A Novel Fuzzy Logic Framework for Model Reliability Evaluation in Permeability Prediction Using GPR

Permeability is a critical parameter in reservoir engineering and hydrocarbon extraction, yet its prediction remains challenging due to inherent uncertainties in subsurface data. While Gaussian Process Regression (GPR) has proven effective in predicting permeability with associated uncertainties, it generates multiple metrics that are difficult to interpret, particularly in high-stakes environments. This study proposes a novel approach using fuzzy logic to compute a single, comprehensive metric that accounts for model reliability. Our method incorporates human input and reasoning into the modelling process, enhancing the model's interpretability and its ability to handle uncertainty. Additionally, we introduce a new visualization technique to simplify the understanding of fuzzy logic outputs for non-technical stakeholders. The proposed methodology demonstrates that GPR achieves a higher reliability level (0.89) compared to traditional machine learning counterparts, which are typically neutral to uncertainties. By providing a comprehensive, transparent, and easily interpretable measure of model reliability, this approach significantly aids in making more informed and responsible decisions in reservoir management. Our framework represents a crucial step towards improving the practical application of advanced machine learning techniques in the oil and gas industry, potentially extending to other fields where uncertainty quantification is vital.