Bayesian Evidential Learning : Prediction without Model Inversion Stanford University , Geological Sciences Department , Stanford

@inproceedings{Caers2017,
 abstract = { The conventional paradigm in decision making under uncertainty is to build several models that match a variety of data, then use these models to generate posterior uncertainty on key decision variables. In the context of reservoir management. This is generally an expensive and difficult task, and often results in models that do not accurate asses the uncertainty of the forecast. The academic idea of generating 100s models that match all the data has never really worked in a practical setting. We propose an alternative re- formulation of the problem, termed Bayesian Evidential Learning (BEL) in which the role of the model is reconsidered. Instead of using the model to match the historical production, and then forecasting, the model is used in combination with Monte Carlo sampling to establish a statistical relationship between the data and prediction variables. The estimated relationship is then used in conjunction with the actual observed data to produce a statistical prediction on any key decision variable. This allows us to quantify posterior uncertainty on the prediction variable without explicit model inversion. The main rationale behind this is that the subsurface model is highly complex and even so still remains a simplified model of the actual subsurface. As statistical relationships can generally only be constructed in low dimensions, compression and dimension reduction of the models themselves would result in oversimplification. Conversely, data and prediction variables are time series data or maps, which are simpler and much more applicable for dimension reduction techniques. In this presentation, I will provide an overview of the BEL approach, the various strategies for Monte Carlo, dimension reduction, prior model falsification, global sensitivity analysis and non-linear regression methods that can be used for real world applications involving engineering the subsurface. },
 author = { Caers, Jef },
 booktitle = { 2017 Ring Meeting },
 number = { September },
 pages = { 2018 },
 publisher = { ASGA },
 title = { Bayesian Evidential Learning : Prediction without Model Inversion Stanford University , Geological Sciences Department , Stanford },
 year = { 2017 }
}