Reservoir Development and Design Optimization

Optimization of reservoir development requires many evaluations of the possible combinations of the decision variables, such as the reservoir properties, well locations and production scheduling parameters, to obtain the best economical strategies. Running a simulator for such a large number of evaluations may be impractical due to the computation time involved. In this study, a hybrid Genetic Algorithm (GA) was developed. The optimization algorithm integrated economic analysis, simulation and project design. The layout of 33 new wells for a real oil field development project was proposed by the company’s project team. The algorithm was used to determine an optimal relocation of the wells, by evaluating an objective function from a cash flow analysis for the production profile obtained from simulation at each iteration. The wells were allowed to be placed anywhere in the reservoir and could be vertical or horizontal and, if horizontal, any direction in the same layer could be considered. A total of 99 decision variables were used to solve this problem. All the real restrictions were included. Based on the cost of the sea bottom flowlines a second level optimization found the best platform location. The results were compared against the proposed solution, and showed that the algorithm performed very well finding an optimized well distribution. A reduction of the total number of new wells was found as part of the solution. An improvement of about 6% in the project profit was found, representing about US$70 million additional income. Introduction The main task of a reservoir engineer is to develop a scheme to produce as much hydrocarbon as possible within economic and physical limits. The solution of this kind of problem encompasses two main entities: the field production system and the geological reservoir. Each of these entities presents a wide set of decision variables and the choice of their values is an optimization problem. In view of the large number of decision variables it is not feasible to try to enumerate all possible combinations. Analysis tools encoded in computer programs can spend hours or days of processing for a single run, depending on their sophistication and features. Also, it can be costly to prepare the input data if many hypotheses are going to be considered. A typical reservoir development involves many variables that affect the operational schedule involved in its management. These variables are usually used as input to a reservoir simulator that generates a forecast of the production profile. Using this forecast, the production engineer has to consider several hypotheses to achieve the best strategy for the field development. Also, each hypothesis can generate others, and so the overall process is one of generating a hypothesis tree. More and more data are generated and analyzed. The solution of these problems requires the effort of several people as well as considerable computer work and physical time. An optimization procedure requires the characterization of the function to be optimized (minimized or maximized), known as the objective function, as well as the choice of an appropriate optimizing method. The complexity of predicting hydrocarbon production profiles requires the use of reservoir simulators. So, the simulator must be part of the evaluation of the objective function. This work concerns the optimization of characteristic petroleum production problems considering economic factors. A hybrid algorithm based on direct methods such as Genetic Algorithm (GA), polytope search and Tabu search was developed. Hybrid techniques were found to improve the overall method. The objective function consisted of a cash flow analysis for production profiles obtained from simulation runs. The optimizing procedure was able to interface with commercial simulators (generating their input data and retrieving the results) that worked as data generators for the objective function evaluation. These hybrid mathematical approaches were found to be successful in obtaining the optimal solution with less time and work than existing techniques. These approaches can speed up the study of a hydrocarbon reservoir development plan and SPE 38895 Reservoir Development and Design Optimization Antonio C. Bittencourt, SPE, Petrobras, and Roland N. Horne, SPE, Stanford University 2 A.C.BITTENCOURT, R.N. HORNE SPE 38895 allow consideration of a wider range of hypotheses. The engineer can also keep track of economics during the study, allowing better project decisions. A real project was optimized using two approaches: the first one had the proposed solution inserted in the initial population and the second one did not. The first approach achieved the better solution, albeit at the cost of a larger number of simulations due to premature convergence earlier in the calculation. The Hybrid Algorithm Approach Optimization concerns the optimal solution determination using an oriented search towards the best possible value. Algorithms used in the optimization procedure are problem dependent, so it is necessary to investigate the strengths, weaknesses and ranges of applicability of each. One of the objectives of this work was to develop a hybrid algorithm to overcome the limitations of individual approaches, and to take advantage of the particular strengths of each. The principal procedure used in the hybrid was the Genetic Algorithm, combined with a polytope search between generations and an initial distribution based on the Fang algorithm. Brief descriptions of the component algorithms will be summarized here. Genetic Algorithm (GA) is a robust search method based on analogies to biology and genetics. Survival of the fittest among a population of individuals, selection criteria, and reproduction strategies are concepts copied from natural life and used as operators in this artificial environment. Applications have been found for GA in business, engineering and science. The GA has four advantageous features: 1. GA begins the search with a population of parameter realizations, rather than a single realization as most of the conventional optimization methods might. In this way, the search domain is covered in a random distribution. 2. The realizations are perturbed by probabilistic rules rather than deterministic ones. 3. The parameter itself is not manipulated directly by the GA operators. GA would alter the chromosome (or individual or string) that is a pattern of zeros and ones representing the whole set of parameters put all together in one binary entity. For binary alphabets, the smaller piece of a chromosome is called a bit. 4. Only function evaluations are used rather than derivatives or other secondary descriptors. The fitness is defined as