I certify that I have read this report and that in my opinion it is fully adequate, in scope and in quality, as partial fulfillment of the degree of Master of Science in Petroleum Engineering.

No abstract

TX 75083-3836, U.S.A., fax 01-972-952-9435. AbstractDetermining the optimum location of the wells is a critical and crucial decision to be made during a field development plan. The quality of the decision is strongly dependent upon the amount of the information available to the decision-maker at the time the decision is made. Knowing that the development phase of a reservoir is a dynamic period in which different categories of information are added to system from distinct sources, one should make the well placement decisions considering these time-dependent contributions of information. This study proposes an approach that addresses the value of time-dependent information to achieve better decisions in terms of reduced uncertainty and increased probable Net Present Value (NPV). A Hybrid Genetic Algorithm 1 (HGA) was used as the optimization method to find the best locations of the wells. A utility framework, that enables the assessment of the uncertainty of the wellplacement decisions, was used to find the optimum decisions for different risk attitudes. Through this new approach, production history data obtained from the wells, as they are drilled, are integrated into the well placement decisions. Unlike previous approaches, well placement optimization is coupled with recursive history matching steps. To test the results of the proposed approach, an example reservoir was investigated with multiple realizations, all of which match the history response of the reference. At each step of optimization, a reduction in the uncertainty of the multiple realizations was observed, as production history became available. Problem StatementMotivation. Well placement decisions made during the early stage of exploration and development activities have significant impacts on the future recovery and profitability of the project. In addition, these early decisions have the capability to improve the later placement decisions by providing more information (greater certainty). Therefore, recovery and efficient use of information add value beyond the amount oil recovered. In this respect, the quality of the placement decision is dependent upon the amount, quality and efficient use of the information at the time of decision. Production data are a component the information available.Knowing that the reservoir development phase is a dynamic period in which different types of data are added to the system through time, one should make time-dependent (the subsequent well placement) decisions considering this timedependent information flow.

This paper presents an integrated production model construction and forecasting workflow along with three practical real field applications from the Jack asset located in deepwater Gulf of Mexico. Integrated production modeling is a composite modeling strategy that couples subsurface (material balance or simulation) models to a surface network model via well-bore models. The objective of using an integrated production model (IPM) is to predict the reservoir performance, while honoring mechanical design constraints of the surface network. The integrated production model construction process consists of five steps, which are framing, modeling, static quality check, initialization, and dynamic quality check followed by forecasting. An IPM was built for Jack and used as the primary forecasting method forevaluation of artificial lift alternatives (gas lift, sea floor boosting, and electric submersible pumps)identifying key artificial lift design parameters using Experimental Design, andsupporting injection facility design. In all of the studies, a black oil flow simulation model was used as the subsurface model and coupled to a surface network model. Background Deepwater field developments bring unique challenges for engineering design and project economics. A field development plan consist of various focus (surface and subsurface) decisions which require multiple engineering disciplines to work on their sub-domains concurrently (Figure 1). Reservoir engineers focus on well count, placement, and recovery mechanism decisions whereas production/completion engineers seek out feasible artificial lift alternatives and well design. At the same time, facility/subsea engineers study the subsea lay-out, facility size, and topsides design. Although these surface and subsurface decisions appear separable, they are coupled via common design limitations. A reliable production/ economics forecast will rely on the physical limitations of the overall system. This motivates project teams to use a common forecasting platform to support inter-related field development decisions.

Decisions made during the reservoir development phase have major impact on the project economics. Due to reservoir complexity and severe nonlinear interactions between decision parameters, automated optimization algorithms may be needed to supplement intuitive reservoir engineering analysis. Previous algorithms, proposed for well placement optimization, search for optimum well locations without full consideration of engineering knowledge and practice. This may result in degradation of optimization efficiency (number of simulations) and output unrealistic suboptimal well configurations. This study proposes an approach that superimposes engineering knowledge and practice on an automated optimization routine (the Hybrid Genetic Algorithm, HGA) to maximize the value and efficiency of field development optimization. The new approach is called the Fixed Pattern Approach (FPA), since it enforces the HGA to locate thewells on a user defined pattern inspiring from the idea that the wells are more likely to be drilled in specific patterns (line, staggered-line, five-spot etc.). The FPA essentially incorporates reservoir engineering concepts within the parameterization of optimization problem and reduces the search-space size considerably by eliminating the simulation of infeasible development scenarios. The FPA was applied to a deepwater field to get a quick understanding about the possible optimum development scenarios. In the scope of this study, finding the optimum development plan was defined as determining the optimum number and location of producers and injectors, as well as production rates and the starting time of injection that maximizes the Net Present Value (NPV) of the project. These parameters were optimized simultaneously using a black oil reservoir simulator as the objective function evaluator. Comparison with respect to the conventional window approach showed that FPA significantly reduced the CPU time and resulted in practical and economical well locations. Uncertainty in the geological properties (porosity, permeability) and reservoir structure was also addressed for this field by performing optimization on five different models constructed by a Plackett-Burman experimental design. The new methodology recommended an optimum field development plan while taking the uncertainty of property distributions and reservoir structure into account. Introduction Genetic Algorithms (GAs) have been used as global search engines in various optimization problems[1, 2] since they have been introduced by Holland[3] (1975). Hybridization of GAs with the hill-climbing methods[4] (simplex) improved GAs' local search capability. Aanonsen et al.[5] (1995) optimized the well locations using several regression techniques and showed that simple statistical methods could be adequate to find optimal solutions in some cases. Bittencourt and Horne[6] (1997) showed that optimization efficiency of GAs were improved when GAs were hybridized with the polytope algorithm and tabu search. They called this hybrid optimization technique the Hybrid Genetic Algorithm (HGA). Guyaguler et al.[7] (2000) minimized the number of simulations of an injector location optimization by coupling a HGA with a kriging proxy. Yeten et al.[8] (2002) optimized the trajectory of non-conventional wells by combining GA with hill-climbing methods and an Artificial Neural Network (ANN) proxy. Ozdogan and Horne[9] (2004) optimized the recovery of oil and information from the wells by embedding recursive probabilistic history matching steps into a HGA. They showed that early stage well locations could be optimized in order to reduce the uncertainty associated with the later stage wells. Previous approaches utilized various types of proxies to minimize the computational effort but none of them focused on the geometrical limitations of the well placement problem.