Leveraging Factor Analysis Machine Learning Workflow Concurrent with Advanced Volumetric Analysis to Improve Porosity-Permeability Transform in Complex Carbonate Reservoir

Abstract

Abstract Permeability determination is critical in understanding the viability of a project as it is often used as an economic indicator in the infill well placement, production strategy and enhanced oil recovery strategies. Often, well tests are planned, and core analysis are performed to evaluate the flow capability of the reservoir, but it may not be sufficient for heterogenous and complex carbonate formation. Hence, to determine the permeability, we often employ correlations such as resistivity-permeability relationship, intrinsic permeability estimation from geochemical data and most common and widely used is the porosity-permeability (poro-perm) relationship. Poro-perm relationship relies on the basis that all pores contribute to fluid flow. However, any heterogeneity, such as presence of isolated pores could cause this poro-perm relationship to fail. Hence, this paper aims to address the challenges associated with the quantification of the isolated pores in the formation. The case study gas well, Well M, is in offshore of Sarawak, Malaysia. The nuclear magnetic resonance (NMR) logs are acquired to quantify porosity and permeability in addition to basic quad-combo and wireline formation tester (WFT) sampling. The direct porosity-permeability transform obtained from NMR Timur-Coates equation shows distinct disagreement by a factor of up to 100 with the mobility obtained from WFT. This discrepancy could be due to the incorrect assumption that all pores are interconnected, but in reality, some of the pores might be isolated porosity. To unravel this complex problem, an advanced analysis incorporating the quad-combo data and NMR data is carried out in the volumetric solver. Since sonic is generally less sensitive to spherical pores, deviation seen between sonic porosity and total porosity is interpreted as the presence of spherical pore. After analyzing the core, it was found that these spherical pores are isolated in nature, hence sonic could be used as a quantification of isolated pores inside the formation. In addition, an unsupervised machine learning algorithm, NMR factor analysis (NMR FA) was performed on the NMR T2 Distribution to fully characterize the formation by analyzing the fluid residing in the pores. This was done via concurrent analysis of the NMR signal modelling. By leveraging machine learning of the NMR data, many of the critical information that would otherwise go undetected were extracted successfully. Lastly, the factor analysis result was blindly compared to advanced volumetric analysis, and both methodologies yield the approximate the same volumes of isolated porosity in the formation of interest (R2 = 0.886). After the quantification of the isolated pores were successfully carried out and confirmed, a reliable poro-perm transform was established. To conclude, poro-perm estimate in this field was enhanced and the permeability uncertainty is greatly reduced. Subsequently, the result from this workflow can be used as a quick preliminary justification on the reservoir flow capability derived from NMR on the new play zone. This will ultimately lead to an earlier input to the production strategy decision and the net present value (NPV) can be maximized accordingly.