Abstract
Reservoir waterflooding is one of the cheapest means of producing hydrocarbon from
underground formation to the surface. A properly formulated control and optimization strategy
will not only solve the process inevitable problems but will also lead the process to optimal
operation. Previous optimization studies are model-based, but reservoirs are highly complex,
and therefore cannot be described and predicted accurately using models. To counteract the
effects of reservoir model/system mismatch, feedback control was suggested to be included in
the optimization framework. In this work the principle of self-optimizing control (SOC) is used
to derive controlled variable (CV) based on synthetic data. We have previously implemented
this methodology on a very small reservoir. The present work extends the implementation to a
realistically sized reservoir. In the methodology, the CV is formulated via a single regression
step in which a measurement function is used to approximate the gradient of the objective
function with respect to control. The developed CV is firstly implemented on a nominal model
and then to various uncertain cases. The performance of the method is compared to that of
open-loop solution technique, OC (based on optimal control theory) and then to a benchmark
case. The developed CV is found to be robust in the presence of uncertainties. In one of the
cases considered, the SOC method is found to be better than OC solution procedure by about
24.03%.