Scholarlink

<p>Mobile Edge Computing (MEC) has emerged as a promising paradigm to offload computationally intensive tasks from resource-constrained IoT and mobile devices to proximal edge servers. However, existing solutions often rely on static pricing and centralized architectures, thereby limiting adaptability, energy efficiency, and user engagement. To address these challenges, we propose <span>FLBDP</span> (Federated Learning-Based Dynamic Pricing). This novel four-phase framework integrates wireless access modeling, federated learning, incentive design, and dynamic pricing to optimize computation offloading decisions. In Phase I, wireless access is modeled using Frequency Division Multiple Access (FDMA) to prevent interference and to compute per-user uplink data rates. Offloading decisions are made based on task complexity, device capability, and communication conditions. Phase II employs a federated learning strategy in which each device locally trains a model and transmits only updated parameters to the edge server. An attention-based aggregation mechanism is used to generate a global model that guides pricing policies. Phase III introduces a reputation-based incentive mechanism in which participants receive dynamic reputation scores based on the quality of their model, the timeliness of their training, and the similarity of their gradients. These scores determine eligibility for reward contracts, which are designed to encourage consistent participation. In Phase IV, the global model informs a dynamic pricing strategy that incorporates edge resource cost, network conditions, user demand, and energy usage. The pricing mechanism strikes a balance between execution efficiency, affordability, and profitability. Extensive simulations demonstrate that <span>FLBDP</span> effectively balances system efficiency and reliability. Compared with static pricing, it reduces task execution time by 22.5% and energy consumption by 16.0%. Furthermore, in terms of service stability, <span>FLBDP</span> outperforms the DQN-based approach by achieving a 54.5% reduction in queuing delay and 75.0% fewer dropped tasks, proving its robustness in handling dynamic workloads compared to baseline methods.</p>

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

Towards dynamic pricing for computation offloading in mobile edge computing: a federated learning approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月4日

查看全文
  • Roya Jahed,
  • Shahram Jamali,
  • Reza Fotohi

摘要

Mobile Edge Computing (MEC) has emerged as a promising paradigm to offload computationally intensive tasks from resource-constrained IoT and mobile devices to proximal edge servers. However, existing solutions often rely on static pricing and centralized architectures, thereby limiting adaptability, energy efficiency, and user engagement. To address these challenges, we propose FLBDP (Federated Learning-Based Dynamic Pricing). This novel four-phase framework integrates wireless access modeling, federated learning, incentive design, and dynamic pricing to optimize computation offloading decisions. In Phase I, wireless access is modeled using Frequency Division Multiple Access (FDMA) to prevent interference and to compute per-user uplink data rates. Offloading decisions are made based on task complexity, device capability, and communication conditions. Phase II employs a federated learning strategy in which each device locally trains a model and transmits only updated parameters to the edge server. An attention-based aggregation mechanism is used to generate a global model that guides pricing policies. Phase III introduces a reputation-based incentive mechanism in which participants receive dynamic reputation scores based on the quality of their model, the timeliness of their training, and the similarity of their gradients. These scores determine eligibility for reward contracts, which are designed to encourage consistent participation. In Phase IV, the global model informs a dynamic pricing strategy that incorporates edge resource cost, network conditions, user demand, and energy usage. The pricing mechanism strikes a balance between execution efficiency, affordability, and profitability. Extensive simulations demonstrate that FLBDP effectively balances system efficiency and reliability. Compared with static pricing, it reduces task execution time by 22.5% and energy consumption by 16.0%. Furthermore, in terms of service stability, FLBDP outperforms the DQN-based approach by achieving a 54.5% reduction in queuing delay and 75.0% fewer dropped tasks, proving its robustness in handling dynamic workloads compared to baseline methods.