Unsupervised Machine Learning Methods as a Powerful Tool for Lithological Pattern Recognition Using Wells Logs Data

Abstract The prediction or identification process of lithological facies is one of the pillars of reservoir modeling for any field. However, the lack of representativity of cored data leads to over- or under-prediction of certain types of lithologies, especially because the acquired cored sections hold the best reservoir properties, imprinting bias to the data analysis. In this study, we will present a carbonate platform reservoir model, constructed by following a sequence stratigraphy framework, using 4th order sequence stratigraphy boundaries for reservoir zonation. Hence, each zone contains both high and low reservoir quality, but with the lack of representativity of low-quality zones, it is necessary to identify a different technique that helps to delineate more accurately the extension of each of them (Fig. 1). This strategy is applied to identify groups of wells that may sustain similar sedimentary features and indirectly detect places where certain changes might occur.