Experimental Comparison On Ann-based Solvers To Tsp Problems

TSP is one of the most typical NP-complete problems (Non-Deterministic Complete Problem) in combinational optimization, and also valuable both theoretically and practically. Since there is no polynomial time algorithm able to solve NP-complete problems, many intelligent optimization algorithms have been developed in order to achieve a solution to the TSP, such as simulated annealing algorithm, genetic algorithm and neural network. ANN methods are widely researched for their ability to handle large scale TSPs with low computation complexity.In this paper, we do experimental study and experimental comparisons on TSP measured by 2-dimensional Euclidean distance. We not only develope a computer program but also do experiments on 16 TSPs from the TSPLIB and 48 TSPs generated randomly by computer. We also compare our algorithm with KNIES, SA, Budinich and ESOM, and the experimental results show that for the large-scale TSP problems, both on deviation from the optimal path and on time performance, ORC_SOM algorithm is superior to KNIES, SA, Budinich and ESOM.