Comprehensive Logs Evaluation Method Of Tight Gas Reservoir

Tight gas accounts for about 30% of China's natural gas production and plays an important role in China's energy structure.Compared with conventional natural gas reservoirs,tight gas reservoirs usually have low porosity,low permeability,large burial depth,high diagenesis intensity,compact lithology,complex pore structure and strong heterogeneity,which bring comprehensive evaluation of reservoirs a huge challenge.Using relevant logging data,combined with intelligent algorithms such as machine learning,to establish a fine logs interpretation model,carry out comprehensive logs evaluation technology research has both theoretical research significance and urgent practical value.In this thesis,the study of the comprehensive evaluation method of tight sandstone logs is carried out with the study of the Xiashihezi Formation,Shanxi Formation and Taiyuan Formation in the Daniudi gas field of Ordos.Firstly,based on the petroleum geology,petrophysical experiments and conventional logs data in the study area,the study of reservoir “four properties” relationship and logging response characteristics are carried out,prepare for finding fluid identification methods and logs interpretation medels for tight gas reservoirs.Secondly,combined with the test results of the target layers,the fluid identification method is studied based on the logging response mechanism,the curve overlap method,calculation of acoustic logs curve method,Piccket plate method,normal distribution method,porosity difference ratio method,resistivity difference ratio method were established;After the fine core is homing,the core data is equally weighted by the appropriate resolution window length.the empirical relationship between core measurement porosity and acoustic wave and density logs values is established,and the parameter processing module is formed,which is connected to the Forward.NET platform and meets the needs of large-scale processing.Finally,based on BPNN,SVM,KNN single learner and ensemble learner,the reservoir fluid identification is realized,and the performance difference between different classification learners is compared.The machine learning method is used to effectively solve the tight gas reservoir fluid identification.Based on BPNN,SVM,GPR single learner and ensemble learner,we have realized the quantitative prediction of reservoir parameters,compared the performance difference between different regression learners,and explored a new method for predicting the parameters of tight gas reservoirs.In this thesis,a suitable fluid identification method and logs interpretation model are established.The gas layer recognition accuracy and the calculation accuracy of the reservoir parameters are improved,and the comprehensive evaluation of the tight gas reservoir is realized.The research results provide a new perspective for comprehensive evaluation of tight gas reservoir logs and provide technical support for further expansion of exploration results.