Efficient Ensemble-Based Closed-Loop Production Optimization

Abstract

With the advances in smart well technology, substantially higher oil recovery can be achieved by intelligently managing the operations in a closed-loop optimization framework. The closed-loop optimization consists of two parts: geological model updating and production optimization. Both of these parts require gradient information to minimize or maximize an objective function: squared data mismatch or the net present value (or other quantities depending on financial goals), respectively. Alternatively, an ensemble-based method can acquire the gradient information through the correlations provided by the ensemble. Computation of the optimal controls in this way is nearly independent of the number of control variables, reservoir simulator and simulation solver. In this paper, we propose an ensemble-based closed-loop optimization method that combines a novel ensemble-based optimization scheme (EnOpt) with the ensemble Kalman filter (EnKF). The EnKF has recently been found suitable for sequential data assimilation in large-scale nonlinear dynamics. It adjusts reservoir model variables to honor observations and propagates uncertainty in time. The EnOpt optimizes the expectation of the net present value based on the updated reservoir models. The proposed method is fairly robust, completely adjoint-free and can be readily used with any reservoir simulator. The ensemble-based closed-loop optimization method is illustrated with a waterflood example subject to uncertain reservoir description. Results are compared with other possible reservoir operation scenarios, such as, wells with no controls, reactive control, and optimization with known geology. The comparison shows that the ensemble-based closed-loop optimization is able to history match the main geological features and increase the net present value to a level comparable with the hypothetical case of optimizing based on known geology.

Optimisation of costs and carbon savings in relation to the economic dispatch problem as associated with power system operation

“…Even a few ensemble members have the ability to demonstrate the large-scale covariance behaviour of the system [John and Mandel, 2008]. EnKF has been widely used for assimilation of the real-time production data, providing both the estimate of a model and the corresponding uncertainty, by keeping track of the whole ensemble [Begum, 2009, Chen et al, 2009, Jahangiri, 2012.…”

“…In addition, a closed-loop based feedback optimisation control is proposed for continuous updates and optimisations in the ED problem. The EnOpt, a production optimisation model that combines the EnKF and optimisation that was initially developed for the SmartWell technologies in the oil industry [Chen et al, 2009], is applied in order to develop optimal operating strategies in optimising the ED problem.…”

“…Such data assimilation aims to minimise the mismatch between the model prediction and the production data [Chen et al, 2009]. The resultant assimilation data from EnKF is further utilised in the EnOpt, which estimates expectation of the objective function based on the assimilated data along with optimised control constraints [Chen et al, 2009, Jahangiri, 2012, Nwaozo, 2006.…”

“…Such data assimilation aims to minimise the mismatch between the model prediction and the production data [Chen et al, 2009]. The resultant assimilation data from EnKF is further utilised in the EnOpt, which estimates expectation of the objective function based on the assimilated data along with optimised control constraints [Chen et al, 2009, Jahangiri, 2012, Nwaozo, 2006. At the end of the EnOpt simulation, the best control settings are the optimal control variables that minimise the objective function [Chen et al, 2009, Jahangiri, 2012, Nwaozo, 2006.…”

Model-reduced gradient-based history matching

“…Another method uses an ensemble of control variables and computes the cross-covariance between the control variables and the objective function which can then be used as an approximate gradient. This "ensemble optimization" technique was proposed for recovery optimization by Lorentzen et al [28] and thereafter refined by Chen et al [10]. For history matching, a similar approximate gradient-based technique has gained enormous popularity over the past years in the form of the ensemble Kalman filter (EnKF); see [16] and [1] for recent overviews.…”