Encapsulated Evolution strategies for the determination of group contribution model parameters in order to predict thermodynamic properties

Abstract

The computation of parameters for group contribution models in order to predict thermodynamic properties usually leads to a multiparameter optimization problem. The model parameters are calculated using a regression method and applying certain error criteria. A complex objective function occurs for which an optimization algorithm has to find the global minimum. For simple increment or group contribution models it is often sufficient to use deterministically working optimization algorithms. However, if the model contains parameters in complex terms such as sums of exponential expressions, the optimization problem will be a non-linear regression problem and the search of the global optimum becomes rather difficult. In this paper we report, that conventional multimembered (Μ,λ)- and (Μ+λ.)-Evolution Strategies could not cope with such non-linear regression problems without further ado, whereas multimembered encapsulated Evolution Strategies with multi-dimensional step length control are better suited for the optimization problem considered here.