TITLE: Uncertainty quantification for oil reservoir predictions: stochastic optimisation and machine learning. SPEAKER: Vasily Demyanov (Institute pf Petroleum Engineering, Heriot-Watt University, Edinburgh) ABSTRACT: Prediction of petroleum reservoir performance is a challenging task aimed to support high-value development decisions. The challenge is complicated with immense amount of uncertainty associated with all aspects of reservoir exploration and development. The models used for forecasting how much oil is to be recovered are subject to various types of uncertainty in modelling equations, solution errors, interpretational choices and observation errors. Therefore, robust decisions should be based on an adequate evaluation of uncertainty associated with model prediction, which can be rigorously evaluated in a Bayesian way. Stochastic optimisation is commonly used to calibrate prediction models to available observations (historical oil production data). Inferring from multiple solution of the inverse problem provides a way to quantify uncertainty of the predictions. One of the challenges of the inverse problem is maintaining the model realism though out the calibration process, so the obtained solution still agrees with the basic geological knowledge and considered interpretations. This is achieved by using proper priors derived from a wide domain of analogue geological data with machine learning classifiers. Another challenge is to be able to ingrate a vast domain of related reservoir and geological data and knowledge into the model to ensure the important uncertain aspects of the model are not overlooked. This data are commonly of different scale, type and origin. Therefore, it is vital to make the best possible use of this data to be able to reproduce the essentially multi-scale character of petroleum reservoirs at the right level of complexity. The talk will demonstrate how some of these challenges are addressed in real and synthetic filed example applications of stochastic evolutionary optimisation algorithms and machine learning techniques, such us support vector regression and multiple kernel learning. Vasily Demyanov. Biographical Note: Dr Vasily Demyanov is a lecturer in Geostatistics at Heriot-Watt University (Edinburgh). He lectures geostatistics for MSc students at the Institute of Petroleum Engineering since 2004. Vasily’s research interests are wide spread between spatial statistics, stochastic optimisation, Bayesian statistics and uncertainty quantification. He leads research in machine learning application for petroleum reservoir modelling. He is a co-author of over 50 publications, including books: Geostatistics: Theory and Practice (Nauka, 2010, in Russian), Advanced Mapping of Environmental Data %G–%@ Geostatistics, Machine Learning and Bayesian Maximum Entropy (Wiley, 2008). V. Demyanov is an Associate editor for Computers and Geosciences Elsevier journal. V. Demyanov has obtained his 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 he worked with the University of St. Andrews (2000-2002) and Nuclear Safety Institute, Moscow (1994).