Application of Ensemble-based Optimization on UNISIM-I
MetadataShow full item record
In reservoir management, it is challenging to obtain an efficient production schedule and maximize the profits. An optimization workflow is usually used in maximizing/minimizing the production objective. However, production optimization is not an easy task and could be time-consuming since the reservoir and the production optimization itself consist of complex subsystems and uncertainties. Thus, many studies have been done to propose optimization methods that are efficient and yet practical in finding the optimal strategy. Most of these methods usually focus on the gradient-based approaches, where the information from gradients of the objective function with respect to control parameters is used in finding the optimal solutions. One of the gradient-based methods that recently has gained popularity in petroleum production optimization is Ensemble-based Optimization (EnOpt). In EnOpt, the gradient is approximated using a linear regression between an ensemble of control vectors and their corresponding objective function values. Thus, the computational cost of the method relies on the number of realizations in the control ensemble and is nearly independent of the number of control parameters. Moreover, the EnOpt can be used with any reservoir simulator without modification to the simulator. Many publications have demonstrated that EnOpt gave a good optimized-result on different reservoir models and recovery techniques. In this thesis, we study the benefits of the EnOpt applied to waterflooding optimization problems using realistic reservoir data. In particular, the EnOpt is used to optimize the waterflooding process in a benchmark reservoir, namely UNISIM-I. The objective of this optimization is to maximize the expected net present value (NPV) over 20 years of production. The control parameters are injection and production rates in injector and producer wells. We consider two optimization problems: random initial control settings and extended production from the production history. The EnOpt was successful in finding optimal solutions in both cases with significantly cheaper computational cost required in gradient calculations. In addition, we study the effect of discount rate use in calculating the NPV: the short-term EnOpt uses high discount rate, whereas the long-term EnOpt sets discount rate equal to zero. The different discount rates result in different optimal solutions. The high discount rate results in an increase of cash flow in the early stages of the production time while low to no discount rate results in an increase of cumulative cash flow throughout the production time.
Plukmonton, Pattanapong (2017). Application of Ensemble-based Optimization on UNISIM-I. Master's thesis, Texas A & M University. Available electronically from