Risk Analysis With History Matching Using Experimental Design or Artificial Neural Networks

Abstract

AbstractOil companies have increasingly adopted Risk Analysis in decision-making processes. Most risks are related to reservoir uncertainties. Risk Analysis, even combining possible values of the input parameters (uncertainty variables) in the flow simulations, can create unrealistic models and compute wrong results (decision variable). Some reservoirs have dynamic data that can reasonably constrain the model increasing the Risk Analysis confidence.Three methodologies based on Experimental Design and Response Surface Modeling (RSM) to model flow simulations are proposed. In the first method, one RSM is built to model the decision variable and another RSM is built to represent an Objective Function that takes into account dynamic data. Then, a relationship among the uncertainty variables obtained from the RSM of the Objective Function is applied to the RSM of the decision variable to constrain the model, enabling the Risk Analysis with History Match. In the second methodology, a maximum tolerance of the Objective Function is used as a filter to select the models taken on Risk Analysis. The third method is similar to the second one, but Artificial Neural Network replaces RSM to model the flow simulation. Even though requiring more simulation runs, such approach seems to be skillful to deal with non-linearities.The methodologies were applied to two different reservoirs in Campos Basin, Brazil: a semi-synthetic case where dynamic data were simulated to represent the behavior of an appraisal phase field and a real case where the reservoir produces for more than five years. Significant reduction on the decision variable uncertainty is observed showing the importance of constraining the models when dynamic data are available.The methods show that it is possible to improve the quality of the Risk Analysis, constraining the possible range of the uncertainty variables by taking into account dynamic data.