Advancing Relative Permeability Estimation Through Data-Driven Modeling

Abstract

AbstractAvailability of large amounts of data helps in developing data-driven models using state of the art Artificial intelligence (AI) methodologies. Relative permeability is an important parameter used by reservoir engineers and are usually accurately obtained from laboratory experiments, which are relatively expensive. Therefore, AI can play an important role in developing models to predict relative permeability accurately without extensive lab procedures. Accordingly, this work presents application of two AI algorithms namely, Artificial Neural Network (ANN) and Adaptive Neuro-Fuzzy Inference System (ANFIS). Furthermore, two novel mathematical correlations are extracted from the ANN model to predict relative permeability of oil/water in oil- and water-wet environments. The input data, obtained from literature, for the development of AI models include porosity, rock absolute permeability, initial water saturation, residual oil saturation, wettability index and water saturation.A customized workflow is applied to ensure proper data is fed into the AI models. In addition, a rigorous sensitivity analysis is performed within the framework of this workflow. This analysis involves running multiple realizations with varying number of neurons, resulting in various weights and bias for the ANN model. Moreover, ANFIS model is tuned using various cluster sizes to result in the most optimum value. Finally, the optimized ANN and ANFIS models are compared using the Root Mean Squared Error (RMSE) and correlation coefficient (R2) analysis when applied to a blind dataset comprising of more than 300 data points. The analysis illustrates that the ANN model is relatively better in predicting relative permeability values to both, oil, and water. On the other hand, analysis of the ANFIS model shows that it yields high error values when tested on unseen dataset. Also, unlike the ANN mode, it does not provide an actual mathematical correlation. This work presents alternate data-driven artificial intelligence models which will lead to quicker and cheaper relative permeability estimates.