Modeling and analysis of the thermal conductivities of air saturated sandstone, quartz and limestone using computational intelligence

Abstract

Accurate experimental determination of the effective thermal conductivity (ETC) of porous reservoir rocks (especially under high temperature and pressure conditions) is a difficult problem and often a time-consuming and costly process. This study firstly examines the ability of the theoretical and empirical correlations for estimating the air saturated sandstone, quartz and limestone ETCs based on the models available in the literature. Optimal values of constant parameters of these correlations are found using the genetic algorithm (GA) technique. Empirical correlations have acceptable accuracy; however, they are not applicable in wide ranges of temperature and/or pressure. Also, each equation is dependent on the composition of the porous rock. In other words, they are not generalized correlations. The ability of artificial neural networks (ANN) and adaptive neuro-fuzzy inference system (ANFIS) as two generalized models are also investigated utilizing 872 experimental data points for a wide range of pressure and temperature. Temperature, pressure, porosity and bulk density are considered as the inputs of the mentioned models. An optimal topology of multi-layer perceptron neural network model (MLPNN) is determined via 10-fold cross-validation method. The total average absolute relative deviation (AARD (%)) of the developed ANN and ANFIS models for estimation of ETC are obtained 2.91% and 3.80%, respectively. The results show that the optimal ANN model is able to estimate ETC with the higher accuracy than the other correlations.