Multi-constrained optimization using bio-inspired approaches

Abstract

This thesis focuses on the optimization of constrained problems using bio-inspired algorithms, a growing field in artificial intelligence and computational optimization. Constrained optimization problems (COPs) are pervasive in real-world applications, requiring innovative methods to efficiently handle the inherent complexities. The main objective of this research is to develop and evaluate nature-inspired techniques, particularly focusing on Differential Evolution (DE) to solve various single objective constrained optimization problems. Also our research explore the application of the grey wolf optimizer bio-inspired algorithm in solving the Traveling Salesman Problem. This study begins by addressing the key challenges in COPs, focusing on handling constraint violation and maintaining feasibility within complex search spaces. To address these challenges, a novel adaptive coordinate system based on constrained differential evolution is proposed, where the DE is adaptively performed in either an Eigen coordinate system or original coordinate system. This flexibility enables the algorithm to dynamically direct the search, enhancing its ability to explore and exploit promising feasible regions, resulting in improved convergence rates and solution quality. Additionally, an enhanced Greedy Discrete Grey Wolf Optimizer (GD-GWO) is developed for discrete optimization, demonstrating superior performance against multiple benchmark instances of the Traveling Salesman Problem (TSP). The experimental results highlight the effectiveness of these bio-inspired algorithms in balancing exploration and exploitation, showing competitive performance compared to state-of-the-art techniques. The findings suggest that the proposed methods not only provide promising results for constrained optimization but also offer a foundation for solving more complex, real-world problems. Future work will investigate broader applications, multi-objective optimization, and further enhancements of these algorithms for better scalability and adaptability in diverse problem domains.