Hledat publikace

Steam-cracking is very important industrial process which transforms wide range of hydrocarbon feedstocks to valued petrochemical commodities, such as ethylene, propylene and butadiene. Mathematical model of the process is very useful tool for industrial producers for many purposes. Continuous increasing variability of the feedstock enforces model creators to extend existing models by new components, under the time pressure. The creation of reaction scheme for a new component is tedious process requiring a lot of highly-qualified human resources and it is prone to human errors. But, in the case of steam-cracking, individual reactions can be generalized by several templates. Therefore, the reaction scheme and corresponding mathematical description can be generated automatically. Generated models usually contain hundreds of reactions even for such a simple component, such as n-heptane is. But for the modeling purposes, the knowledge of kinetic parameters is critical. In this paper, a method for kinetic parameters estimation for automatically generated steam-cracking model is proposed. It is based on the classification of reactions to several classes by their nature, such as hydrogen abstraction, β-scission, etc. A “template” for every reaction class is related to basic kinetic parameters. Inside these classes, the group contribution approach was applied. It uses various factors of frequency factor and increments of activation energies connected with structure differences from the “template” around the reaction center. Values of these generalized parameters were estimated based on available data about “typical” reactions and later were optimized using a set of experimental data measured in the micro-scale on a pyrolysis gas chromatograph. As more pyrolysis data were added to the dataset, model described products distribution very well for the cracking of normal and branched alkanes. First discrepancy was indicated when a data about subsided cycloalkanes were added into the dataset. Based on residual analysis, several improves were made in the estimation method and the optimization process continued. To check the generalization ability of the method, experimental data obtained on two different reactors were added to the dataset. Final model predicts the behavior of selected data well. Developed model provides certain possibility for “structural” generalization of the kinetic parameters for the pyrolysis which could be potentially utilized for the prediction of products distribution for unknown feedstock molecules.