陈晓红1,2,3,4,龚思远1,2,袁依格2*,曹文治3,4,王傅强1 (1.中南大学商学院,长沙410083;2.湘江实验室,长沙410205;3.湖南工商大学前沿交叉学院,长沙410205;4.湖南工商大学管理科学与工程学院,长沙410205) 摘要:随着数字孪生、工业物联网、边缘智能与元宇宙等虚实技术的快速演进,虚实融合已成为推动智能社会构建与产业体系重塑的核心驱动力。算力作为虚实融合的底层支撑要素,正在从单一集中式计算资源向多层协同、智能调度与安全可信的复杂系统加速演化。本文系统梳理了新型算力体系的发展现状与关键特征,指出当前算力体系正呈现“云‒边‒端”一体化演进的趋势,智能算力正成为算力结构升级的核心引擎,区域算力布局逐步形成了差异化与协同并重的格局,虚实融合驱动下的算力应用模式亦呈现出多样化、泛在化与自主化的特征;在虚实融合应用背景下,进一步分析了支撑新型算力体系构建的关键技术,包括虚实融合驱动的算力体系架构设计、面向虚实融合场景的关键技术要素,从体系架构与算力编排两方面揭示了算力供需匹配的逻辑基础;通过对混合计算架构的研究,重点探讨了虚实融合的算力体系在异构协同、低延迟高带宽保障、多源数据安全与隐私保护等方面面临的挑战;针对上述挑战,提出了构建泛在智能算网、发展可信算力体系、突破异构协同壁垒、完善安全治理机制与培育虚实算力生态等新型算力体系发展重点方向,为未来虚实算力体系的建设、产业生态优化以及算力资源配置策略提供理论参考与战略支撑。 关键词:虚实融合;算力体系;混合计算架构;“云‒边‒端”协同;隐私保护中图分类号:F420文献标识码:A Development and Application of ComputingArchitecture for Virtual‒Reality Integration Chen Xiaohong1,2,3,4,Gong Siyuan1,2,Yuan Yige2*,Cao Wenzhi3,4,Wang Fuqiang (1.Business School,Central South University,Changsha410083,China;2.Xiangjiang Laboratory,Changsha410205,China;3.Advanced Interdisciplinary Studies,Hunan University of Technology and Business,Changsha410205,China;4.School ofManagement Science and Engineering,Hunan University of Technology and Business,Changsha410205,China) Abstract:With the rapid evolution of virtual reality technologies such as digital twins,industrial Internet of Things,edge intelligence,and metaverse,virtual ‒ reality integration has become a core driving force for the construction of an intelligent society and thereshaping of industrial systems.Computing power,as the underlying supporting element of virtual ‒ reality integration,is rapidlyevolving from a single centralized computing resource to a complex system characterized by multi-layered collaboration,intelligentscheduling,security,and trustworthiness.This study reviews the current development status and key characteristics of new computingpower systems,pointing out that current computing power systems are showing a trend of cloud‒edge‒device integration,intelligentcomputing power is becoming the core engine for upgrading computing power structures,regional computing power layouts are 面向虚实融合的算力架构发展与应用探讨 gradually forming a pattern that emphasizes both differentiation and collaboration,and computing power application models driven byvirtual‒reality integration are also showing diversified,ubiquitous,and autonomous characteristics.In the context of virtual‒realityintegration,this study further analyzes the key technologies supporting the construction of new computing power systems,includingthe architecture design of computing power systems driven by virtual‒reality integration and the key technical elements for virtual‒reality integration scenarios,revealing the logical basis for matching computing power supply and demand from both systemarchitecture and computing power orchestration perspectives.Through research on hybrid computing architectures,this study focuseson discussing their practical bottlenecks in areas such as heterogeneous collaboration,low-latency and high-bandwidth assurance,multi-source data security,and privacy protection.To address the aforementioned bottlenecks,we propose key development directionsfor new computing power systems,including constructing ubiquitous intelligent computing networks,developing trusted computingpower systems,breaking heterogeneous collaboration barriers,improving security governance mechanisms,and cultivating a virtual‒reality computing power ecosystem.These will provide theoretical references and strategic support for the construction of futurevirtual‒reality computing power systems,optimization of industrial ecosystems,and allocation of computing power resources. Keywords:virtual‒reality integration;computing system;hybrid computing architecture;cloud‒edge‒device collaboration;privacyprotection 射,实现对生产流程的预测性维护与工艺优化[10];工业物联网结合边缘计算节点,支撑起设备互联与本地智能决策[11];元宇宙作为虚实融合的前沿领域之一,则借助扩展现实(XR)与大模型技术,在虚拟社交、沉浸式培训、数字文旅等方面拓展人机交互边界[12]。尤其是在制造业、医疗、教育等领域,基于生成式大模型的虚拟专家系统、数字孪生人体与智能教学助手等应用,逐步重塑传统行业的运行逻辑与服务模式[13]。然而,虚实融合的深入发展对算力体系提出了更高要求。一方面,不同应用对算力的需求存在显著差异,数字孪生依赖高精度建模与实时数据闭环,工业物联网强调边缘端低时延响应,而元宇宙则追求多模态融合与沉浸交互。另一方面,以大模型为代表的生成式AI在推动场景智能化的同时,也因其大规模参数推理与动态负载调度特性,加剧了算力资源在规模、结构与能效方面的压力[14]。目前,尽管异构并行、存算一体与分布式推理等技术已取得初步进展,但在芯片自主性、三维图形引擎、算法优化与终端生态构建等方面仍存在明显短板,制约了虚实融合系统的大规模推广。 一、前言 随着数字技术与实体经济的深度融合,以数字孪生、工业物联网、边缘智能与元宇宙等为代表的虚实融合技术群正推动社会迈入“人机物”三元协同的新阶段[1]。这些新兴应用不仅在工业仿真、智能制造、智慧城市等领域展现出广泛潜力,也因其对实时交互、多模态数据处理与智能决策的高要求,对现有算力架构提出了前所未有的挑战[2]。在国家政策的积极推动下,我国已将虚拟现实、人工智能(AI)、工业互联网等列为数字经济重点产业,并积极推动算力网络与新型基础设施建设[3]。上海、深圳等地也陆续发布支持元宇宙与数字孪生等融合应用发展的行动计划,为相关技术的协同创新与场景落地提供了制度保障[4]。与此同时,国际竞争态势加剧,美国、日本、韩国等国家积极布局开放式创新生态,推动关键技术的研发与应用示范[5]。 目前,我国现有算力建设已形成多极协同、场景化的区域布局[6]。在技术上,普遍采用异构融合与高密度并行架构,聚焦行业特定任务。算力体系建设的发展重心正从基础的资源集约化供给,转向体系级的智能化协同与能效优化[7],紧密围绕政务云、AI大模型训练、工业仿真、医疗影像等具体行业需求,呈现出极强的应用导向与任务特化趋势[8]。此演进也表明,我国算力体系正经历从早期“资源集中供给”向未来“智能协同编排”的关键跃迁,重心已从硬件规模堆叠转向体系级的智能化、自适应与绿色可持续发展[9]。 虚实融合场景的共性特征主要体现在异构算力的高效协同,低延迟、高带宽的通信保障,多源异构数据的融合处理与隐私保护等方面,亟需构建与之匹配的绿色、弹性、可持续的新型算力体系。在此背景下,本文围绕虚实融合驱动的算力需求演变,总结算力体系演进趋势,结合算力体系发展与应用现状,系统梳理当前算力架构在面对数字孪生、工业物联网、边缘计算及元宇宙等多元场景时所面临的关键挑