面向异构AI训练任务的无线通信与计算资源联合调度方法OA

For edge-device collaborative AI training tasks in wireless networks,scenarios such as vehicular cooperative perception,industrial visual inspection,and background model updating for low-power IoT devices exhibit significant differences in split-layer feature size,computational load,device capability,and SLA constraints.Edge-side collaborative training therefore needs to perform resource scheduling under limited wireless bandwidth and edge computing resources.Due to device computing heterogeneity and wireless channel fluctuation,non-cooperative training or static resource allocation methods can hardly meet diverse training requirements,which may lead to latency violations or inefficient resource utilization.To address this issue,this paper proposes a joint wireless communication and computing resource scheduling method for heterogeneous AI training tasks.Based on Split Federated Learning(SFL),the model split point,wireless bandwidth,and edge computing frequency are integrated into a unified optimization framework.The end-to-end training latency and system energy consumption models are constructed,and task workload profiles together with differentiated service sensitivity coefficients are introduced to characterize the resource requirements of different training tasks.Simulation results show that,compared with the Comm-Only-SFL baseline that optimizes only communication resources,the proposed method reduces the average system cost by 38.89％,while saving training latency and total system energy consumption by 21.75％and 31.98％,respectively.These results verify the effectiveness of the proposed method for heterogeneous AI training tasks.