基于两阶段优化的并发流任务计算卸载策略

计算卸载作为移动边缘计算中最关键的技术之一而备受研究人员的关注，然而现有研究较少同时考虑拓扑结构、优化目标多样性及计算资源竞争的特性。针对移动边缘计算场景下的并发型数据流任务计算卸载及资源竞争问题，设计一种基于并发型数据流任务的多目标计算卸载混合整数模型，并给出一种基于多目标优化和多属性决策的两阶段优化框架对该模型进行求解。在多目标优化阶段，提出改进动态多种群并行NSGA-II（DMP-NSGA-II）算法，包括多种群多交叉策略、动态调整种群规模与二次局部搜索的改进策略，以解决局部收敛和全局搜索难以平衡的问题，同时设计一种基于混合式求解框架的DMP-NSGA-II算法求解多目标混合整数模型。在多属性决策阶段，提出一种基于模糊C均值聚类和灰关联投影法的后验选解方法，以选出在不同偏好下具有代表性的最优卸载决策。在测试函数和模型实例上的实验结果表明，设计的两阶段优化框架能够有效地求解所提出的模型，在ZDT系列测试函数上DMP-NSGA-II算法的HV和SP指标表现全面优于NSGA-II、MOEA/D和MOEA/D-DE算法，在模型实例上DMP-NSGA-II算法的Meantime和Meanenergy指标相较于基于混合式求解框架的NSGA-II算法，分别提升了30.1%和8.9%。

Computing Offloading Strategy for Concurrent Flow Tasks Based on Two-stage Optimization

Recently, computing offloading has attracted the attention of researchers as one of the most critical technologies in mobile edge computing. However, the existing research rarely considers the application topology, diversity of optimization objectives, and characteristics of the computing resource competition simultaneously.Therefore, a multi-objective computing offloading mixed integer model based on concurrent data-flow tasks is designed in this study to solve the problem of the computing offloading and resource competition of concurrent data-flow tasks in the Mobile Edge Computing (MEC) scene, and a two-stage optimization framework based on multi-objective optimization and multi-attribute decision-making is designed to solve it. First, an improved Dynamic Multi-Population parallel NSGA-II (DMP-NSGA-II) algorithm is proposed in the multi-objective optimization stage, including a multi-population multi-crossover strategy, a dynamic adjustment of the population size, and an improved strategy for quadratic local search, to solve the problem of the difficult balance between the local convergence and global search.Second, a DMP-NSGA-II algorithm based on a hybrid solution framework is designed to efficiently solve the multi-objective mixed integer model.Finally, a posterior solution selection method based on the Fuzzy C-Means clustering and Grey Relational Projection (FCM-GRP) method is designed to select the representative optimal unloading decision under different preferences.The results of the two simulation experiments on the test function and model example show that the designed two-stage optimization framework can effectively solve the proposed model.For the ZDT series of test functions, the HV and SP performances of the DMP-NSGA-II algorithm are significantly better than those of the NSGA-II, MOEA/D and MOEA/D-DE algorithms.For the model examples, the mean time and mean energy performances of the DMP-NSGA-II algorithm based on the hybrid solution framework are increased by 30.1% and 8.9%, respectively, compared with the NSGA-II algorithm based on the hybrid solution framework.