Adaptive elitist-ant system for solving combinatorial optimization problems

Abstract

This work investigates the effects achieved by using an adaptive elite pool over a basic population-based approach, such as, the ant system approach, to different combinatorial optimization problems (COPs). Optimization methods are generally considered a good methodology for solving COPs, while population-based approaches are considered better for exploring the search space and local-based approaches are considered better for exploiting the search space. However, some of these approaches are also hybridized in order to balance between exploiting and exploring the search space. Using an adaptive elite pool offers further balance while also adaptively tuning the importance parameter. Taken together, these three processes may improve and enhance the efficiency of any given search. In this work, three COPs (i.e. CVRP, STSP and 0-1 MKP) are used as test domains. In order to evaluate the effectiveness of using an adaptive elite pool, a comparison is made among the original ant system algorithm, the hybrid Elitist-Ant system algorithm, and other approaches drawn from the literature. Experimental results indicate that using an adaptive elite pool enhances the algorithm and is often able to produce better-quality solutions than other approaches. Where the adaptive elitist-ant system approach obtained 10 out of 14 best known results for CVRP, 20 out of 29 optimal solution results for STSP, 9 out of 11 best known results for 0-1 MKP, and generally better than other individuals approaches in term of better, worse, and similar results.