Evolutionary algorithm; optimisation; non-linear regression; parameter-estimation

A lot of new optimisation methods are developed in order to achieve better results when solving the global optimisation problem. A group of algorithms inspired by natural systems are called Evolutionary Algorithms (EA). One of the most popular EA used in many theoretical and practical studies and applications is known as Differential Evolution (DE). State-of-the-art optimisation methods provide very good results on many benchmark optimisation problems where the true solution is known. Nevertheless, the efficiency of state-of-the-art algorithms on real optimisation problems is not possible to evaluate because the true solution is not known. This experimental study is focused on application and comparison of several state-of-the-art EAs on a simple real optimisation problem, especially DE. The problem of estimation of parameters in non-linear regression is used to evaluate selected well-known approaches. In 2007, several EAs were applied to this optimisation problem. Methods used in this study are outperformed by newly proposed methods in many problems of various benchmark sets. For this purpose, well-performing EAs are selected to show the efficiency of these methods on this simple real optimisation problem. The known set of 27 non-linear regression models (NIST) is used to evaluate employed algorithms.