Soft computing-based nonlinear fusion algorithms for describing non-Darcy flow in porous media

By increasing the velocity of flow in coarse grain materials, local turbulences are often imposed to the flow. As a result, the flow regime through rockfill' structures deviates significantly from linear Darcy law; and nonlinear or non-Darcy flow equations will be applicable. Even though the structures of these nonlinear equations have some physical justifications, empirical studies are still necessary to estimate the parameters of these equations amid a great deal of uncertainty inherent to this estimation process. Consequently, none of the current empirical equations alone seem to be able to model the flow process exactly. In recent years, soft computing, in contrast to classical modeling techniques, has been advocated as a hybrid approach to intelligent paradigms such as neural networks, fuzzy logic, and neuro-fuzzy systems that aim to handle the uncertainties and vagueness in such systems. In this paper, we investigate several soft computing paradigms to combine three of the most commonly validated and utilized empirical solutions in the current literature. In this way, the estimates from the three empirical equations drive a soft computing-based fusion algorithm. The results show that soft computing-based approaches provide a powerful paradigm with a strong ability to model reality. Specifically, this paper concludes that cascade correlation neural networks provide the best fusion algorithm with the highest accuracy among the considered conventional alternatives as well as several other soft computing paradigms.