A new transition rule for ant colony optimization algorithms: application to pipe network optimization problems

Ant algorithms are now being used more and more to solve optimization problems other than those for which they were originally developed. The method has been shown to outperform other general purpose optimization algorithms including genetic algorithms when applied to some benchmark combinatorial optimization problems. Application of these methods to real world engineering problems should, however, await further improvements regarding the practicality of their application to these problems. The sensitivity analysis required to determine the controlling parameters of the ant method is one of the main shortcomings of the ant algorithms for practical use. Premature convergence of the method, often encountered with an elitist strategy of pheromone updating, is another problem to be addressed before any industrial use of the method is expected. It is shown in this article that the conventional transition rule used in ant algorithms is responsible for the stagnation phenomenon. A new transition rule is, therefore, developed as a remedy for the premature convergence problem. The proposed transition rule is shown to overcome the stagnation problem leading to high quality solutions. The resulting ant algorithms are also found to be less sensitive to the sensitive to the sensitivity indexes, requiring less computational effort for the determination of these parameters. The efficiency and effectiveness of the proposed rule and the resulting algorithm is tested on some pipe network optimization benchmark problems and the results are compared with the existing results using ant algorithms and other evolutionary methods.