Tight sandstone reservoir characterization and validity evaluation faced great challenge due to the complicated pore structure and strong heterogeneity. Generally, the nuclear magnetic resonance (NMR) logging was used to characterize pore structure to improve tight sandstone reservoirs evaluation. The limitation of quantity of NMR data and effect of saturated hydrocarbon to NMR T2spectra made it impossible to be widely used, especially in development wells. In this study, to quantitatively characterize pore structure in development wells to improve validity evaluation in Triassic Chang 8 Formation, in Ansai Region, eastern Ordos Basin, a method of constructing pseudo capillary pressure from conventional well logging data based on machine learning method was proposed. Based on the analysis of the morphological feature of mercury injection capillary pressure (MICP) from core samples, we concluded that the used mercury injection pressures were uniform for all core samples, the morphological difference can be determined by the mercury injection saturation under every mercury inject pressure increment. The relationships among mercury injection saturations with conventional well logging data were extracted by using the data mining and decision tree techniques. Finally, the porosity, density, shaly content, and the ratio of deep and shallow resistivity were found to be sensitive to mercury injection saturations at every mercury injection pressure, and they were chosen as the input parameters to establish mercury injection saturation prediction model. 86 clusters of core analysis data (accounting for 78.6% of the total)were used as the training samples, and the rest was retained as samples for verification. The random forest algorithm was used to establish the models of predicting mercury injection saturation from conventional well logging data. Combining the predicted mercury injection saturations with fixed mercury injection pressures, pseudo capillary injection curves were well constructed, and the pore structure evaluation parameters, e.g., the average capillary pressure, the threshold pressure, and the median pore throat radius, were calculated. Our target tight sandstone reservoirs were classified into three clusters based on the constructed capillary pressure curves, the corresponding pore structure parameters and drill stem test (DST) data. The first and second types of formations were effective, and they can be exploited after some development program was used. The third type of formation was deficient. Our research method and technique were well used to improve tight sandstone reservoirs characterization and evaluation, and it would also be valuable in indicating the distribution of effective tight sandstones.
Tight Sandstone Reservoir Pore Structure Characterization from Conventional Well Logging Data Based on Machine Learning Method
Li, Fei, Zhang, Wenjing, Li, Weibing, Chen, Zhen, Sun, Bowen, Chi, Ruiqiang, Li, Gaoren, and Liang Xiao. "Tight Sandstone Reservoir Pore Structure Characterization from Conventional Well Logging Data Based on Machine Learning Method." Paper presented at the SPWLA 27th Formation Evaluation Symposium of Japan, Virtual, September 2022.
Download citation file: