关键词: 指针网络, 遗传算法, 初始种群, 旅行商问题

Abstract:

Aimed to solve the problem that traditional genetic algorithm with poor-quality initial population easily traps into local optimum and has slow convergence speed when applied to traveling salesmen problem, this paper proposes a new algorithm model using pointer network to improve genetic algorithm for solving traveling salesmen problem. Pointer network is a effective deep learning neural network for solving combinatorial optimization problem which can provides high quality near-optimal solution. The new model uses the superior generation retention strategy and hamming distance roulette strategy on solution set from pointer network to make it become a high-quality and diversity initial population. Simulation experiment is compared new model with tradition genetic algorithm on various test instances from TSPLIB. Experimental results show that performance of new model is better than tradition genetic algorithm in both convergence speed and optimizing capacity. It is proved that pointer network hybrid genetic algorithm model is advantage on solving Traveling Salesmen Problem（TSP）.

Key words: pointer network, genetic algorithm, initial population, Traveling Salesmen Problem（TSP）