Abstract:Our objective is to devise an efficient optimization algorithm for estimating optimal well controls during the production optimization step of closed-loop reservoir management where the optimization algorithm does not require an adjoint method for the calculation of derivatives. Research within the TUPREP group strongly suggests that gradient-based algorithms will be the most efficient and reliable methods for estimating well controls to maximize the net present value (NPV), or cumulative oil recovery of an oil field. However, most major oil companies use commercial reservoir simulators to make future predictions of production, and such commercial simulators generally lack efficient gradient-based production optimization algorithms where computational efficiency demands that the gradient of the cost function (e.g., NPV or cumulative oil production over the life of the field) be computed by an adjoint method. Adding an adjoint solution without access to the source code is not feasible. Therefore, it is important to implement optimization algorithms that can be easily coupled with a commercial simulator simply by using the simulator as a black box. While many such gradient-free optimization algorithms are available, based on the excellent performance of gradient-based methods, it is desirable to develop a procedure that mimics the property of methods that use a gradient. One such method is the simultaneous perturbation stochastic approximation (SPSA) method which estimates a stochastic gradient by a procedure that is far more efficient than estimating the derivative with a finite difference approximation. The purpose of this study is to develop a computationally efficient version of the SPSA algorithm for the constrained production optimization problem involved in the optimal control step of closed-loop reservoir management. The SPSA is a stochastic optimization algorithm, which obtains its stochastic gradient by simultaneously perturbing all components of the well control vector in a stochastic way. Although the SPSA gradient is stochastic, it is always uphill for a sufficiently small perturbation size and its expectation is equal to the true gradient as the perturbation size goes to zero. The most challenging problem in production optimization is to honor state-control constraints which are nonlinear. The SPSA algorithm has been applied to production optimization problems with only bound constraints and implementations used in previous published work were computationally inefficient. In this study, the focus will be developing and implementing a computationally efficient SPSA algorithm for production optimization problems with general equality, inequality and bound constraints. Here, the bound constraints are converted into the new unbounded control variables using a log-transformation. Although the results are not shown in this work, simply truncating control variables at their bounds consistently yields slightly lower NPVs values than are obtained with the log-transform method used here. All constraints are incorporated into the objective function to be maximized using the augmented Lagrangian function. The advantage of this method is that the general inequality and equality constraints can conceptually be handled using suitable values of the penalty parameter and Lagrange multipliers. In addition to the basic SPSA algorithm, which uses samples from the symmetric Bernoulli distribution as perturbations, a SPSA type algorithm that uses samples from Gaussian distribution is also tested. To improve the efficiency of the SPSA algorithm, a combined search direction SPSA algorithm is proposed, in which the search direction is a combination of the current SPSA gradient and the "best" SPSA gradient or some "good" former SPSA gradients calculated in the last few iterations.

Comparison of SQP and AL algorithms for deterministic constrained production optimization of hydrocarbon reservoirs

Quantifying the Robustness of Optimal Reservoir Operation for the Xinanjiang-Fuchunjiang Reservoir Cascade

A GLOBALLY AND SUPERLIEARLY CONVERGENT FEASIBLE SQP ALGORITHM FOR DEGENERATE CONSTRAINED OPTIMIZATION

Optimization of refinery hydrogen network based on chance constrained programming

A Comparison of Arithmetic Operations for Dynamic Process Optimization Approach

Application of spsa-type algorithms to production optimization

Robust Constrained Optimization of Short- and Long-Term Net Present Value for Closed-Loop Reservoir Management

Maximization of A Dynamic Quadratic Interpolation Model for Production Optimization

A method for developing and calibrating optimization techniques for oil production management strategy applications

Production Optimization in Closed-Loop Reservoir Management

A comparative study of SQP-type algorithms for nonlinear and nonconvex mixed-integer optimization

A QCQP-based splitting SQP algorithm for two-block nonconvex constrained optimization problems with application

Operation Optimization of Distillation Column Based on Reduced SQP Algorithm and Hybrid Derivative Method

Design and Implementation of Large-scale Nonlinear Optimization Software UniOptima

A Sequential-Quadratic-Programming-Filter Algorithm with a Modified Stochastic Gradient for Robust Life-Cycle Optimization Problems with Nonlinear State Constraints

Improved SQP and SLSQP Algorithms for Feasible Path-based Process Optimisation

Optimization of reaction parameters based on rSQP and hybrid automatic differentiation algorithm

On the Performance of SQP Methods for Nonlinear Optimization

A simultaneous perturbation stochastic approximation algorithm for coupled well placement and control optimization under geologic uncertainty

Uncertainty Quantification of Reservoir Performance Predictions Using A Stochastic Optimization Algorithm