Pattern Recognition for Wettability Alteration with Surfactants in Carbonate Reservoirs by Using Machine Learning

Abstract

Abstract Surfactants are widely applied agents to interact with the adsorbates on carbonate rock surface, which could alter wettability from oil-wetness to water-wetness and thus enhance oil recovery. Surfactant huff-puff is often conducted to achieve wettability alteration and has been applied under various conditions. Currently, many investigations have been reported to apply conventional data analysis methods to analyze different sets of surfactant huff-puff projects. Yet the application of machine learning algorithms to reveal the inherent patterns is rarely reported. In this study, we integrate principal component analysis (PCA) with hierarchical clustering algorithm (HCA) to uncover the hidden patterns embedded in global surfactant huff-puff dataset. PCA is effective to transform the original data space to principal components space, where five principal components could represent the information of original nine reservoir and fluid parameters and maintain around 90% of total variance. Based on the transformed data space, HCA is implemented with optimized structure, including Euclidean distance measure, Ward’s linkage method and three clusters. It shows that HCA obtains a normalized mutual information score, V-measure score, and Fowlkes-Mallows index of 0.75, 0.75 and 0.83, which suggests the effectiveness and reliability of the newly proposed PCA/HCA procedure. The PCA/HCA is able to group similar surfactant huff-puff treatments into the same cluster. Three distinct cluster patterns are found in the surfactant huff-puff dataset. The second cluster includes most of the high temperature and high salinity carbonate rocks, while the third cluster includes chalk rocks with fairly low permeability and high porosity. The first cluster contains surfactant treatments conducted with low temperature and low divalent concentrations. Besides, the analog reasoning results show that the PCA/HCA is able to provide valuable experiences for designing surfactant treatments and predicting future performance for new candidate surfactant treatment projects. This procedure could greatly save time and cost.