Transformation of Geochemical Log Data to Mineralogy Using Genetic Algorithms

J. H. Fang, C. L. Karr, D. A. Stanley

There is a pressing need for developing a technique for transforming geochemical data obtained with the newly developed geochemical log tool (GLT) into constituent minerals of formations. A novel technique of genetic algorithm (GA) is found to be both appropriate and efficaceous. GA is a global optimization method that searches for solutions based on an analogy between optimization and natural selection. In this approach, the problem is represented as binary strings of 0's and 1's. Initially, the population of solutions is generated randomly and at each subsequent iteration three probabilistic processes are applied. The first, reproduction, imposes a survival of the fittest criterion to select a new population of strings or solutions; the second, crossover,</ M> produces an efficient exchange of information between surviving solutions; the third, mutation, introduces a purely random element that maintains diversity in the new population. Compared to the traditional approaches of least squares and linear programming, the genetic algorithm does not require the calculation of partial derivatives and produces the same or better results. The method is illustrated with two examples--a sandstone and a volcanic rock.