Fuzzy machine learning contribution in reservoir characterization from well-logging data

Abstract

This thesis presents a comprehensive exploration of the integration of Neuro-Fuzzy Systems (NFS) within the domain of reservoir characterisation, with a specific focus on the analysis of petrophysical data in both conventional and unconventional reservoirs, notably within the Algerian Sahara region. Leveraging recent advancements in machine learning, neural networks, and fuzzy logic, this research elucidates the pivotal role of NFS as hybrid machine learning systems in augmenting reservoir characterisation methodologies. Drawing upon two peer-reviewed publications, this thesis embarks on an elaborate work to contextualize the latest developments in NFS within the broader domain of machine learning applications in reservoir characterisation. In the first foundational chapter, we delineate the fundamental principles underpinning machine learning, fuzzy logic, and the amalgamation thereof in the form of Neuro-Fuzzy Systems. Through a rigorous exposition, the theoretical underpinnings and operational mechanisms of these paradigms are elucidated, laying the groundwork for subsequent chapters. A meticulous examination of contemporary machine learning applications in reservoir characterisation forms the essence of chapter two. By synthesising existing literature, we distinguish prevalent methodologies, challenges, and advancements in employing machine learning techniques for reservoir characterisation tasks, thereby providing a comprehensive overview of the current status. Building upon the theoretical framework established in preceding chapters, Chapter 3 explores the application of unsupervised fuzzy logic methods for lithology classification. Through empirical investigations, the efficacy of fuzzy logic algorithms in delineating lithological boundaries is assessed, contributing to enhanced reservoir characterisation workflows. Chapter four undertakes the task of predicting porosity and permeability in a conventional reservoir situated within the Algerian Sahara region. Leveraging machine learning techniques, predictive models are developed to accurately estimate these critical reservoir properties, thereby facilitating informed decision-making in petroleum exploration and production endeavours. In the concluding chapter, the research findings are synthesized, and key insights gleaned from the empirical investigations are elucidated. Moreover, recommendations for future research endeavours aimed at further enhancing the efficacy and applicability of automated methods in predicting hydrocarbon reservoir properties are delineated, underscoring the imperative for continued interdisciplinary collaboration and innovation in the field of reservoir characterisation