| 摘要: |
| 针对边缘计算任务卸载中的位置隐私泄露问题，使用位置K-匿名技术为进行任务卸载的智能终端设备生成K-匿名区域，终端设备的任务通过K-匿名区域中的所有终端设备进行任务卸载，从而保护终端设备的位置隐私。提出了一种基于强化学习的位置隐私保护任务卸载（Reinforcement Learning Based Location Privacy Protection Mechanism for Task Offloading，RL-LPTO)算法，通过部署Actor和Critic网络来优化任务卸载决策，同时保护位置隐私，并在每个终端设备上设计了双部分Actor网络结构以实现任务转发和卸载决策，从而实现对智能体的训练并优化任务卸载的时延和能耗。仿真实验结果表明，RL-LPTO算法在保护位置隐私的同时将任务卸载性能的代价降低为基准算代价平均值的55%。 |
| 关键词: 边缘计算；位置隐私；K-匿名；任务卸载 强化学习 |
| DOI：10.20079/j.issn.1001-893x.240325002 |
|
| 基金项目:重庆市教育委员会在渝本科高校与中国科学院战略课题(HZ2021015） |
|
| Reinforcement Learning Algorithm for Task Offloading withLocation Privacy Protection |
| ZHANG Jian,HAN Yanni,AN Wei,TAO Tao,WANG Xuejian,FAN Dongyuan |
| (1.Institute of Information Engineering,Chinese Academy of Sciences,Beijing 100093,China;2.School of Cyber Security,University of Chinese Academy of Sciences,Beijing 100049,China;3.Key Laboratory of Cyberspace Security Defense，Chinese Academy of Sciences,Beijing 100093,China;4.China Mobile Information Technology Center,Beijing 100083,China) |
| Abstract: |
| For the problem of location privacy leakage in edge computing task offloading,position K-anonymous technology is used for task offloading of intelligent terminal equipment to generate K-anonymous area,and terminal equipment task is offloaded through all terminal equipment in K-anonymous area,to protect the location privacy of terminal devices.A reinforcement learning based location privacy protection mechanism for task offloading (RL-LPTO) is proposed,which deploys Actor and Critic networks to optimize task offloading decisions while preserving location privacy. A dual-part Actor network structure is designed on each terminal device to enable both task forwarding and offloading decisions, facilitating agent training and optimizing latency and energy consumption. Simulation results show that the RL-LPTO algorithm reduces the performance cost of task offloading to 55% of the average cost of baseline algorithms, while effectively protecting location privacy. |
| Key words: edge computing location privacy K-anonymity task offloading reinforcement learning |
