基于机器学习智能决策系统的古龙页岩油储层总有机碳含量定量表征及智能预测

Quantitative characterization and intelligent prediction of TOC of Gulong shale oil reservoir based on machine learning intelligent decision system

总有机碳含量(w(TOC))是评价页岩储层含油性的一个重要参数,w(TOC)的定量表征与智能预测对于页岩油地质甜点标准确定、储量评估和开发方案设计具有重要作用。将岩心自动归位技术、基于相关系数的层次聚类算法和机器学习自动化技术融合为机器学习智能决策系统,可以解决w(TOC)智能预测的不确定性问题。该系统已经应用于古龙页岩油储层w(TOC)参数的定量表征及智能预测。结果表明:(1)将滑动窗口法岩心归位技术应用于井A2,确定4.250 m为岩心归位最优距离,归位后w(TOC)和声波时差(Δt)的相关系数从0.06提高到0.55;(2)应用基于相关系数的层次聚类算法挖掘出w(TOC)与声波时差(Δt)、密度(ρ_(DEN))和中子孔隙度(φ_(N))具有明显的相关性,与自然伽马(q_(AOI))、浅侧向电阻率(R_(LIS))、微球形聚焦电阻率(R_(MSFL))和深侧向电阻率(R_(MSFL))不具有明显相关性;(3)基于粒子群机器学习自动优化技术从6种回归算法中优选出随机森林算法及附带组合参数将古龙页岩油储层w(TOC)参数预测精度提高到81.7%。机器学习智能决策系统可以系统化降低古龙页岩油储层w(TOC)参数预测的不确定性,明显提高预测精度和计算效率。

Total organic carbon content(w(TOC))is one of important parameters for evaluating content of shale oil reservoirs,and study of quantitative characterization and intelligent prediction of TOC plays an important role in determining geological sweet spot criteria for shale oil,reserves calculation and development plan design.Automatic core location technology,hierarchical clustering algorithm based on correlation coefficient and machine learning automation technology are integrated into a machine learning intelligent decision system to solve the uncertainty problem of intelligent prediction of w(TOC)parameters.The systematic approach is applied in quantitative characterization and intelligent prediction of TOC parameters of shale oil reservoirs in Qingshankou Formation in Gulong area in Daqing Oilfield.The results show that:(1)Core homing technique by sliding window method determines optimal distance of 4.250 m core homing,and correlation coefficient between w(TOC)and acoustic time difference(Δt)can be improved from 0.06 to 0.55.(2)Hierarchical clustering algorithm based on correlation coefficient mines that TOC parameters have obvious correlation with acoustic time difference(Δt),density(ρ_(DEN))and neutron porosity(φ_(N))logs,but not with natural gamma(q_(API)),shallow lateral resistivity(R_(LIS)),microspherical focused resistivity(R_(MSFL))and deep lateral resistivity(R_(LLD)).(3)Random forest algorithm and accompanying combination of parameters from 6 regression algorithms based on particle swarm machine learning automatic optimization techniques improve the prediction accuracy of Gulong shale oil w(TOC)parameters to 81.7%.The machine learning intelligent decision system can systematically reduce prediction uncertainty of w(TOC)parameters of Gulong Shale oil reservoir and significantly improve prediction accuracy and computational efficiency.