Machine Learning-Based Prediction of UCS in Carbonate Reservoirs: A Data-Driven Approach to Enhance Geomechanical Modeling in a Field in Middle East

As oil and gas wells reach the end of their productive life, it becomes essential to carry out the well abandonment in a safe and environmentally responsible manner. There are stages to this process beginning with the review of the integrity status of the well and then there are measures stringent isolation passes to completely prevent any hydrocarbons from leaking and secures fluid migration all within regulatory and environmental compliance. This study focuses on Field G, a complex asset comprising both clastic and carbonate reservoirs, and presents a comprehensive geomechanical assessment to evaluate caprock integrity for potential re-use of existing wells. One of the main goals is to see if upgrading or sidetracking existing wells could eliminate the need for new drilling, cutting costs and reducing environmental damage. To feed into reservoir simulation, a comprehensive geomechanical workflow was developed integrating well log interpretation, synthetic log generation and the derivation of important static and dynamic rock properties. Core plugs were recovered from four wells throughout the field, and laboratory testing (uniaxial and triaxial compressive strength experiments) was then conducted. This core-derived UCS measurements provided the basis for model calibration and validation. To improve UCS prediction from uncored intervals, machine learning models (such as Random Forest, Gradient Boosting and Neural Networks) were trained using the log data. This study compared the performance of these models with well-established empirical correlations. The AI-driven models showed considerable improvements in predictive performance with R2 > 0.93 than empirical methods, making way for better characterization of rock strength over the entire reservoir. The parameters derived from the integrated model were further validated based on a maximum chocking pressure envelope with formation integrity and leak-off test (LOT) data to build an average field-wide representative pressure profile. Identification of multiple sealing intervals and validations of the corresponding geomechanical characteristics using the trained UCS models. Some time earlier, these observations played a critical role in assessing the potential validity of henceforth performing a series of partial well abandonments and reservoir re-entries Ultimately, this study not only supports sustainable abandonment strategies and cost-effective field development but also provides a transferable AI-based UCS prediction framework that can benefit other nearby assets with limited core availability.