神经网络TSP问题仿真分析

描述了Hopfield神经网络和自组织特征映射神经网络解决TSP问题时的求解过程和仿真算法.通过对两种算法的仿真比较,得出以下结论:对于较大规模的TSP问题,SOFM模型的寻优结果要优于HNN模型寻优结果;HNN对网络模型参数和初始条件具有很强的依赖性且调整参数组合非常困难,而SOFM的参数设置和调整相对要简单得多;SOFM算法对待解决问题的拓扑分布不敏感,而HNN算法的收敛性对待求解问题的自身分布有很强的依赖性;当待求解问题的数目增大时,SOFM算法的运算时间增加缓慢,而HNN算法的运算时间增加较快.因此,在解决TSP问题时,自组织特征映射神经网络比Hopfield神经网络的效率高,随着问题规模的增大,其优势更为明显.

Simulation of Traveling Salesman Problem Based on HNN and SOFM

Hopfield neural network and self-organizing feature map neural network are utilized to solve traveling salesman problem.The arithmetic of software is given.Through comparing the two algorithms,SOFM method has the following advantages.First,as to large scale TSP,SOFM neural network generates sub-optional solutions,they are better than the results generated by Hopfield neural network(HNN).Second, HNN method depends on parameters and initial state strongly,but setting and adjusting parameters are quite easy in SOFM algorithm .Third,HNN method is sensitive to the topology of the problem but SOFM does not.Finally,when the scale of the problem increasing, the increase in the operation time for SOFM arithmetic is slow but that of HNN method is fast.The efficiency of SOFM neural network is higher than HNN in solving traveling salesman problems.The advantage is obvious as increase in the problem's scale.