Non-linear Heterogeneous Ensemble Model for Permeability Prediction of Oil Reservoirs

Abstract

The ensemble computational model, which uses machine learning techniques to learn partial solutions of a given problem and combines the solutions to obtain a complete solution to the original problem, is increasingly used in both classification and regression problems. Typically the ensemble model performs better than the individual models and ensures better reliability of the model. The investigations of literature reveal that support vector machine (SVM), artificial neural network (ANN) and adaptive neuro-fuzzy inference system (ANFIS) are adopted individually to characterize oil reservoirs. In this paper, an ensemble of SVM, ANN and ANFIS is proposed to predict the permeability of oil reservoirs by using real-life well logs. An ANN model is used to implement a non-linear ensemble strategy. The individual models in the ensemble approach are trained with the datasets obtained by using bagging sampling principle. In this sampling approach, partial datasets for training each model are selected randomly. The simulation results reveal that the proposed ensemble of heterogeneous models outperforms the individual models with respect to correlation coefficient, root mean square error, and execution time. The proposed ensemble model exploits the expertise of each model and increases the generalization capability which is highly desirable for reservoir characterization that requires accurate predictions for maintaining efficient exploration and management of oil resources.