Log Evaluation Of Tight Sandstone Gas Reservoirs

Tight sandstone gas reservoirs have strong heterogeneity and complex pore structure,and are difficult to be evaluated with petrophysical logs.Taking the Middle Jurassic Shaximiao Formation tight sandstone gas reservoir in Jinhua-zhongtaishan area of Sichuan Basin as the object,combined with petrophysical experiments and production test data,research is carried out on the determination of macroscopic petrophysical parameters,evaluation of microscopic pore structure,reservoir classification and productivity prediction,from three aspects: empirical model,machine learning and theoretical analysis.Guided by the empirical model,reservoir evaluation methods are studied with conventional logging,electrical imaging logging and acoustic characteristics respectively.Using conventional logs,interpretation models of reservoir parameters suitable for the main sand bodies in the study area are optimized and established,a reservoir classification chart is constructed based on neutron-density envelope difference and compressionalshear velocity ratio,and open flow prediction formulas of horizontal and vertical well sections are fitted respectively based on the constructed apparent productivity index of single layer with neutron-density envelope area and permeability;according to the measurement principle,volume model and parallel conductivity theory,the resistivity calibration and image segmentation of electrical imaging log are carried out,and the resistivity component of sandstone is extracted,which improves the estimation of permeability of porous clastic rock reservoir;based on the variable saturation acoustic experiment,a locally enhanced gas reservoir identification chart is constructed with the compressional-shear velocity ratio and compressional slowness,and then a segmented water saturation model and productivity prediction model are developed.Through verification with core analysis and gas test results,the above empirical models and charts have achieved good application effects.Based on Bayesian regularization neural network(BRNN),machine learning modeling research is carried out in permeability prediction,pore size distribution reconstruction,reservoir classification and productivity prediction.Taking core porosity,conventional logs and some constructed parameters as input items,a unified BRNN model is constructed and optimized according to the data distribution characteristics,petrophysical significance and regional geological laws,and the prediction accuracy and the generalization performance are both satisfactory for the multi-stage tight sandstone reservoir with permeability across 6 orders of magnitude and variable porositypermeability relationship;based on numerical preprocessing such as validity-consistency correction,interpolation and decimation,two BRNN models are established respectively taking the corresponding core nuclear magnetic resonance T2 spectrum and petrophysical logs as the main input items,and the pore size distribution or pseudo capillary pressure curve is accurately constructed through the training and prediction of pore radius and mercury saturation;BRNN classification and regression models are established respectively with selected or constructed characteristic parameters such as neutrondensity envelope difference,cumulative pore thickness and cumulative hydrocarbon thickness,and identification of reservoir types and prediction of gas reservoir productivity are then realized.In the aspect of theoretical analysis,the problem of determining pore fractal dimension based on mercury injection capillary pressure(MICP)is discussed.Based on Yu-Xia model and Washburn equation,an analytical model for calculating pore fractal dimension with MICP is strictly derived.The fractal turning point is determined with the maximum of the moving variance ratio of mercury saturation,and then the global,macropore and micropore fractal dimensions are fitted nonlinearly.It is found that for relatively high permeability reservoirs,a larger global fractal dimension may correspond to a more favorable pore structure and better reservoir quality.The combination of fractal model related parameters,especially the global fractal dimension and entry pressure,can effectively identify reservoir types.Through the comparison of theoretical analysis and calculation results,it is considered that the commonly used He-Hua and Shen-Li models are equivalent to the simplification and approximation of the proposed model.A series of methods and numerical processing procedures developed here provide reliable basis and technical support for the fine description,reserve calculation and plan design in exploration and development of the tight sandstone gas reservoirs in the study area,and can also provide reference for log evaluation and petrophysical analysis of similar reservoirs in other areas and even other types of complex heterogeneous reservoirs.