A hybrid intelligent model for reservoir production and associated dynamic risks

Abstract

This research presents a hybrid model to predict oil production and to provide a dynamic risk profile of the production system. The introduced predictive approach combines a multilayer perceptron (MLP)-artificial neural network (ANN) model with a hybrid connectionist strategy (BN-DBN), which comprises a Bayesian network (BN) model and a dynamic Bayesian network (DBN) model. The proposed hybrid methodology (MLP-BN-DBN) is designed to find the correlations between the input and output data to forecast the desired oil production rate. The MLP model captures the variabilities in the fluid and rock properties, model's uncertainties, and the effects of pressure maintenance on the production process. The BN model uses the 3σ mathematical rule to promptly signal the arrival of any production rate change and captures the pressure maintenance impact using the early warning source indexes. The DBN model provides a dynamic risk profile of the production system using the observed evidence and reservoir production hyperbolic decline concept.The proposed methodology offers the field operators better opportunity to obtain real time estimate of the likelihood of any impending production loss at any time during production operations. The model exhibits a high capability of oil production prediction with the minimum, average, and maximum percentage errors of 0.01%, 6.57%, and 15.28%, respectively. The developed hybrid model serves as a risk monitoring system. The model is cost-effective and eases the computational burden of history matching processes and bridges the gaps in the existing systems for oilfield development dynamic risk forecast and production predictions. Hence, the proposed methodology offers a multipurpose tool for dynamic risk assessment and for proper reservoir production optimization.