Parameter Estimation Strategies in Thermodynamics

Abstract

Many thermodynamic models used in practice are at least partially empirical and thus require the determination of certain parameters using experimental data. However, due to the complexity of the models involved as well as the inhomogeneity of available data, a straightforward application of basic methods often does not yield a satisfactory result. This work compares three different strategies for the numerical solution of parameter estimation problems, including errors both in the input and in the output variables. Additionally, the new idea to apply multi-criteria optimization techniques to parameter estimation problems is presented. Finally, strategies for the estimation and propagation of the model errors are discussed.

Modeling, Simulation, and Optimization 4.0 for a Distillation Column

“…For solving the parameter estimation problem, there exist several options for the objective function to be optimized (cf. ). If the inputs of the simulation are set equal to the measured values, estimating the parameters can be done by solving a standard least‐squares problem.…”

Application of Quantum Chemistry Insights to the Prediction of Phase Equilibria in Associating Systems

“…Thermodynamic properties such as speed of sound, second virial coefficient, enthalpy of vaporization, and heat capacity can be included when parameterizing both nonassociating and associating systems. , Practical challenges with this approach include the availability of experimental data as well as assigning weights to each additional property in the objective function. Strategies are available to guide the optimal selection of weights (Pareto optimal solutions visualized on a Pareto front), but these techniques have not yet been widely applied to parameterizing thermodynamic models. , …”

Some Aspects of Combining Data and Models in Process Engineering

“…For example, the thermodynamic behavior of mixture properties is often described by models derived from excess enthalpy or equation of state models. The parameters of these models have been adjusted by regression techniques to measurement data , , from which not only estimates of the parameters themselves, but also of their statistical uncertainties result. It would be highly desirable to have all these regression results directly available in simulation of separation devices which make use of the thermodynamic quantities.…”