Evaluation Method of Total Organic Carbon Content in Shale Based on Stacking Algorithm Ensemble Learning
Total organic carbon content(TOC)is an important parameter for shale oil reservoir evaluation.However,traditional log evaluation methods for TOC have low accuracy and poor universality.Machine learning models have improved the prediction accuracy of TOC to a certain extent,but the results are unstable.In order to further improve the prediction accuracy of total organic carbon content in shale oil reservoirs,based on the physical characteristics of organic matter rocks and the logging response characteristics of different total organic carbon content,deep lateral resistivity,acoustic time difference,compensated neutron,and density logging curves are selected as sensitive logging responses for total organic carbon content.These are used as input features,and the total organic carbon content in core analysis is used as the expected output value.Decision tree models,support vector regression machine models,and BP(Back Propagation)neural network models are established,and Stacking algorithm ensemble learning models are established based on decision trees and support vector regression machines as meta models.The effectiveness of different models in predicting total organic carbon content is validated using core sample data from block B of oilfield A and actual well data.The results show that the ensemble learning model based on Stacking algorithm has the highest prediction accuracy for total organic carbon content compared with decision tree model,support vector regression machine model,BP neural network model,and improved ΔlgR method.Therefore,the ensemble learning model based on Stacking algorithm is the most effective method for calculating the total organic carbon content in the study area,which lays the foundation for accurately evaluating the hydrocarbon generation potential of shale oil reservoirs,ensuring efficient exploitation and resource utilization of shale oil reservoirs.
万方数据
维普
