|dc.description.abstract||This thesis considers the problem of residue upgrading operations in an oil
refinery. Visbreaking is a residue-upgrading process that improves profitability
of a refinery. The economics of converting the heavy residue into the lighter and
more valuable streams, coupled with the installation of a modem visbreaker unit
at the Engen Refinery in Durban, provides sufficient motives to develop a
mathematical model to simulate the unit's capability and estimate the economics
of the visbreaking process and fuel oil operations. Furthermore, the proposed
model should provide a crude-dependent visbreaking yield that can be used in
the refinery's global linear programme (LP), employed to evaluate and select the
crude and to optimise refinery's operations.
Traditionally, kinetically based models have been used to simulate and
study the refining reaction processes. In this case, due to the complexity of the
process and some unknown reactions, the performances of existing visbreaking
simulators are not fully satisfactory. Consequently, a neural network model of
the visbreaking process and fuel oil blending operation is developed.
The proposed model is called the adaptive visbreaker paradigm, since it is
formed using neuroengineering, a technique that fabricates empirically-based
neural network models. The network operates in supervised mode to predict the
visbreaking yields and the residue quality.
It was observed that due to the fluctuation in the quality of feedstock, and
plant operating conditions, the prediction accuracy of the model needs to be
improved. To improve the system's predictability, a network reciprocation
procedure has been devised. Network reciprocation is a mechanism that controls
and selects the input data used in the training of a neural network system.
Implementation of the proposed procedure results in a considerable
improvement in the performance ofthe network.
To facilitate the interaction between the simulation and optimisation
routines, an integrated system to incorporate the fuel oil blending with the
neurally-based module is constructed. Under an integrated system, the
economics of altering the models' decision variables can be monitored.
To account for the visbreakability of the various petroleum crudes, the yield
predicted by the adaptive visbreaker paradigm should enter into the visbreaker,s
sub-model of the global refinery LP. To achieve this, a mechanism to calculate
and update the visbreaking yields of various crude oils is also developed.
The computational results produced by the adaptive visbreaker paradigm
prove that the economics of the visbreaking process is a multi-dimensional
variable, greatly influenced by the feed quality and the unit's operating
condition. The results presented show the feasibility of applying the proposed
model to predict the cracking reaction yields. Furthermore, the model allows a
dynamic monitoring of the residue properties as applicable to fuel oil blending
In summary, the combination of the proposed models forms an integrated
decision support system suitable for studying the visbreaking and associated
operations, and to provide a visbreaking yield pattern that can be incorporated
into the global refinery LP model. Using an integrated decision support system,
refinery planners are able to see through the complex interactions between
business and the manufacturing process by performing predictive studies using