军事知识服务平台
基于集中式反向拍卖的物理-虚拟同步资源分配联合决策机制
A Reverse Auction-Based Joint Decision-Making Mechanism for Physical-Virtual Synchronization Resource Allocation
摘要 实时物理-虚拟同步是实现高效的数字孪生辅助边缘计算网络(即边缘数字孪生网络)的核心任务,涉及同步对匹配、资源分配、同步优化与任务卸载等一系列相互耦合的环节.该任务需要综合考虑边缘数字孪生网络中的同步对特性、边缘节点的计算与通信能力以及通信覆盖、异构资源容量和时延等复杂本地约束条件,本质上是一种复杂的多维度联合决策问题.然而,现有工作尚未对这些紧密耦合的联合决策优化问题进行系统分析与有效解决.针对此问题,本文提出了一种基于集中式反向拍卖的物理-虚拟同步资源分配联合决策机制(ARAPVS),实现了边缘数字孪生网络系统的稳定、实时物理-虚拟同步.该机制由边缘数字孪生网络中的云服务中心(CSC)进行统一任务调度与资源分配,其核心目标是在满足物理约束的前提下,最大化包含同步实时性收益在内的系统社会福利.具体而言,该机制首先引入数字孪生状态陈旧度模型以量化同步质量,并构建广义成本函数;其次,采用匈牙利-库恩(KM)算法结合机会成本策略求解任务与异构资源束的最小广义权完美匹配,有效解决了多用户资源争用问题;最后,设计了基于 VCG 的支付规则以确保边缘节点如实报价并获得非负收益.理论分析证明了该机制的真实性与个体理性,并分析了其计算复杂度.Python 平台的大量仿真表明:与随机分配(RA)、贪心分配(GA)、McAfee 机制(MM)以及最新的深度强化学习算法(DDPG)相比,ARAPVS 不仅能够有效降低边缘节点总成本,更在平均累积陈旧度和社会福利方面表现出显著优势.在不同规模的同步对和边缘节点网络下,ARAPVS 均表现出最优的综合性能和良好的可扩展性,验证了所提机制的有效性.
Abstract:
With the rapid evolution of mobile edge computing and the upcoming sixth-generation wireless networks,the digital twin-aided edge network has emerged as a transformative paradigm for bridging the physical and virtual worlds.In such networks,real-time physical-virtual syn-chronization is the fundamental task required to maintain high-fidelity digital replicas and enable precise closed-loop control.This synchronization process is inherently complex,involving two tightly coupled directions:physical-to-virtual state reporting and virtual-to-physical decision feed-back.However,achieving stable and real-time synchronization in dynamic edge environments re-mains a daunting challenge.This challenge arises from the heterogeneity of edge resources,strin-gent latency requirements,and the complex interdependencies between synchronization pair matching,resource allocation,and task offloading.Existing research often treats these dimen-sions as isolated sub-problems,failing to provide a systematic and integrated optimization frame-work for these tightly coupled decisions. To address these issues,this paper proposes ARAPVS,a centralized joint decision-making mechanism based on a reverse auction model.The objective is to maximize total social welfare while satisfying constraints including limited communication coverage,heterogeneous resource capacities,and maximum tolerable delay thresholds.At the architectural level,the system fol-lows the design principle of control and data plane separation to avoid communication bottle-necks.The cloud service center acts as a centralized controller and market clearing authority,processing only lightweight metadata via a dedicated signaling channel.Once an auction decision is made,it issues admission tokens to winning nodes.Actual high-bandwidth synchronization da-ta is transmitted directly between physical devices and edge nodes via authorized peer-to-point links,ensuring system scalability and low latency. The ARAPVS mechanism integrates several key technical innovations.First,we introduce a digital twin state staleness model to quantify synchronization quality,which is incorporated into a generalized cost function.Second,we formulate resource contention as a minimum generalized weight perfect matching problem in a bipartite graph.We employ the Kuhn-Munkres algorithm combined with an opportunity cost strategy to achieve a global optimal matching.Unlike un-weighted matching algorithms,the Kuhn-Munkres algorithm handles weighted edges with cubic time complexity,ensuring both theoretical optimality and computational feasibility for large-scale networks.Third,we design a pricing rule based on the Vickrey-Clarke-Groves mechanism.Edge nodes serve as resource sellers submitting competitive bids,while synchronization pairs act as buyers.The cloud service center automatically calculates settlement prices based on social oppor-tunity cost,ensuring that truthful bidding is a weakly dominant strategy. Rigorous theoretical analysis confirms the mechanism satisfies individual rationality and in-centive compatibility.Extensive numerical simulations on the Python platform demonstrate the superiority of ARAPVS.We compared our mechanism with several baselines,including random allocation,greedy allocation,the McAfee mechanism,and deep reinforcement learning algo-rithms.Quantitative results indicate that ARAPVS significantly reduces total costs and maintains the lowest average accumulative staleness.Notably,the mechanism exhibits robust convergence and sub-linear execution time growth even as the network scales up to hundreds of edge nodes.This research provides a solid theoretical and technical foundation for enabling efficient,real-time synchronization in next-generation edge digital twin ecosystems.
刊 名 计算机学报 ISTICEIPKU
年,卷(期): 2026, 49(9)
分类号: TP18
机标分类号: TP391 TN929.11 O157.5
在线出版日期: 2026年10月8日