A log-based lithofacies identification method based on random forest and sedimentary microfacies characteristics:A case study of tight sandstones in the second member of the Xujiahe Formation in the Xinchang area
Tight sandstones serve as significant oil and gas reservoirs.Their lithofacies identification can assist in further understanding the developmental characteristics of reservoirs.Combining core observations with log data processing,this study analyzed the lithofacies and sedimentary microfacies characteristics of tight sandstones in the Xinchang area and the internal relationships between lithofacies and sedimentary microfacies.Moreover,it constructed a random forest classification model with geological implications through data min-ing of sedimentary microfacies characteristics.The results show that:(1)Tight sandstones in the Xinchang area can be classified into seven typical lithofacies,including mudstone,siltstone with ripple lamination,massive fine sandstone,fine sandstone with parallel bed-ding,massive medium-to coarse-grained sandstone,and medium-to coarse-grained sandstone with parallel/cross bedding;(2)The sedi-mentary microfacies in the Xinchang area consist primarily of subaqueous distributary channel,subaqueous distributary bay,river-mouth bar,and prodeltaic mud,which are closely associated with the sedimentation of lithofacies;(3)In the classification model,the relative centroid(RM),root mean square deviation(GS),average median(AM),and average slope(M)of the gamma ray(GR)curve can be used as the characteristic parameters of sedimentary microfacies to increase the number of characteristics in the dataset;(4)Considering the characteristics of sedimentary microfacies,especially the energy and turbulence of water bodies,can significantly enhance the per-formance of the random forest classification model.Overall,the results of this study provide a novel approach for lithofacies identification using machine learning methods and a significant reference for oil and gas exploration in tight sandstones.
国家科技期刊平台
NSTL
万方数据
