人工神经网络求解TSP问题的改进算法研究

Study on Improved Method of Neural Network to Solve TSP

在算法优化问题的研究中,旅行商问题(TSP)是组合优化领域里的一种描述简单而难以处理的NP完全难题,为解决Hopfield神经网络求解TSP问题时易出现无效解和收敛性能差的问题,提出一种对能量函数"行"、"列"项进行严格约束并在神经元动态方程中使用软限幅函数的改进算法。在参数优化方面进行了分析并选取了最优参数值,与经典Hopfield神经网络TSP求解方法进行比较。对10个城市仿真研究,实验结果表明:改进算法能使网络函数达到全局搜索从而避免无效解的产生,求得的最优解个数多于原始算法,迭代次数少且易达到有效解。

In the study of algorithms optimization problems, the Traveling Salesman Problem （TSP） is described as a simple and difficult to deal with NP complete problem in the field of combinatorial optimization. For the Hopfield network in solving the TSP often getting un valid and poor convergence problems, we proposed an im proved algorithm about strictly bound the energy function＇s ＂row＂ ,＂ column＂ items and used the soft limiter function in neuronal dynamics equations. In the parameter optimization aspects, we analyzed and selected the optimal parame ter value, compared with the classic Hopfield neural network to solve the TSP problem. Simulation was carried out with 10 cities and the experimental results show that the improved algorithm can achieve global searching to avoid generating un valid, the number of optimal solutions is more than the original algorithm, and easy to achieve effec tive solution with fewer iterations.