A population-based, steady-state procedure for real-parameter optimization

Despite the existence of a number of procedures for real-parameter optimization using evolutionary algorithms, there is still a need of a systematic and unbiased comparison of dierent approaches on a carefully chosen set of test problems. In this paper, we develop a steady-state, population-based search algorithm which allows the main search principles to be independently designed. The algorithm so developed is applied to a set of 25 test problems and results on 10 and 30 dimensions are presented. Although the proposed procedure cannot nd the exact optimum within the specied number of function evaluations, in most problems, the algorithm show steady progress towards the optimum. Moreover, it is also observed that the performance of the algorithm does not get aected by the rotation of the problems and embedded noise in function description. It would be interesting to compare the results with other contemporary evolutionary and classical optimization methods on the same set of test problems.