Higher Dimensional Cost Function for Synthesis of Evolutionary Algorithms by means of Symbolic Regression

This contribution deals with a new idea of how to create evolutionary algorithms by means of symbolic regression and Analytic Programming. The motivation was not only to tune some existing algorithms to their better performance, but also to find a new robust evolutionary algorithm. In this study operators of Differential Evolution (DE), SelfOrganizing Migrating Algortithm (SOMA), Hill Climbing (HC) and Simulated Annealing (SA) were used during a process of Analytic Programming. The results showed that AP was able to find successful as well as the original DE or SOMA. The cost function includes not only success in unimodal and multimodal benchmark function but also rules concerned to cost function evaluations. Results were tested on 16 benchmark functions in 2D, 20 D and 100 dimensional versions, i.e. 192 test, each was 100 times repeated and each of 100 repetitions has around 200 000 cost function evaluations. The results are presented in tabular and graphic form.