Prediction of Porosity and Sand Fraction from Well Log Data using ANN and ANFIS: a comparative study

Reservoir characterization is a difficult problem due to nonlinear and heterogeneous physical properties of the subsurface. In this context, we present a case study to compare Artificial Neural Network (ANN) with Adaptive Neuro Fuzzy Inference System (ANFIS) for predicting two reservoir characteristics: porosity and sand fraction from well log data. The predictor variables are gamma ray content (GR), density (RHOB), P-sonic (DT), and neutron porosity (NPHI) logs. We use the data from a hydrocarbon field located in western part of India. It has been shown that while prediction results from both the models are comparable in terms of performance evaluators such as root mean square error, correlation coefficient etc., ANFIS has an inherent edge over ANN to describe and model uncertainties; whereas the computational complexity is high for ANFIS, making ANN a natural choice for the dataset under consideration. The research concludes that choice of the prediction model is closely associated with the nature of the dataset, and newer and complex methodologies need not always perform better in terms of accuracy of the result and computational complexity.