Performance Comparison of Derivativefree Optimization Methods

Structural optimization for parameters and non-parameters-based modelling are adopted in almost all area of science and engineering due to their inherent advantages in contrary to manual tuning. Typically, in the easiest case regression analysis is used to develop a data-based model for the future applications and forecasting. However, the aforementioned modelling approach may be suitable for well-defined data (i.e., elastic model) but might be difficult for complex data type where linear type model may show very poor performances. As a result, the adaptation of nonlinear type or complex models (e.g. a combination of linear and nonlinear parts) are unavoidable. For instance, modeling of a hysteresis type behavior would not be possible via any linear type models while a Bouc-Wen type nonlinear model would be a better choice. Hence, to avoid early mentioned issues and to achieve better performances, the derivative-free or searchalgorithm optimization might be suitable alternatives. Herein, due to the underlying advantages, the derivative-free search algorithms have been adopted to optimize model performances and interpret both linear and nonlinear type data. To be precise, two different derivative-free algorithms namely, (i) Nelder-Mead Simplex Method, and (ii) Genetic Algorithm have been studied. The outcome of this study shows that in both former mentioned cases search-based algorithms have better performance in terms of capturing the behavior of the true data via optimization. The real-life applications of optimization algorithms are numerous as these tools are used in many branches of science and engineering.