Predicting Rock Properties from Formation Fluid Measurements: Examples, Challenges, and Future Possibilities

Abstract

Digital transformation is unleashing unprecedented potentials to maximize subsurface data utility and explore various possibilities. The petroleum industry has witnessed for a long time the prediction of reservoir rock properties from wireline data. Recent efforts have explored the same with drilling surface parameters. While basic and advanced mud gas data have been used extensively for reservoir fluid characterization, there is the question of whether their utility could be extended to reservoir rock characterization. There is currently no empirical correlation or established relationship between mud gas measurements and rock properties. This gave us the motivation to embark on this research effort. In this paper, we will review the physics behind the relationship between mud gas measurements and reservoir properties especially porosity, and how the measurements are carried out. We will also discuss a number of research questions on the need to extend the utility of mud gas data beyond the traditional reservoir fluid typing. A physics-based workflow that follows the machine learning modeling approach will be presented and discussed. Example results of two successful applications of predicting porosity and hydrocarbon volume fraction from only mud gas data and a combination of drilling surface parameters and mud gas data respectively will be discussed. The applications are based on state-of-the-art machine learning methods including random forest and support vector machines. We used correlation coefficient, mean squared error, and mean percentage error as the models’ performance evaluation metrics. All the metrics indicate that the models gave acceptable results, proved the feasibility of this proposal, and gave impetus for exploring more possibilities. The challenges faced while implementing the proposed methodology and answering the research questions will be shared. Moving forward, other possibilities of predicting various rock properties and their respective implications will be discussed. There is immense value in utilizing only mud gas data or using it to complement drilling surface parameters to predict reservoir rock properties, especially in real time. With dedicated research, incorporating the physics of the problem, and leveraging the power of machine learning, the industry could be witnessing the emergence of a new era of "real-time reservoir characterization".