基于改进随机森林的火山岩测井岩性识别.

The lithology of carboniferous volcanic rocks in Junggar Basin is complicated. It is difficult to divide lithology by conventional methods when the number of thin sections and cores of one lithology is obviously less than that of other lithology. In order to solve the above problems and improve the accuracy of volcanic rock lithology identification, synthetic minority oversampling technique (SMOTE) algorithm was used to increase the number of samples of a few lithologic categories to solve the problem of data imbalance. The optimal parameter combination was determined by grid search and K-fold cross validation method, and the lithologic identification of volcanic rocks based on improved random forest was carried out. By analyzing the data of volcanic rock core, thin section and logging response characteristics, the lithologic crossplot was established, and the importance of logging parameters sensitive to lithology in the study area was determined. The application of case data shows that the improved random forest algorithm effectively solves the influence of the traditional random forest algorithm due to the unbalance of lithological sample types and the small amount of data, and the accuracy rate of volcanic rock lithology identification increases from 87% to 94%, which provides reference for the lithology identification of volcanic rock in the case of unbalanced samples. [ABSTRACT FROM AUTHOR]