction of the empirical basis, is considered to be crucial for generating a global basis. Here the concept of information entropy is used to optimally select the parameter range for the sampling region. The resulting POD/Galerkin technique is successfully applied to the optimal reduction of the model of an autothermal tubular reactor with recycle. The empirical basis is constructed with data sampled in the highest entropy (chaotic) regime. As expected, the higher the entropy of the sampled orbits, the closer the approximation to the original model. POD/Galerkin is shown to perform better than the yet efficient, classic approach of orthogonal collocation on finite elements, especially in the early transient.