No abstract

In contrast to many successful applications of ant colony optimization, the theoretical foundation is rather weak. It greatly limits the application in practical problems. One problem, called solution quality evaluation, is how to quantify the performance of the algorithm. It is hardly solved by theoretical methods. Experimental analysis method based on the analysis of search space and characteristic of algorithm itself is proposed in this paper. As algorithm runs, it would produce a large number of feasible solutions. After preprocessing, they were clustered according to distance. Then, good enough set was partitioned by the results of clustering. Last, evaluation result of ordinal performance was got by using relative knowledge of statistics. As the method only uses feasible solution produced by optimization algorithm, it is independent to specific algorithm. Therefore, the proposed method can be adopted by other intelligent optimization algorithms. The method is demonstrated through traveling salesman problem.