Aiming at Travel Salesman Problem (TSP) that ant colony algorithm is easy to fall into local optima and slow convergence, a multi-colony ant algorithm using both generative adversarial nets (GAN) and adaptive stagnation avoidance strategy (GAACO) is proposed. First, to improve the convergence speed of the algorithm, we introduce a GAN model based on the game between convergence speed and solution quality. Then, to overcome premature convergence, an adaptive stagnation avoidance strategy is proposed. The strategy consists of two parts: (1) information entropy. It is used to measure the diversity of GAACO; (2) a cooperative game model. When the value of information entropy is less than threshold value, the cooperative game model will be used to select the appropriate pheromone matrix for different colonies to improve the accuracy. Finally, to further accelerate the convergence of the algorithm, the initial pheromone matrix is preprocessed to increase the pheromone of the optimal path for each iteration in the early stage. And according to reinforcement learning method, each colony increases the pheromone of the global optimal path at the end of each iteration. Extensive experiments with numerous instances in the TSPLIB standard library show that the proposed methods significantly outperform the state-of-the-art multi-colony ant colony optimization algorithms, especially in the large-scale TSPs.

中文翻译：

---

使用生成对抗网络和自适应停滞避免策略的多群体蚂蚁算法

针对旅行商问题(TSP)蚁群算法容易陷入局部最优、收敛速度慢的问题，提出了一种同时使用生成对抗网络(GAN)和自适应停滞避免策略(GAACO)的多蚁群算法。首先，为了提高算法的收敛速度，我们引入了一个基于收敛速度和求解质量博弈的 GAN 模型。然后，为了克服早熟收敛，提出了一种自适应停滞避免策略。该策略由两部分组成：（1）信息熵。用于衡量 GAACO 的多样性；(2)合作博弈模型。当信息熵值小于阈值时，将采用合作博弈模型为不同的菌落选择合适的信息素矩阵以提高准确率。最后，为了进一步加快算法的收敛速度，前期对初始信息素矩阵进行了预处理，增加了每次迭代的最优路径的信息素。并且根据强化学习方法，每个群体在每次迭代结束时增加全局最优路径的信息素。TSPLIB 标准库中大量实例的大量实验表明，所提出的方法明显优于最先进的多菌落蚁群优化算法，尤其是在大规模 TSP 中。每个群体在每次迭代结束时增加全局最优路径的信息素。TSPLIB 标准库中大量实例的大量实验表明，所提出的方法明显优于最先进的多菌落蚁群优化算法，尤其是在大规模 TSP 中。每个群体在每次迭代结束时增加全局最优路径的信息素。TSPLIB 标准库中大量实例的大量实验表明，所提出的方法明显优于最先进的多菌落蚁群优化算法，尤其是在大规模 TSP 中。