Multi-objective methods for history matching, uncertainty prediction and optimisation in reservoir modelling

Robust decision-making and reliable forecasting uncertainty are the two key factors to the success of modern reservoir development and management. The reason is straightforward: signi1cant capital investments are involved (i.e. hundreds of millions of dollars or more) by an incomplete understanding of the oil and gas reservoirs developed. Hence, wellinformed decision-making with a good knowledge of the reservoir has always been a critical component in the risk-based oil and gas industry. The research in this thesis focuses on developing solutions for robust decision-making and reliable forecasting using multi-objective methods to history matching and reservoir development optimisation within a Bayesian approach for uncertainty quanti1cation. The multi-objective approach on history matching can 1nd an ensemble of diverse set and good matched models. This diverse set of good matched models is essential for reliable and yet realistic uncertainty prediction of future 1eld behaviours. Additionally, the models from multi-objective history matching also can be used in the reservoir development optimisation to obtain robust decision under uncertainty. Several challenges in the framework of multi-objective history matching, uncertainty quanti1cation and optimisation have been identi1ed and investigated in this thesis. These challenges include: (1) impact of the uncertainty in the model parameterisation on the forecast reliability; (2) history matching e6ciency in case of many matched-criteria and the way they can be grouped into multiple objectives; (3) the problem with a high number of objectives; and (4) reservoir development optimisation under uncertainty. The thesis proposes solutions for each of the challenges mentioned above through extensive studies on both synthetic and real 1eld cases supported by rigour statistical evaluations. The opportunity o:ered by multi-objective to generate an ensemble of a diverse set of good matched models has been explored to handle the 1rst challenge. A novel technique on how to group and select the objective grouping properly for multi-objective history matching has been proposed to address the second challenge. A recently proposed manyobjective optimisation algorithm has been applied to cope with the third challenge. Finally, a new workflow for reservoir development optimisation under uncertainty to obtain robust and reliable uncertainty estimation in the optimisation forecast is proposed.