An extended equation of state modeling method. I. Pure fluids.

A new technique is proposed here to represent the thermodynamic surface of a pure fluid in the fundamental form. The peculiarity of the present method is the extension of a generic equation of state for the target fluid, which is assumed as basic equation, through the distortion of its independent variables by individual shape functions which are represented by a neural network function approximator. The basic equation of state for the target fluid can have the simple functional form of a cubic equation, as for instance the Soave-Redlich-Kwong one assumed in the present study. A set of nine fluids including hydrocarbons, haloalkane refrigerants, and strongly polar substances has been considered. For each of them the model has been regressed and then validated against volumetric and caloric properties generated in the vapor, liquid, and supercritical regions from highly accurate dedicated equations of state. In comparison with the underlying cubic equation of state, the prediction accuracy improves by a factor between 10 and 100, depending on the property and on the region. It has been verified that about hundred density experimental points, together with from ten to twenty coexistence data, are sufficient to guarantee high prediction accuracy for different thermodynamic properties. The method is a promising modeling technique for the heuristic development of multiparameter dedicated equations of state from experimental data.