面向调控信息新鲜度保障的电力至简物联网资源优化

Dispatching and control information freshness guaranteed resource optimization in simplified power Internet of things

信息新鲜度对分布式能源调控模型训练精度具有重要影响。信息新鲜度较差会导致训练模型损失值增加,降低调控可靠性与经济性,影响能量实时供需平衡。电力至简物联网能够为分布式能源调控提供即插即用、多模态融合的通信支撑,但仍面临跨域资源优化与模型训练适配性差、调控信息新鲜度难以保障等挑战。针对上述问题,提出基于调控信息新鲜度感知的通信与计算资源协同优化算法,通过赤字虚拟队列演进感知调控信息新鲜度偏差。在此基础上,利用深度Q网络学习信道分配与批量规模联合优化策略,最小化模型损失函数,保障调控信息新鲜度长期约束。仿真结果表明,相较于基于联邦深度强化学习的低时延资源分配算法与自适应联邦学习批量规模优化算法,所提算法使全局损失函数降低57.19%和24.60%,信息新鲜度提高35.34%和49.05%。

Information freshness conducts an important impact on the training accuracy of the distributed energy dis-patching and control model.Poor dispatching and control information freshness will increase the loss function of the training model,reduce the reliability and economy of dispatching and control,and effect the real-time balance of energy supply and demand.Simplified power Internet of things can provide plug-and-play and multi-mode fusion communica-tion support for distributed energy dispatching and control,but it still faces challenges of the inadaptability between cross-domain resource optimization and model training,and the difficulty in guaranteeing dispatching and control infor-mation freshness.To solve the above challenges,an information freshness aware-based communication-and-computation collaborative optimization algorithm(IFAC3O)was proposed,and the information freshness deviation was regulated by the awareness of deficit virtual queue evolution.On this basis,IFAC3O leveraged deep Q network and dispatching and control information freshness awareness to learn the channel allocation and batch size joint optimization strategy,thereby minimizing model loss function while guaranteeing long-term dispatching and control information freshness constraints.Compared with the federated DRL based low-latency resource allocation algorithm and adaptive federated learning-based batch size optimization algorithm,IFAC3O can reduce global loss function by 63.29%and 38.88%as well as improve information freshness by 20.59%and 57.69%.