Application Of Principal Component Analysis And Support Vector Machine In Lithology Recognition Of Conglomerate

The lithology recognition of conglomerate reservoir is one of the most basic work in the field of petroleum exploration and development.The conglomerate body is a common lithologic oil and gas reservoir,and the oil-rich conglomerate formation are of great value for exploitation.However,the rock types of conglomerate are various,and they are characterized by wide range of particle size distribution and strong heterogeneity,which cause great difficulties in sedimentary facies division,reservoir unit recognition and reserve evaluation of conglomerate body.Although the traditional drilling coring technologyis the most intuitive,but because of the high cost and long time consuming of drilling coring technology,it is obviously unrealistic to drill coring in many wells,and it cannot provide lithology information for all well segments of each well.It is also well known thatthe conventional logging is the most common data in wells,and the responses of conventional log of different lithology strata are different.Then,how to build a bridge between the core and the conventional logging data,and how to use the logging data to recognize the formation lithology of the whole well section on the basis of the core,which is always an important part of the logging interpretation.Therefore,based on the mathematical methods and well logging data,this paper takes conglomerates of Triassic Baikouquan Formation in Mahu Sag,Junggar Basin as tthe primary study subject,and discusses the method and application of lithology recognition of conglomerate.Based on core observations,combined with existing geological data,the lithology of the conglomerate can be divided into three categories and nine subclasses in this studyarea.The three categories are: sandstone,conglomerate and mudstone.Among them,conglomerate includes four types of lithology: coarse conglomerate,largemedium conglomerate,small-medium conglomerate,fine conglomerate;Sandstone consists of four types of lithology: coarse sandstone,medium sandstone,fine sandstone and siltstone.Firstly,this paper uses five logging curves to identify lithology,which are acoustic transittime(AC),compensated neutron(CNL),density curve(DEN),deep resistivity(RT)and natural gamma ray(GR),respectively.Because the dimensions of each parameter are different in conventional logging data,andit is easy to interfere with lithological recognition results.So the data normalization method is used to preprocess the original logging data in this paper,and in order to make the range of log curve values between 0 and 1.Secondly,there are several logging curves in conventional logging data,and the lithology information reflected has certain correlation between them.They tend to causeinformation redundancy,which increases the complexity of data calculation in the process of lithology recognition.In this paper,the method of Principal Component Analysis is used to extract the characteristic principal component,and principal component with eigenvalue greater than 1 is selected as the input variable of lithology recognition model to reduce the dimension of data.In terms of lithology recognition,Support Vector Machine method has its own advantages,such as its simple structure,fast learning,good adaptability and strong generalization ability,and it can solve the nonlinear problem under the condition of less samples.Therefore,a lithology recognition model is established by using the method of Support Vector Machine,and it can classify the conglomerate lithology one by one.After the classification of large class lithology,the small class lithology is further identified on this basis.Finally,the lithology recognition results of the conglomerate of Principal Component Analysis and Support Vector Machine method are compared with those of discriminant analysis method and BP neural network method.In the process of large class lithology recognition,the lithology recognition accuracy rate of Support Vector Machine method is 88.05%,which is 13.7% and 12.46% higher than that of discriminant analysis method and BP neural network method.In the process of lithology recognition of conglomerate,the lithology recognition accuracy rateof Support Vector Machine method is 79%,which is 22.5% and 20% higher than that of discriminant analysis method and BP neural network method.In the process of sandstone lithology recognition,the lithology recognition accuracy rate of Support Vector Machine method is 68.02%,which is 14.42% and 5.86% higher than that of discriminant analysis method and BP neural network method,respectively.The results show that the lithology recognition results of Principal Component Analysis method and Support Vector Machine method are better than those of discriminant analysis method and BP neural network method.Through comparative analysis with coring observation results,the accuracy rateof Principal Component Analysis and Support Vector Machine lithology recognition method is up to 73% in the processof large group.At the same time,the accuracy rate of this method is up to 61.35% in the process of conglomerate classification,and the accuracy rateof this method is up to 58.61% in the process of sandstone classification.Therefore,the lithology recognition method of Principal Component Analysisand Support Vector Machinecan accurately recognize thelithology of conglomeratein a certain extent.It can also effectively predict the lithology of uncorked well sections,and provide reference for the subtle reservoir exploration,petroleum development and lithology recognition of conglomerate.