Research on lithology identification based on two-stage genetic algorithm

Abstract

A two-stage genetic evolution algorithm based on gene fragments optimization (GF-GA) is proposed, which optimizes multi-parameter weights in logging lithology identification. First, gene fragments composed of multivariate weights corresponding to certain category are preselected to build the gene fragment library; then gene recombination and genetic evolution are carried out on the basis of the gene fragment library. GF-GA greatly reduces the search space redundancy of the genetic algorithm and improves the optimization efficiency by retaining the genetic diversity. The experiments of the KNN and fuzzy recognition methods of multivariable weighting in the lithology identification of complex carbonate rocks show that the GF-GA outperforms the traditional genetic algorithm (GA) in accuracy and time-consuming. Specifically, the GF-GA weighted model has an 2.5% to 3.5% accuracy improvement compared with the traditional GA weighted model, and the convergence speed is doubled. Moreover, compared with the conventional equal weight model, the accuracy of lithology identification is increased by 10% to 20%. Therefore, GF-GA is an optimization model with wide prospects in application.