Heirarchical geological realism in history matching for reliable reservoir uncertainty predictions
Kuznetsova, Alexandra Anatolievna
MetadataShow full item record
The oil and gas industry has been always associated with huge risks. To minimise these risks, one is looking for the reliable reservoir performance predictions to make better field development decisions. The great challenge associated with reliable predictions is to account for the essential geological uncertainties and propagate them through the engineering model validation process. In this thesis, we propose a new methodology to improve the reliability of reservoir predictions under the Bayesian framework. The first step of the methodology applies the new hierarchical approach to account for essential geological uncertainties from different levels of geological data in facies modelling. As the result of the hierarchical approach, we evaluate the prior range of different geological uncertainties. Facies models greatly affect simulation results but it’s a great challenge to history match them whilst maintaining geological realism. Therefore, next step of the methodology is aiming to improve geological realism during history matching. We propose to combine metric space approach and machine learning classification to evaluate geological relations between multiple geological scenarios and parameters combination and propagate them into history matching. Multidimensional scaling was used to analyse the similarity of the facies models in the metric space. Results of different machine learning classification methods – k-means clustering, Support Vector Machines, Random Forest – were compared to include the ones that performed better into history matching. The reservoir predictions under uncertainty were performed by evaluating the Posterior Probability Distribution under the Bayesian framework and estimating the Credible Intervals (P10, P50, P90). The methodology was applied to a synthetic case study based on a real reservoir of the West Coast of Africa (offshore turbidite reservoir). The main results show that the proposed methodology was able to improve the geological realistic facies model representation during history matching and uncertainty quantification. Some additional controls of facies architecture and facies connectivity modelling could be introduced to improve the quality of the facies realisations.