基于机器学习的总有机碳含量测井预测方法对比研究OACSTPCD

The lithology of the first member of the Maokou formation in the Middle Permian of Central Sichuan is complex and the longitudinal variation is large.There are many influencing factors and difficulty in the prediction of total organic carbon content,so it is particularly important to explore the most suitable high-precision prediction method for the prediction of total organic carbon content in this area.In this paper,the sensitive parameters for the prediction of total organic carbon content are selected based on the Pearson correlation coefficient matrix,and three machine learning methods are used to model the total organic carbon content.By using three machine learning methods:multi-layer feed forward neural network,support vector machine and XGBoost algorithm,the effect of calculating total organic carbon content is compared with the conventional Δlog R method.The results show that natural gamma,resistivity,acoustic time difference,compensated density and compensated neutron log data with high correlation with total organic carbon content are selected as input features by using the sensitive feature mining method.The Δlog R calculation model with R2=0.624 8,the BP neural network prediction model with R2=0.814 4,the support vector machine prediction model with R2=0.702 9 and the XGBoost prediction model with R2=0.937 0 were established.The analysis of the practical application effect showed that the conventional Δlog R method had poor accuracy in calculating the total organic carbon content and could not achieve the expected effect.The predicted value of the XGBoost prediction model is in the best agreement with the measured value and has the highest reliability.In this study,a prediction model of total organic carbon content with high prediction accuracy and strong generalization ability is established,compared and optimized,which provided an effective method for predicting the total organic carbon content of complex carbonate rocks in the study area.