Reshaping Reservoir Insights: The Breakthrough Potential of Principal Component-Driven Machine Learning

Abstract The oil and gas industry generates a vast volume of logging data during all phases from exploration to production, resulting in an accumulation of high-dimensional logging data. Concluding insightful outcomes from the gathered information is a daunting task due to the multi-dimensionality constraints in data visualization (three axes of space, x, y and z) and data manipulation. The goal is to use a technique to simplify data visualization from high dimensionality into lower dimensionality while preserving features that are important for big data analysis. One of the most dominant methods to reduce data dimensionality is principal component analysis (PCA), which is used to simplify multivariate datasets acquired by well logging tools (gamma ray, density, neutron porosity, resistivity, etc.) to a smaller number of factors (PCs). The workflow focuses on PCA applications to investigate potentially meaningful petrophysical analysis, reservoir characterization and rock typing, and multi-linear regression analysis for data prediction. Logging data dimensionality was reduced to produce a set of principal components (PC1, PC2…PCx) where x is equal or close to the number of the used variables. The results indicate that the first two components (PC1 and PC2) represent the majority of the data patterns, and the graphical display of these two components (PCA biplot) shows distinctive clusters that can be assigned to separate electro-facies. Results indicate that PC1 behavior correlates well with lithology variations and can be utilized for well-to-well correlation. Linear regression analysis revealed a strong predictive relevance between PCA components and well logging variables, allowing the R-squared regression technique to predict a result curve from PCA input curves. PC1 showed a strong response to substantial lithological variations, whereas higher principal components are more connected to the formation fluid and could be used to complement the standard petrophysical analysis. The paper presents an innovative workflow based on PCA components that utilize a range of dimensional reduction approaches to analyze data in a more descriptive form to identify hidden trends and insights as an approach to machine learning and artificial intelligence.