New optimization algorithm based on free dynamic schema

Abstract

In this paper, we describe and test a new evolutionary algorithm based on the notion of a schema, which is designed to solve global optimization problems. We call it Free Dynamic Schema (FDS). It is a more refined variant of our previous DSC, DSDSC and MDSDSC algorithms. FDS processes two populations which are partially composed of the same chromosomes. The algorithm divides each population into several groups to which various genetic operators are applied: free dynamic schema, dissimilarity, similarity, and dynamic dissimilarity. Also, some new chromosomes are regenerated randomly. The FDS algorithm is applied to 22 test functions in 2, 4 and 10 dimensions. It is also compared with the classical GA, CMA-ES and DE algorithms. Moreover, the FDS algorithm is compared with the BA and PSA algorithms for some functions. In most cases, we have found the FDS algorithm to be superior to the classical GA and BA.