Abstract
The paper presents the experimental results of some tests conducted with the purpose to gradually and cumulatively improve the classical DE scheme in both efficiency and success rate. The modifications consisted in the randomization of the scaling factor (a simple jitter scheme), a more efficient Random Greedy Selection scheme, an adaptive scheme for the crossover probability and a resetting mechanism for the agents. After each modification step, experiments have been conducted on a set of 11 scalable, multimodal, continuous optimization functions in order to analyze the improvements and decide the new improvement direction. Finally, only the initial classical scheme and the constructed Fast Self-Adaptive DE (FSA-DE) variant were compared with the purpose of testing their performance degradation with the increase of the search space dimension. The experimental results demonstrated the superiority of the proposed FSA-DE variant.