This paper is centered in the optimization of well placement using Genetic Algorithms. A Simple Genetic Algorithm program has been developed and it has been used to optimize well placement in two case studies. The effects of different internal parameters on the algorithm performance have been analyzed and a base case configuration suggested. Future modifications are analyzed.
The objective of obtaining the maximum profit from investments in oil fields, is everyday more demanding. With easy fields being more scarce every day, new costly frontiers being developed, and many of the world producing provinces reaching maturity (North sea, western Texas, Gulf coast), the need to develop new tools, which allow the reservoir engineer to optimize reservoir performance, is becoming more important every day.
The problem of optimizing an oil field is an extraordinarily complex one. In it there are many variables to consider, geological variables like reservoir architecture, production variables such as well placement, well number, type of platform, platform position, etc, and monetary variables like oil and gas prices. All these variables, together with reservoir geological uncertainty, make difficult the determination of the objective function and its restrictions.
In these conditions, numerical simulators are required to evaluate the objective function. The lack of analytical solutions in most cases makes traditional function optimization impossible. Furthermore the non-linearity and non-continuity of oil field optimization problem, limit traditional methods like, simplex or gradient, Udias1, Jefferys9 and Goldberg2.
The use of numerical multiphase flow simulators allows the reservoir engineer to assess the behavior of the reservoir under different scenarios. However the setup of each scenario takes time and the sheer number of possible alternatives makes this task lengthy and costly. In order to automate this process, several Artificial Intelligence methods like Neural Networks and Genetic Algorithms have been developed. Genetic Algorithms use Darwin's survival of the fittest theory to evaluate different scenarios, which evolve to an optimum solution.
Genetic Algorithms are search algorithms based on the mechanics of natural selection. They combine survival of the fittest with a structured yet stochastic exchange of information to increase the efficiency of an otherwise purely random search. These methods were first used in the sixties by biologists to simulate evolution, based mainly on mutation patterns. The first attempt to use Genetic Algorithms to optimize complex problems was in the seventies by John Holland, in which the basic theory was formulated demonstrating the ability of bit chains to represent complex problems, and the capacity of simple transformations to improve those chains. In his work Holland13, demonstrated that it was possible to find an optimal individual evaluating only a very small fraction of the population (900 individuals out of a population of 2.7*1011, in his studies).
It is useful to comment on the main differences between genetic algorithms and traditional methods:
Direct manipulation of a coding
Search from a population and not from a point
Search via sampling, blind search.
Search using stochastic operators.
All these differences make Genetic Algorithm able to overcome many of the limitations of the traditional methods, especially the continuity and derivability of the objective function.