基于Stacking算法集成学习的页岩油储层总有机碳含量评价方法

Evaluation Method of Total Organic Carbon Content in Shale Based on Stacking Algorithm Ensemble Learning

总有机碳含量(TOC)是页岩油储层评价的重要参数,而传统总有机碳含量测井评价方法精度较低且普适性较差,机器学习模型在一定程度上提高了总有机碳含量预测精度,但结果存在不稳定性。为了进一步提高页岩油储层总有机碳含量预测精度,基于有机质岩石物理特征和不同总有机碳含量测井响应特征的深入分析,优选出深侧向电阻率、声波时差、补偿中子和密度测井曲线作为总有机碳含量的敏感测井响应,并将其作为输入特征,以岩心分析总有机碳含量作为期望输出值,分别建立了决策树模型、支持向量回归机模型、BP(Back Propagation)神经网络模型,并建立了以决策树模型为基模型、支持向量回归机模型为元模型的Stacking算法集成学习模型。利用B油田A区块的岩心样本数据和实际井数据对不同模型预测总有机碳含量结果进行了验证,结果表明,基于Stacking算法的集成学习模型的总有机碳含量预测精度最高,相较于决策树模型、支持向量回归机模型、BP神经网络模型和改进的ΔlgR法,预测精度有较大提高。因此,基于Stacking算法的集成学习模型为该研究区最有效的总有机碳含量计算方法,这为准确地评估页岩油储层的生烃潜力、确保页岩油储层的高效开采及资源利用奠定了基础。

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 improvedΔlgR 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.