Scholarlink

<p>Vehicle-to-Cloud (V2C) federated learning (FL) is a paradigm training model of intelligent transportation systems (ITS), however, typical pipelines are not well suited to the vehicular setting: dense gradient transmissions, synchronous aggregation, and standardized client involvement, together compound communication price, on-board battery consumption, convergence time, and increase operational carbon emissions. The current protocols (FedAvg, FedProx, SCAFFOLD, FedCPF, Batch-Aggregate) focus on efficiency or privacy separately but seldom a combination of mobility, energy, and carbon trade-off of V2C. VERDE-FL (Vehicular Energy- and Resource-aware Distributed Edge Federated Learning) is a lightweight green aggregation protocol that we propose to be used in hierarchical V2C systems (vehicles, RSUs, edge servers, cloud). VERDE-FL combines seven modules: energy-adaptive top-k sparsification, mobility-aware scheduling through dwell-time prediction, priority-aware aggregation with a trust-energy-data score, carbon-aware participation control, hierarchical pre-aggregation of the edges, hash-chained verifiable aggregation, and selective global redistribution. VERDE-FL achieved a smaller uplink volume (-21.2 to 35.4%), vehicle energy (-20.2 to 29.4%), aggregation latency (-17.9%), and CO 2e emissions (-24.1%) on a SUMO–OMNeT++–Veins–PyTorch–TFF testbed using the VeReMi dataset, with higher accuracy (2.3–4.2%) The protocol is resilient to model-poisoning, replay, and gradient-leakage attacks, and provides a viable implementation basis of sustainable V2C learning.</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

Lightweight Aggregation Protocols for Green Federated Learning in Vehicle-to-Cloud Systems

  • 期刊论文

  • 发表日期：

    2026年7月24日

查看全文
  • Sujit R. Wakchaure,
  • Neha Sangram Patil,
  • Pankaj Agarwal,
  • N. S. L. Kumar Kurumeti,
  • Ganta Jacob Victor,
  • B. Muthukumar

摘要

Vehicle-to-Cloud (V2C) federated learning (FL) is a paradigm training model of intelligent transportation systems (ITS), however, typical pipelines are not well suited to the vehicular setting: dense gradient transmissions, synchronous aggregation, and standardized client involvement, together compound communication price, on-board battery consumption, convergence time, and increase operational carbon emissions. The current protocols (FedAvg, FedProx, SCAFFOLD, FedCPF, Batch-Aggregate) focus on efficiency or privacy separately but seldom a combination of mobility, energy, and carbon trade-off of V2C. VERDE-FL (Vehicular Energy- and Resource-aware Distributed Edge Federated Learning) is a lightweight green aggregation protocol that we propose to be used in hierarchical V2C systems (vehicles, RSUs, edge servers, cloud). VERDE-FL combines seven modules: energy-adaptive top-k sparsification, mobility-aware scheduling through dwell-time prediction, priority-aware aggregation with a trust-energy-data score, carbon-aware participation control, hierarchical pre-aggregation of the edges, hash-chained verifiable aggregation, and selective global redistribution. VERDE-FL achieved a smaller uplink volume (-21.2 to 35.4%), vehicle energy (-20.2 to 29.4%), aggregation latency (-17.9%), and CO 2e emissions (-24.1%) on a SUMO–OMNeT++–Veins–PyTorch–TFF testbed using the VeReMi dataset, with higher accuracy (2.3–4.2%) The protocol is resilient to model-poisoning, replay, and gradient-leakage attacks, and provides a viable implementation basis of sustainable V2C learning.