Reactive Max-min Ant System: an Experimental Analysis of the Combination with K-opt Local Searche

Ant colony optimization (ACO) is a stochastic search method for solving NP-hard problems. The exploration versus exploitation dilemma rises in ACO search. Reactive max-min ant system algorithm is a recent proposition to automate the exploration and exploitation. It memorizes the search regions in terms of reactive heuristics to be harnessed after restart, which is to avoid the arbitrary exploration later. This paper examined the assumption that local heuristics are useless when combined with local search especially when it applied for combinatorial optimization problems with rugged fitness landscape. Results showed that coupling reactive heuristics with k-Opt local search algorithms produces higher quality solutions and more robust search than max-min ant system algorithm. Well-known combinatorial optimization problems are used in experiments, i.e. traveling salesman and quadratic assignment problems. The benchmarking data for both problems are taken from TSPLIB and QAPLIB respectively.