Optimization of Upstream Offshore Oilfield Production Planning under Uncertainty and Downstream Crude Oil Scheduling at Refinery Front-End

In this work, we have attempted to solve two problems concerning the planning and
scheduling of crude oil operations: first, on the upstream production planning of
crude oil from offshore sources and second, on the scheduling of downstream
processing of crude oil at the refinery front-end.
The first part is on the offshore oilfield infrastructures planning under both
exogenous uncertainty and endogeneous decision-dependent uncertainty. A model
representative of the oilfield that is able to select the best routes to obtain the desired
objective function is considered. The methodology used is by firstly developing a
deterministic model andmodeling it with GAMS, followed by a stochastic one. The
results obtained show a high accuracy representation in which the uncertainties in
both the exogenous and endogeneous uncertainties in planning are accounted for.
The stochastic model is a more thorough representation of the problem because it
considers all the uncertainties along with the associated probabilities. Having
validated the model formulation and solution obtained with results for standard
problems reported in the literature, we believe that the model can be a tool to assist
upper-level management in preliminary decision-making on an optimal plan for
crude oil production from an offshore operation.
The second part is onthe scheduling of crude oiloperations at a refinery front-end. A
technique for obtaining globally optimal schedules for the flow of crude is
developed. Acontinuous time model based on transfer events is used to represent the
scheduling problem and this model is a nonconvex MINLP model which presents
multiple local optima. We implement a branch-and-contract algorithm that aims at
reducing the size of the search region. In order to obtain a global optimum solution
of the problem, an outer-approximation algorithm is proposed, whereby lower and
upper bounds on the global optimum are generated, which are converged to a
specified tolerance. The solution obtained from the LB-MILP model, i.e., the
decision variables (binary variables), was used to obtain a feasible solution for model
UB-NLP. This solution is the upper bound solution. The application of the proposed
algorithm shows significant reduction in the computational effort involved in solving the problem. Slack variables are introduced to overcome the integer infeasibility
problem. The optimization model is developed using GAMS and an optimal solution
is found with no logical constraints conflicts or error.
The main contribution on this work in the first part is to conduct an extensive study
onthe implementation ofthe model formulation in Iyer et al. (1998). As well, in the
second part, we are focused on investigating effective implementation strategies of
the model formulation and solution strategy in Karuppiah et al. (2008) using our
choice of the modeling platform GAMS and the best numerical solvers that are
available. Hence, most of the exposition on the model formulation and solution
algorithms are taken directly from the original papers so as to provide the readers
with the most accurate information possible.
VI