Petroleum refining processes might be considered on different levels. Usually feedstock for refining unit includes
variety of components which yields to multi-product outflow containing desired and undesired components.
Commonly units operate under high temperature and pressure, which significantly contributes to cost of final
product and impose restrictions to process operation parameters. Majority of processes utilize heterogeneous
catalysts, which are also sensitive to process conditions. Nevertheless, typical refining process consists of many
auxiliary units, the performances of which can significantly affect optimal conditions. All of these increases
complexity of refining process.
In such scenario, application of multi-objective optimization becomes essential for improvement of unit
performances. To carry out MOO of refining process it is vital to formulate real-life objectives in addition to
implementation of relevant constraints.
Critical literature review showed a growing interest for use of genetic algorithms in multi-objective optimization
of petroleum refining processes in the last several years.
In Bhutani et al.10 authors performed a multi-objective optimization of an industrial hydrocracking unit, which is
used to process heavy distillates to valuable products in presence of hydrogen. The unit considered mainly consisted
of two reactors in series – a hydrotreater (HT) and a hydrocracker (HC). Both are packed bed reactors with 2 and 4
beds respectively.
Authors utilized a simplified model for HT and first-principle model for HC. Reaction products were lumped into
8 components (e.g. liquefied petroleum gas (LPG), light naphtha, heavy naphtha, etc.). Kinetic scheme were
developed for these pseudo components. The HC modeling was based on the following assumptions: plug-flow
reactor without axial diffusion, adiabatic, steady state operation.
Objectives were chosen based on industrial priorities and they were applied to maximize diesel, kerosene and
naphtha production and to minimize off-gases, LPG production and hydrogen consumption. Due to many important
objectives MOO problem was divided into 3 two-objective cases for simplicity: