| Tight sandstone reservoir has become a hot field of oil and gas exploration and development in China.It has the characteristics of complex pore structure and strong reservoir heterogeneity,which leads to insufficient calculation accuracy of reservoir physical parameters and difficult identification of fluid properties.Based on the data of drilling coring,analysis and testing,logging,oil testing and production performance,and guided by the theories of machine learning,geophysical logging and reservoir geology,this paper systematically studies the intelligent logging reservoir evaluation methods,improves the accuracy and efficiency of logging reservoir evaluation,and provides theoretical guidance for the efficient development of tight sandstone reservoirs.Based on drilling coring and laboratory analysis data,the basic characteristics and four-property relationship of tight sandstone reservoirs in Chang 8 member of Chuankou target area in Zhenjing oilfield are clarified.The rock types of Chang 8 member are diverse,mainly lithic feldspar sandstone,feldspar lithic sandstone and feldspar lithic quartz sandstone.The reservoir has poor physical properties and belongs to ultra-low porosity and ultra-low permeability tight sandstone reservoir.The oil-bearing property of the reservoir is relatively good,mainly oil spots and oil immersion.The difference of physical properties of different lithologies in the reservoir is small,the oil-bearing property is positively correlated with the physical properties,and the discrimination of logging response characteristics of different fluids is small.The geological prior knowledge is integrated into the intelligent algorithm,and the reservoir physical parameters and fluid intelligent prediction model are established.The porosity prediction model based on gated recurrent unit neural network(GRU)was constructed with the porosity label of NMR logging interpretation.Taking the core physical property analysis data as the label,the genetic algorithm is used to optimize the parameters of the extreme gradient lifting algorithm,and the GA-XGBoost permeability intelligent prediction model is constructed.A method for calculating water saturation of tight sandstone reservoirs with variable rock electrical parameters is proposed,and a fluid discrimination model based on convolutional long short-term memory network(Conv LSTM)is established.The experimental analysis shows that the accuracy of GRU porosity prediction model is improved by 12.2 % compared with the multiple regression model.The GA algorithm can effectively optimize the parameters of the XGBoost algorithm.Compared with the single XGBoost model,the accuracy of the GA-XGBoost permeability prediction model is improved by 14.1 %.The accuracy of Conv LSTM fluid discrimination model reaches 82.3 %,and the prediction conclusion is in line with geological understanding. |
