Research On Parameter Tuning Of Ant Colony Optimization

The researchers found that the determination of algorithm parameter is moretime-consuming than the design and implementation of the algorithm. This phenomenonis particularly prominent in the application of heuristic algorithm and metaheuristicalgorithm. The appropriate parameter setting has a vital impact on the performance of thealgorithm. The purpose of parameter tuning is to make the parameter tuning of algorithmsimplify from the perspective of systematic parameter adjustment. As a branch of SwarmIntelligence, inspired by ant foraging process choosing the shortest path, Ant ColonyOptimization (ACO) is a swarm intelligence stochastic search algorithm to mimic thebiological behavior of ants. With its advantages in solving the complex optimizationproblem, ACO is applied to a number of areas. As a result, it is important in both theoryand practice to consider parameter tuning of ACO from the systematic perspective.In this paper, the performance and principle of basic ant colony algorithm areanalyzed comprehensively and the effect of the parameter selection on theperformance of ant colony algorithm is thoroughly discussed. Three parametershave more affect on the algorithm, which are the heuristic factor α,the expectationheuristic factorβ,and the information persistent factorρ.On the basis of above,we propose a two-stage method to determine the optimal combination of parametersin ACO and impose this improved algorithm for solving the traveling salesmanproblem (TSP) to test the feasibility. The main work and innovation research resultsare as follows:(1)In the off-line stage we exploit the knowledge gained in an a priori tuningphase, where parameter values are further optimized based on specific TSPinstances. This approach use uniform design method to convert the problem ofparameter setting into the uniform design of multi-factor and multi-level and set thevalues of algorithm parameters with fewer tests quickly.(2)In the on-line stage we adapt the parameter values and find the appropriatesetting while solving an instance. Combining the advantages of uniform designmethod with chaos we draw chaotic perturbation into parameter tuning in ant colony optimization to avoid the search being trapped in local optimum. Theon-line tuning stage adapt typically very few key parameter of an algorithm basedon feedback from the search process to achieve the improvement of overallperformance. This paper not only adopt chaos theory to adjust static parameters ofthe algorithm but also combine uniform design method and chaos to adjust thealgorithm parameter dynamically on the basic of the results in the off-line stage.The experimental results show that this method is effective, showing that chaosadjustment and the mixing mechanism are superior.The above research plays an active role in the development of ACO, and offersan important policy for parameter tuning in ACO in practice to improve theeffectiveness of parameter optimization and the promotion and application of antcolony optimization.