Optimizing oil recovery subject to operational constraints requires accurate reservoir models with high predictive capabilities. Specifically, during the immiscible injection processes such as water flooding or gas injection, changes occurring in the flow directions and proportions can affect the long term recovery factors dramatically. Monitoring such changes, as reflected in the rate and pressure measurements, in real time, can provide us with information that can prevent long term losses and help in optimal oil recovery. Such a monitoring platform is tied to a predictive model and should incorporates historic and the incoming data as it becomes available. In this dissertation, we used a continuous pressure interference testing to characterize a 1-D reservoir and subsequently obtained the permeability profile and were able to link the fluid front movement to the pressure difference variations across the reservoir. To compensate for the typical scarcity of pressure data as opposed to the abundance of rate data we introduced a novel methodology for efficient characterization, prediction, and optimization of large water-flooding operations. This is based on ensemble based closed-loop production optimization (EnOpt) and capacitance resistive model (CRM) as its linear underlying dynamical system where the injection rates are the driving force or input signal and the production rates are the dynamical variables or output signals. The production rate data are assimilated in real-time by an ensemble Kalman filter for characterization of the reservoir. Simultaneously, the most up-to-date characterization of the reservoir produces optimal values to set the injection well rates to maximize the net present value of the reservoir. Basing the EnOpt method on CRM as opposed to reservoir simulation is computationally efficient even if limited geological data is available from numerous operating wells. Furthermore, nonlinear effects associated with saturation movements (ignored in previous publish works) can be incorporated through evolving reservoir parameters. Synthetic and real field examples are used to demonstrate how EnOpt/CRM can match and predict oil and water production rates to probe successes and limitations of our methodology in terms of the reliability of characterization, improvement in optimization, and sensitivity to the choice of starting parameters.
展开▼