NOMA-MEC网络中基于优先级的多任务卸载策略

Priority-based multitasking offloading policy in NOMA-MEC network

针对NOMA-MEC网络中多任务卸载引起的资源分配不均、卸载成本过高等问题,考虑任务的异构性和网络环境的动态时变性,以最小化平均卸载成本为目标,面向超密集异构边缘网络提出了一种结合任务优先级的部分卸载策略。首先,充分利用资源,使用二分法模型化卸载比例的封闭解,将卸载问题解耦为任务优先级划分和基于服务器的信道资源分配两个子问题;然后,针对异构的任务,构建多维度任务优先级分类准则,提出了一种基于层次分析的支持向量机(analytic hierarchy-support vector machine, AH-SVM)任务分类方法,通过为不同的任务特征进行权重赋值,实现多任务优先级划分;最后,考虑动态环境下的信道质量,提出了一种面向信道资源分配和最佳卸载位置的NOMA信道增益深度双Q网络(NOMA channel gain deep double Q network, NCG-DDQN)任务卸载算法,有效降低了用户的平均卸载成本。实验结果表明,该算法在任务分类准确率和平均卸载成本方面较其他算法性能均有提升,同时验证了所提算法在高低优先级任务卸载过程中命中率的有效性。

To solve the problems of uneven resource allocation and excessive offloading cost when performing multi-task offloading in NOMA-MEC networks,this paper proposed a partial offloading strategy based on task priority for NOMA-assisted edge offloading networks in ultra-dense heterogeneous scenarios,taking into account the heterogeneous nature of the tasks and the dynamic time-varying nature of the network environment,with the goal of minimizing the average offloading cost.Firstly,to fully utilize the resources,it used the closed solution of the offloading ratio using a binary approach and decoupled into two subproblems:task priority partitioning and channel resource allocation problem based on server selection.Secondly,it constructed a multidimensional task priority measurement criterion for heterogeneous task,and proposed an analytic hierarchy-support vector machine(AH-SVM)task classification method based on hierarchical analysis to achieve task priority classification by assigning weight values to different task features.Finally,it proposed a NOMA channel gain deep double Q network(NCG-DDQN)algorithm for channel resource allocation and optimal offloading location to minimize the average offloading cost of users,considering the channel quality in dynamic environment.The simulation results show that the proposed algorithm performs better than other algorithms in terms of task classification accuracy and average offloading cost,and also verify the effectiveness of the proposed algorithm in terms of hit rate for offloading both high-and low-priority tasks.