Modeling and analysis of thermal conductivity of sandstone at high pressure and temperature using optimal artificial neural networks

Thermal conductivities (TC) of porous media are among the most important information for hydrocarbon bearing reservoir thermal simulation andassessing the efficiency of thermal enhanced oil recovery process, both for the scientific purposes and engineering design. In the present study a novel method for estimation of effective TCs of dry sandstone at a wide range of pressure and temperature has been proposed. Multi-layer perceptron neural network (MLPNN) with optimal configuration was employed to model the effective TCs of sandstone as a function of temperature,pressure, porosity and density. Statistical error analysis confirmed that a MLP network consisting of only one hidden layer composed of fifteen eurons exhibited the best generalization results and therefore can be considered as an optimal topology. The capability of the optimal MLPNN model wasvalidated and benchmarked by its application to experimental effective TCs which were collected from various literatures. The collected experimentaldata were randomly divided into two training and testing data set. Application of the optimized MLP model for 255 experimental effective TCs data gavean absolute average relative deviation percent (AARD%) of 3.63% and 4.47% for the training and testing subsets respectively. The proposed model alsoindicated the 0.97427 for the square of the correlation coefficient (R2) over total data set. Furthermore, the predictive capability of the proposed technique was compared with that of conventional recommended model in the literature. The comparison of the results showed that the proposed neural network is superior to the considered method, with respect to accuracy as well as extrapolation capabilities. The resultsjustify that the proposed optimal MLPNN model can simulate the effective thermal conductivity of sandstones with acceptable error and present excellent agreement with experimental data