Digital Framework for Water Conformance Diagnosis and Optimization Using Machine Learning and Generative AI

This paper presents an innovative autonomous framework for addressing water conformance challenges and recommending optimum water shutoff strategies in oil and gas wells. The introduced approach integrates data management, analytics, machine learning, physics-driven well modelling, and generative AI to establish an end-to-end workflow, which (1) transforms unstructured data from technical reports into structured databases, (2) automates well monitoring and diagnostics, and (3) scales efficiently for deployment across large digital fields.

The presented framework consists of eight modules, which (1) employs RAG to transform raw information into structured databases, (2) applies ensemble unsupervised learning to analyze production anomalies, (3) uses computer vision to diagnose water production patterns, (4) evaluates zonal contribution and well integrity using PLT and CBL/VDL data, (5) integrates physics-driven models for permeability matching, sensitivity analysis, and gain estimation, (6) utilizes PWLA to segment DCA and forecast production trends, and (7) leverages agentic AI to recommend the optimum chemical or mechanical solutions for water conformance challenges.

The proposed framework was applied across multiple wells across the MENA region, demonstrating substantial improvements in efficiency, accuracy, and time savings. Reliable results were achieved where structured data was extracted with accuracy exceeding 0.90. Production anomalies were detected with an F1-score of 0.76. Water production patterns were identified with a mean average precision of 0.84, while DCA forecasting was fitted with an R2 score of 0.86. Moreover, the physics-driven model enabled rapid and accurate matching of permeability and skin for over 5,000 test points within minutes. Furthermore, it permitted nodal analysis for 50+ wells to be completed in mere seconds, significantly improving efficiency of well evaluation. On the other hand, context engineering significantly improved the precision and adaptability of agentic AI and RAG models, while local internal generative models upheld data governance standards. Overall, this comprehensive framework enabled in-depth systematic analysis of well performance, contributing to more robust and informed reservoir characterization and enhancing data-driven decision making while avoiding tedious work.

The introduced methodology represents a state-of-the-art solution that extends beyond traditional analytics and ML classical approaches, which often rely on the availability of structured data. This framework was designed to effectively operate with messy unstructured real field data making it practical for oilfields. In addition, it incorporates agentic AI to recommend WSO products, providing intelligent ways to diagnose and resolve water conformance issues, which significantly impact well performance and operational costs.