AVO Analysis And Fluid Identification Based On Proximal Support Vector Machines

AVO(Amplitude Versus Offset)inversion can obtain a variety of pre-stack attributes,but the identification ability of a single attribute(or factor)on the fluid is weaker than the combination factor,so using multiple parameters or combination factor for fluid identification is one of the development directions of reservoir description at present.Based on the pre-stack seismic data,this paper extracts a variety of AVO attributes and combination factors,and uses approximate support vector machine for machine learning to improve the accuracy of fluid identification.The research includes the following aspects:(1)The principles includes: advantages and disadvantages of support vector machine(SVM)and approximate support vector machine(PSVM)are compared.(2)The fluid identification ability of single AVO attribute and the fluid identification ability of combination factor was analyzed.And the input parameters for PSVM were optimized.(3)Built the flow framework of PSVM multi-class classification algorithm for fluid identification in the actual work area.First,theoretical data was used to verify a series of major functions of PSVM;Secondly,practical well bypass data was used for learning;Finally,the entire section was classified by PSVM.Through the above research,the following conclusion can be reached:(1)The PSVM method based on the traditional SVM significantly improves the computing efficiency and has more advantages in the processing of large sample data.A large number of case studies have proved that both theoretical data and practical well bypass data can be classified by PSVM with its corresponding classification function,and its classification effect is good,which also verifies the significant gas-water identification ability of combined fluid identification factor from the side.(2)The extended fluid identification factor constructed by using AVO angular gathers difference has high practicability and sensitivity.(3)The P-G profile and other parameter profiles calculated by the P-G profile can be obtained by AVO seismic inversion,and the AVO anomalies can be easily separated by using these attribute parameter profiles,which have important research significance to test the consistency and correctness of AVO seismic inversion processing.(4)The results obtained by using PSVM multi-class classification method for multi-attribute fluid identification are more accurate than those obtained by single factor fluid identification,and its division effect is more obvious,which can accurately identify the abnormal response of gas-bearing reservoir.