Research On Lithology Identification Method Of Volcanic Clastic Sand Conglomerate Based On Deep Learning

In recent years,a number of wells in the WN area have been tested in the volcanic clastic sand conglomerate reservoir and obtained industrial oil flow,demonstrating that the volcanic clastic sand conglomerate reservoir in the area has a large exploration potential.Based on the analysis of the logging response characteristics of different conglomerate parent rock compositions,it is found that the logging response characteristics of volcanic clastic sand conglomerate reservoirs are complex and diverse,and the relationship between parent rock compositions and conventional logging curves needs to be analyzed in depth;the area is a set of transitional lithologies with both volcanic rocks and normal clastic sedimentary rocks,as well as mixed volcanic and normal clastic sedimentary rocks.volcanism,fluvial transport and weathering provide sediments for the basin.As the gravels of different grain sizes vary greatly in size during deposition due to the combined effects of fluvial transport and volcanism,the pore structure of different grain size combinations is complex,which further affects the quality of the reservoir.The present study will use micro-resistivity scanning log data to classify and evaluate the sand and gravel particle support type.This paper firstly analyzes the main mineral compositions and parent rock characteristics of the area based on the core thin section analysis data and X diffraction whole rock analysis experimental results.By selecting the conventional logging data,a conglomerate parent rock type identification plate was established,and the conglomerate parent rock compositions were classified into: normal sedimentary rocks,acidic volcanic rocks,medium acidic volcanic rocks,medium basal volcanic rocks and mudstones;then core photos were observed and combined with petrographic experimental data,it was found that the volcanic clastic sand conglomerates in the study area generally have a skeleton containing low resistance conglomerates of acidic volcanic rocks and high resistance conglomerates of normal sedimentation.Based on the micro-resistivity scanning imaging logging data,the Watershed Transform algorithm was used to intelligently quantify the content of high and low resistance gravels by defining the resistivity cutoff threshold in the images,and establishing the lithology identification classification criteria for the study area,classifying the lithology of the study area into conglomerate,sand conglomerate,gravelly sandstone,sandstone and mudstone;by applying the sliding window method to classify the number of gravels in different grain size ranges,generating By applying the sliding window method to classify the number of gravels in different particle sizes,a particle size spectrum was generated to visually characterize the distribution of grains in different particle sizes of sand and gravel.On the basis of the particle size spectrum analysis,the area of irregular gravels is calculated to determine the diameter of gravels,and the number of gravels in a fixed window length is counted,and an innovative particle size area spectrum is constructed to quantitatively evaluate the content of gravels of different grain sizes to obtain the distribution characteristics of gravels of different grain sizes.The oil-bearing characteristics of different grain-size combinations in the cores,combined with the particle size spectra and particle area spectra,form a support type evaluation method based on electro-imaging logging of volcanic clastic sand conglomerates to describe the conglomerate grain support type,and classify the conglomerate grain support type in the study area into same-grade grain support,different-grade grain support and miscellaneous-based grain support.By combining the above research,a "parent rock type+granularity" lithology identification method for volcanic clastic sand conglomerates was innovatively proposed by combining the parent rock composition,lithology classification criteria and conglomerate support type in the study area.To address the problem of low availability of micro-resistivity scanning imaging logs in the study area,this study uses a deep neural network model to model wells in key sections and learn to predict neighboring wells that lack electrical imaging data.Through the above established lithology identification method of volcanic clastic sand conglomerate,the distribution pattern of parent rock types in the study area is summarized based on the test oil production data,and the gravel particle support types of different grainsize combination patterns are combined to jointly evaluate the reservoir quality of volcanic clastic sand conglomerate in the study area and derive the dominant lithology.This method makes full use of the advantages of electrical imaging logging data in conglomerate identification,and provides an important means to finely delineate the lithology identification of volcanic clastic sand conglomerate and the evaluation of grain-size combination patterns.