Automatic Identification Of Deposition Architecture Units From Logging Curves By Artificial Intelligence

Architecture analysis is the key to describe the inhomogeneity inside the reservoir and improve the recovery of oil and gas.The architecture units in the reservoir sense are mainly 3-5 levels.At present,most of the domestic research on automatic identification of architecture units stays at the architecture element limited by the 4-level interface,i.e.,the single deposited microfacies level.Less research has been done on the automatic identification of architecture elements limited by a 3-level interface.Meanwhile,most of the existing machine learning automatic identification methods treat the logging data as discrete data points,without considering the characteristics such as vertical continuity and geometry of the logging curve of the geological body unit.In response to the above problems,this paper takes the 8 member of Xiashihezi Formation in Ct3 area,Ordos Basin is taken as an example and conducts an in-depth study of the artificial intelligence-based method of identifying sedimentary architecture units of logging curves,and carries out the following research work:First,we comprehensively investigate the status of domestic and international research on the identification of sedimentary configuration units of logging curves,analyze the achievements and shortcomings of the current existing artificial intelligence methods,and provide thinking directions and scientific theoretical support for the research of this paper.Secondly,the lithological characteristics and sedimentary microphase spreading characteristics of the target formation in the study area are analyzed.A pre-processing plan is proposed for the characteristics and problems of the logging curves in the study area,including depth alignment,smoothing filtering and standardization of the logging curves.After the curve pre-processing,the rhythmical section of the curve is automatically stratified using the activity function method.Combined with the actual situation in the study area,the parameters of curve morphology features are selected and the automatic extraction of logging curve features in the study area is realized.Then,multiple methods in artificial intelligence are analyzed in depth for automatic architectureal unit identification efficacy.Combined with the configuration unit labels classified by geologists,a learning sample set of sedimentary configuration units in this study area is established,and three kinds of logging curves,namely natural gamma(GR),natural potential(SP)and deep lateral logging(LLD),are selected as experimental data samples,and a reasonable AI model is established for learning by combining the morphological characteristics parameters of logging curves.After the comparison of experimental results,the algorithm with the highest accuracy of sedimentary configuration unit recognition among different AI methods is analyzed to complete the study of automatic recognition of sedimentary configuration unit based on AI in the study area.Finally,an automatic identification method of architecture units based on the "random forest-GR half-amplitude polar point method" is proposed.Based on the morphological parameters of the logging curve,wavelet coefficient polar slope is added to indicate the curve rhythm and characterize the formation rotation information to compensate for the deficiency of the machine learning algorithm in geological understanding.On the premise of identifying the 4-level architecture unit by random forest,the 3-level architecture unit-interlayer inside the braided bar is automatically identified by identifying the half-amplitude extreme value point of the GR curve to realize the refinement of sedimentary architecture unit identification.In conclusion,by incorporating the rhythmic and morphological features of logging curves into pure logging data and using a two-step identification method,the identification accuracy of architecture units can be significantly improved.