In the present PhD thesis, a novel methodology for the kinetic modelling of heavy oil refining processes is developed. The methodology models both the feedstock composition and the process reactions at a molecular level. The composition modelling consists of generating a set of molecules whose properties are close to those obtained from the process feedstock analyses. The set of molecules is generated by a two-step molecular reconstruction algorithm. In the first step, an equimolar set of molecules is built by assembling structural blocks in a stochastic manner. In the second step, the mole fractions of the molecules are adjusted by maximizing an information entropy criterion. The refining process is then simulated by applying, step by step, its main reactions to the set of molecules, by a Monte Carlo method. This methodology has been applied to two refining processes: The hydrotreating (HDT) of Light Cycle Oil (LCO) gas oils and the hydroconversion of vacuum residues (VR). For the HDT of LCO gas oils, the overall properties of the effluent are well predicted. The methodology is also able to predict molecular properties of the effluent that are not accessible from traditional kinetic models. For the hydroconversion of VR, which have more complex molecules than LCO gas oils, the conversion of heavy fractions is correctly predicted. However, the results for the composition of lighter fractions and the desulfurization yield are less accurate. To improve them, one must on one hand include new ring opening reactions and on the other hand refine the feedstock representation by using additional molecular information from the analyses of the process effluents.
