基于图知识迁移的蚁群算法参数选择

Parameters selection for ant colony algorithms based on graph knowledge transfer

同时考虑蚁群算法的所有运行参数,提出一种基于图知识迁移的蚁群算法参数选择方法.首先,将包含知识(蚁群算法的运行参数)的源任务映射到一个高维的迁移空间,并通过迁移权值连接不同的源任务,构造一个模型迁移图;然后,扩展模型迁移图使其包含目标任务,并利用图论的知识学习迁移函数;最后,通过最小二乘法自主地给目标任务分配一个优化的运行参数组合.机器人路径规划问题的仿真结果验证了该方法的智能性、快速性与合理性.

A kind of parameters selection for ant colony algorithms（ACAs） based on graph knowledge transfer is proposed, where all of running parameters are taken into account simultaneously. Firstly, all source tasks containing knowledge （running parameters for ACAs） are mapped onto a high-dimensional transfer space, and transfer weights are used to connect these source tasks. In this way, a model transfer graph is thus constructed. Then, the model transfer graph is extended to include a target task and a transfer function can be obtained according to a graph theory. Finally, a group of optimal parameters for the.. target task can be automatically determined by using a least-squares method. Simulation results involving a robot path planning problem show the intelligence, rapidness and reasonability of the proposed method.