Challenges and Solutions in Stochastic Reservoir Modelling - Geostatistics, Machine Learning, Uncertainty Prediction
Reservoir prediction modelling is subject to many uncertainties associated with the knowledge about the reservoir and the way they are incorporated into the model. Modern reservoir modelling workflows, which are commonly based on geostatistical algorithms, aim to support development decisions by providing adequate reservoir description and predict its performance. Uncertainty about reservoir description needs to be accounted for in modelling workflows to quantify the spread of reservoir predictions and its impact development decisions.
The course aims to build awareness of the impact the modelling choices on the reservoir predictions and their relation to the way uncertainty is incorporated into reservoir modelling workflows. The course addresses the problem of tying the workflow with the expected geological vision of a reservoir subject to uncertainty. This is associated with one of the common issues, when standard assumptions of a workflow are not consistent with the model geology or do not reflect possible variations due to existing uncertainty.
The course demonstrates the implementation of geostatistical concepts and algorithms in geomodelling workflows and the ways uncertainty is accounted for in reservoir description and predictions. The course includes an overview of the state-of-the art conventional techniques and some more novel approaches, in particular machine learning for reservoir description.
Machine learning provides new opportunities in data integration and the model control to tackle the modelling challenges related to non-stationary multi-scale correlation structure and complex connectivity patterns in reservoirs. Novel machine learning techniques are good at capturing dependencies from data, when their parametric description is difficult; and controlling the impact of noisy and ad-hoc data.
The objectives of the course are:
- Provide a practical overview of geostatistical concepts used in modelling workflows and their impact on reservoir models.
- Explain how uncertainty is accounted for and propagated through geomodelling workflows.
- Introduce some machine learning approaches in geomodelling through a series of case studies.
- Demonstrate integration of geomodelling techniques for uncertainty quantification of reservoir predictions.
The outcome of the course is the practical understanding the key geostatistical concepts, their implementation in reservoir modelling, principles of machine learning and its application in reservoir modelling. The course participants will get familiar with a variety of conventional and advanced stochastic modelling algorithms and become aware of the impact from different modelling assumptions.
- Concepts and assumptions of geostatistics
- Stochastic simulation vs interpolation
- Uncertainty quantification workflow for reservoir predication
- Machine learning techniques for spatial reservoir modelling
- Classification - lithofacies
- Regression – ML geomodelling
- Dynamic model update through learning from data
- Data driven integration of relevant data and knowledge
The course is designed for a wide audience of reservoir modellers, geologists and engineers with range of experience from novices to experienced practitioners.
Participants should have a basic knowledge of reservoir modelling and numerical analysis.
About the Instructor
Prof. Vasily Demyanov of Institute of Petroleum Engineering, Heriot- Watt University (Edinburgh), lectures geostatistics and leads industry and government funded research in Geo Data Science and uncertainty quantification for reservoir prediction modelling. He has over 20 years of experience in geostatistics and has published over 100 publications. Vasily has co-authored a number books: Challenges and Solutions in Stochastic Reservoir Modelling – Geostatistics, Machine Learning, and Uncertainty Predictions with EAGE (2018); Geostatistics: Theory and Practice (Nauka, 2010, in Russian). Vasily Demyanov is an Associate editor for Computers and Geosciences Elsevier journal and a guest editor for Mathematical Geosciences Springer Journal special issue on Data Science in Geoscience due in 2019. Vasily’s research interests lie broadly across spatial statistics, machine learning and uncertainty. In particular his research is focused on uncertainty quantification in prediction modelling, inverse modelling for history matching, stochastic optimisation, Bayesian inference, and the problem of integration of reservoir knowledge and relevant data into statistical modelling workflows with machine learning and data analytics approaches.
Vasily Demyanov obtained the first degree in physics from Moscow State University (1994) and a PhD in physics and mathematics from Russian Academy of Sciences (1998) with a thesis on radioactive pollution modelling with geostatistics and artificial neural networks. Prior to joining Heriot-Watt in 2003 he worked with the University of St. Andrews (2000-2002) and Nuclear Safety Institute, Moscow (1994-2000).