Machine Learning for Enhanced Insights into Flow Potential in Overburden

Summary Porosity and permeability are important parameters reflecting flow potential in subsurface. In Standardized practices quantify these parameters via core analysis, well logging, and well testing. Nonetheless, these methods are generally costly and mainly applied to reservoir sections. The emphasis on the identification of flow potential in overburden sections is commensurate with concerns about drilling safety and efficient well plugging operations. The study aims to explore an alternative method to quantifying the abovementioned parameters, specifically porosity, in the overburden sections. By leveraging our in-house database and machine learning (ML) techniques, we established a workflow that could yield a machine learning model for estimating the porosity in the overburden sections. We gathered data for 23 wells from the Norwegian Sea with similar geological properties. After preprocessing, we divided the data into training and testing datasets, with the training dataset being input into the SparkBeyond discovery platform for feature engineering. Through several iterations, we refined our ML model for further testing. Our findings exhibit the significant potential of utilizing ML to quantify the parameters and lay the groundwork for future research in this domain.