Modeling and Analysis of Dry Sandstone, Quarts and Limestone Thermal Conductivities Using Optimal Artificial Neural Networks and Adaptive Neuro-Fuzzy Inference Systems

The effective thermal conductivity (ETC) of porous reservoir rocks depends on various thermo-physical properties. Although accurate laboratory determination of the ETC for porous materials is a difficult problem and often a costly and time-consuming process (especially under high temperature and pressure conditions), measurement of the ETCs of dry and fluid-saturated porous materials, have still been widely conducted in various scientific research fields such as petroleum and natural gas geology. In the present study, two novel methods are proposed to estimate the ETCs of dry sandstone, quarts and limestone utilizing 896 experimental data in a wide range of pressure and temperature. Optimal configurations of multi-layer perceptron neural networks (MLPNN) and adaptive neuro-fuzzy inference systems (ANFIS) are employed to model the ETCs of three conventional rock types as a function of their temperature, pressure, porosity and density. Statistical error analysis confirms that a MLPNN including of only one hidden layer composed of thirteen neurons exhibits the best results (AARD=%3.0066, MSE=0.0206 and R2=0.9851) and also ANFIS model with four, three, three and five input membership functions for temperature, pressure, density and porosity respectively, is selected as optimal structure (AARD=%2.5081, MSE=0.0104 and R2=0.9925). Both these novel models have better results than empirical correlations.

کلیدواژه ها:

Effective thermal conductivity ، artificial neural network ، adaptive neuro-fuzzy inference systems ، sandstone ، quarts ، limestone