Development of a software toolbox for enhancing evolutionary algorithms with opposition-based learning strategies

This thesis initially presents a comprehensive literature review on strategies related to addressing the problem of local optima in evolutionary optimization algorithms. This research encompasses the fundamental concepts of evolutionary optimization, the significance of local optima issues, and the strategies developed to overcome this problem. Subsequently, the focus shifts to Opposition-Based Learning Methods, which are among the prominent approaches in the literature that enhance the exploration capability of algorithms by increasing the diversity of candidate solutions to achieve better results in optimization problems. This strategy is based on the idea of using opposite solutions to optimize a problem. The opposite solution often represents a complementary or deficient version of the initial solution, thereby aiming to increase diversity in the problem domain for improved outcomes. To investigate the impact of Opposition-Based Learning Methods on the performance of evolutionary algorithms, numerous comparative experimental studies have been conducted, leading to a comprehensive discussion. Finally, an open-source software toolkit has been developed, incorporating well-established evolutionary algorithms, benchmark test functions, and eight distinct variants of Opposition-Based Learning (OBL) methods— Classic OBL, Quasi-reflection OBL, Center-Based Sampling OBL, Quasi OBL, Current Optimum OBL, Random OBL, Super OBL, Generalized OBL. This toolkit provides researchers with the ability not only to solve complex optimization problems but also to examine and compare the performance of algorithms and OBL strategies from a rigorous and comparative perspective. Within the scope of the experimental studies, a total of 23 benchmark test functions (F1–F23), selected from standard benchmark sets, were employed to represent problems with varying levels of difficulty and diverse search space characteristics. These functions were specifically chosen to evaluate the overall performance of the proposed approaches. Moreover, the results were subjected to statistical analysis and were visualized through comprehensive graphical representations. In conclusion, the experimental and statistical findings have demonstrated that OBL strategies significantly improve the exploration–exploitation balance of evolutionary algorithms and enhance solution quality when applied in a manner sensitive to algorithm–problem compatibility.