Scholarlink

In the industrial field, the requirements for deterministic and reliable communication cannot be satisfied by current end-edge-cloud collaborative computing (EECC) networks. Time-sensitive networking (TSN) technology not only provides bounded latency guarantees but also gives support to integrate different types of traffic on a single network. However, the development of TSN-based EECC networks remains in its early stage. In light of this, we propose an intelligent end-edge-cloud collaborative deterministic task offloading framework that combines EECC and TSN networks. Based on Distributed Deep Reinforcement Learning (DDRL), a deterministic task offloading algorithm named DDRL-E2CDTO is introduced to address the joint optimization problem of TSN traffic scheduling and end-edge-cloud task offloading. This algorithm employs TSN traffic scheduling (TTS) networks and deterministic task offloading (DTO) networks to train distributed agents to make instant and efficient deterministic decisions of task offloading. Simulation experiments demonstrate that our proposed approaches can obtain significant performance achievements in acceleration of convergence speed and computation rate.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Distributed Deep Reinforcement Learning Based Deterministic Task Offloading in End-Edge-Cloud Collaborative Computing Networks

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yingfei Yao,
  • Shunchun Yao,
  • Nan Zhou,
  • Xiaojun Liang,
  • Chunhua Yang,
  • Weihua Gui

摘要

In the industrial field, the requirements for deterministic and reliable communication cannot be satisfied by current end-edge-cloud collaborative computing (EECC) networks. Time-sensitive networking (TSN) technology not only provides bounded latency guarantees but also gives support to integrate different types of traffic on a single network. However, the development of TSN-based EECC networks remains in its early stage. In light of this, we propose an intelligent end-edge-cloud collaborative deterministic task offloading framework that combines EECC and TSN networks. Based on Distributed Deep Reinforcement Learning (DDRL), a deterministic task offloading algorithm named DDRL-E2CDTO is introduced to address the joint optimization problem of TSN traffic scheduling and end-edge-cloud task offloading. This algorithm employs TSN traffic scheduling (TTS) networks and deterministic task offloading (DTO) networks to train distributed agents to make instant and efficient deterministic decisions of task offloading. Simulation experiments demonstrate that our proposed approaches can obtain significant performance achievements in acceleration of convergence speed and computation rate.