Estimation of Cementation Exponent m, for Complex Carbonates in Mexico, Using Data Science

Abstract This paper presents a data-driven approach using machine learning to create a model for predicting the behavior of the cementation exponent "m" (it is a function of the shape and distribution of pores) in Mexican Naturally Fractured Reservoirs (NFRs), aiming to reduce uncertainty in reserves calculations and risks on decision-making for wells lacking core data information. The purpose is to predict "m" with limited measurements by machine learning models trained with core data and special logs from key wells. Different authors have addressed the ranges of "m" based on the type of porosity or rock texture; however, most of these studies are not from Mexican fields. Information was compiled from 1,154 data points obtained from cores in different porous systems, including matrix, fractures, unconnected vugs, and connected vugs. Probabilistic and statistical theorems were used to handle complex and nonlinear relationships between porosity and the cementation exponent "m". Data science revealed patterns that were not evident or were influenced by multiple factors. Finally, test examples of partitioned porosity data and several measured cementation exponent core well data have been provided to verify the effectiveness and adaptability for the determination of the cementation exponent "m" through the so-called SPP (Statistical Porosity Partitioning) method. This also provides algorithm support for different porous systems (matrix, fractures and vugs), achieving a MAAPE (Mean Arctangent Absolute Percentage Error) of 0.22 between the originals and the predicted values. The first example applying the SPP method is through the development of an equation to analyze a triple-porosity NFR (intergranular, vugs and fractures), successfully predicting the porosity exponent, with an accuracy of around 70% compared to the core-measured data. The second example demonstrates the success of the equation's performance in a dual-porosity NFR (matrix-fractures). Next, the third example was conducted in a well with connected vugs and matrix, achieving a 100% prediction accuracy of the cementation exponent "m" value compared to the core-measured data. Finally, the fourth case shows an example of an exploratory well without any core analysis, and the results of applying our model achieved a better understanding of the reservoir behavior. The input features for cementation exponent identification are effective porosity and discrete porous system log, with resulting values of 1, 2, 3, and 4, depending on the type of NFR present. The SPP method was then applied to wells lacking specific core information to obtain a cementation exponent "m" exclusively for naturally fractured reservoirs (NFR), thus reducing uncertainty in estimation of hydrocarbon reserves.