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

Thermal conductivity (TC) is among the most important characteristics of porous media for hydrocarbon reservoir thermal simulation and evaluating the efficiency of the thermal enhanced oil recovery process. In this study a two-layer artificial neural network (ANN) approach is proposed for estimating the effective TCs of dry and oil saturated sandstone at a wide range of environmental conditions. Temperature, pressure, porosity, bulk density of rock, fluid density and oil saturation are employed as independent variables for prediction of effective TCs of sandstone. Various types of ANN such as multi-layer perceptron (MLP), radial basis function, generalized regression and cascade-forward neural network have been examined and their predictive capabilities are compared. Statistical errors analysis confirms that a two-layer MLP network with seven and 15 hidden neurons are optimal topologies for modeling of TC of oil saturated and dry sandstone, respectively. The predictive capabilities of the optimal MLP models are validated by conventional recommended correlation and a large number of experimental data which were collected from various literatures. The predicted effective TC values have a good agreement with the experimental TC data, i.e., an absolute average relative deviation percent of 2.73% and 3.81% for the overall experimental dataset of oil saturated and dry sandstone, respectively. The results justify the superiority of the optimal MLP networks over the other considered models in simulation of the experimental effective TCs of dry and oil saturated sandstones.