| 摘要: |
| 针对具有依赖关系的计算密集型应用任务面临的卸载决策难题，提出了一种基于优先级的深度优先搜索调度策略。考虑到用户能量受限和移动性，构建了一种联合用户下行能量捕获和上行计算任务卸载的网络模型，并在此基础上建立了端到端优化目标函数。结合任务优先级及时延约束，利用深度强化学习自学习的优势，将任务卸载决策问题建模为马尔科夫模型，并设计了基于任务相关性的Dueling Double DQN（D3QN）算法对问题进行求解。仿真数据表明，所提算法较其他算法能够满足更多用户的时延要求，并能减少9%~10%的任务执行时延。 |
| 关键词: 协同边缘网络 移动边缘计算 计算卸载 深度强化学习 |
| DOI：10.20079/j.issn.1001-893x.220728002 |
|
| 基金项目:国家自然科学基金资助项目(62101277)；江苏省自然科学基金(BK20200822) |
|
| Intelligent Computation Offloading and Resource Optimization Algorithm in Collaborative Edge Networks |
| LI Bin,XU Tiancheng |
| ((1.School of Computer Science,Nanjing University of Information Science and Technology,Nanjing 210044,China;2.Jiangsu Collaborative Innovation Center of Atmospheric Environment and Equipment Technology,Nanjing University of Information Science and Technology,Nanjing 210044,China)) |
| Abstract: |
| In order to solve the problem of offloading decision for computation-intensive tasks in dependency-aware edge networks,a depth-first search scheduling strategy based on task priority is proposed.By taking into account the limited energy and high mobility of users,the network model of joint downlink energy harvesting and uplink computing task offloading is built.Furthermore,the device-to-device optimization objective function is formulated under the constraints of the latency and the task priority and the task offloading problem is modeled as a Markov decision process.By exploiting the advantage of self-learning of deep reinforcement learning,the Dueling Double DQN(D3QN) algorithm based on task dependency is designed to tackle it.Numerical results show that the proposed method can meet the delay requirements of more users and reduce the completion delay up to 9%~10% against other existing schemes. |
| Key words: collaborative edge network mobile edge computing computation offloading deep reinforcement learning |
