Porovnanie optimalizačných algoritmov pri spektrálnom ladení mechanickej sústavy

Abstract

This article deals with a comparison of various, gradient-based optimisation algorithms in terms of their accuracy and effectivity. The compared algorithms are the steepest descent method (SDM) and the most well-know quasi-Newton methods. The presented methods were applied to the spectral tuning of a simple two degree of freedom (DOF) mechanism, in order to evaluate their performance. The obtained results were statistically processed and utilised to compare the algorithms, based on their accuracy and overall effectivity. The results show that the quasi-Newton methods are superior to the SDM in terms of both the accuracy and computing time. Moreover, the overall performance of these methods is also significantly less influenced by the selection of starting point. Thus, the obtained results render the quasi-Newton methods as a significantly better choice, compared to the standard SDM.