Revolutionizing Mud Gas Application: Advances in Fluid Properties Prediction and Developing a Fit for Basin Model

Summary In this work, we present a methodology to accurately predict reservoir fluid properties by leveraging advanced mud gas (AMG) data in combination with machine learning techniques. The approach begins with a classification step using a random forest model to identify the fluid type, achieving an accuracy of around 92%. This classification is critical to guide the subsequent prediction steps. Once the fluid type is identified, we use a Gaussian Process Regression model to estimate the C6+ fraction directly from the AMG measurements. Given the impact of biogenic and biodegraded fluids on the predictive performance, we developed separate models to address these specific fluid types. After obtaining reliable C6+ estimates, we compute the gas-oil ratio (GOR), yielding a robust and consistent solution across different scenarios. A key aspect of our approach is the emphasis on retraining the models for each basin, allowing the methodology to adapt to local geological and fluid variations. This basin-specific retraining improves both the accuracy and the practical applicability of the solution. Overall, the iterative structure of the workflow, combined with the use of tailored models, results in a precise and flexible tool for real-time fluid property prediction during drilling operations.