Pore structure described the macroscopic pore size and microscopic pore connectivity. It heavily determined formation quality and seepage capacity, and thus associated with permeability. Generally, ultra-low permeability to tight sandstone reservoirs were always affected by complicated pore structure and strong heterogeneity. Characterizing pore structure was of great importance in improving tight sandstone reservoir evaluation and validity prediction. Nuclear magnetic resonance (NMR) logging was considered to be valuable in pore structure prediction only in exploration wells because plenty of NMR logging data was acquired in key wells. However, methods that established in exploration wells cannot be directly extended into development wells due to the limitation of quantity of NMR data. In addition, NMR logging was only usable in pore structure characterization in water saturated layers, it cannot be directly used in hydrocarbon-bearing reservoirs. In this study, to establish a widely applicable pore structure characterization method that can be used not only in exploration wells, but also available in development wells to improve formation validity evaluation and high-quality formation identification in Triassic Chang 8 Formation of Shunning Region, Eastern Ordos Basin, we established a technique to synthetize pseudo-Pc curve from geophysical logging data by using deep learning method. This technique was raised based on the morphological feature analysis of mercury injection capillary pressure curves. We found that the applied mercury injection pressures were the same for all core samples during mercury injection experiments, the pore structure difference for all core samples was determined by injected mercury content (SHg) under the same Pc. Hence, once we predicted mercury content under every Pc, pseudo-Pc curve can be synthetized by combining predicted mercury content and known Pc. Constructing pseudo-Pc curve was translated as predicting mercury content. To establish a reasonable model that can be used in development wells, where only conventional logging data was available, we analyzed relationships among mercury contents under every mercury injection pressure and geophysical logging data. This analysis was raised based on heat map of decision tree technique, and the experimental data of 115 core samples that drilled from Triassic Chang 8 Formation in Shunning Region was used. Finally, we found that SHg under 15th capillary pressure was heavily related to porosity and deep and shallow resistivity. Based on this perfect relationship, we established a model to predict 15thSHg from porosity and deep and shallow resistivity by using deep learning method of XGBoost. In this deep learning method, 92 clusters of core analysis data (accounting for 80.0% of the total), were used as training samples, and the rest 20.0% was retained as samples for verification. Meanwhile, relationship between SHgs under two adjacent mercury injection pressures was also closely related. Hence, after SHg under 15th Pc was predicted from conventional logging data, the other SHgs can be calculated by using step iterative method. In addition, considering the used input porosity in XGBoost was also difficult to be estimated based on statistical method, neutron, density, interval transit time (Δt) and delta natural gamma ray (ΔGR) were chosen as input parameters, and XGBoost was used to predict porosity from well logging data. Based on predicted porosity and deep and shallow resistivity, pseudo-Pc curves were consecutively synthetized to characterize pore structure of tight Chang 8 sandstone reservoirs. Meanwhile, pore throat radius distribution, and pore structure evaluation parameters were also calculated, comparison of predicted pore structure evaluation parameters and core derived results illustrated that calculation accuracy reached to 86.4%.

In addition, we determined two pore throat radius cutoffs to classify pore throat radius into three parts, which represented small, intermediate and large pore throat sizes, separately. The relative contents of each type of pore throat sizes were calculated, separately. A parameter of formation validity indication was raised to evaluate formation pore structure. Relationship between formation validity indication and daily liquid production per meter was established, and formations were classified into three types. The first and second types of formations were effective formations that contained substantial hydrocarbon production capacity, and the third type of formation was dry. Our raised method and technique were well used to improve tight reservoirs characterization and evaluation in Chang 8 Formation of Shunning Region, and it would also be valuable in indicating the distribution of effective tight sandstones for formations with similar properties.

You can access this article if you purchase or spend a download.