什么是混合仿真？ What is hybrid simulation?
一句话：单一仿真范式各有盲区——把智能体建模（ABM）、离散事件仿真（DES）与系统动力学（SD）融合进同一个模型，让复杂系统的微观行为与宏观动态被统一刻画。这一页讲清楚它的定义、它与单一范式方法的边界、它的工作流程和它的应用。 In one sentence: every single simulation paradigm has blind spots — hybrid simulation fuses agent-based modelling (ABM), discrete-event simulation (DES), and system dynamics (SD) into one model, so that the micro-level behaviors and macro-level dynamics of complex systems are captured in a unified way. This page covers the definition, the boundaries against single-paradigm methods, the typical workflow, and the applications.
定义与核心思想Definition & core idea
先给严格定义，再给直觉解释。A strict definition first, then the intuition.
混合仿真是在同一个模型中有机融合两种及以上建模范式——通常为智能体建模（ABM）、离散事件仿真（DES）与系统动力学（SD）——使各范式在其最擅长的抽象层级上工作，并通过明确的耦合接口交换信息，从而统一刻画复杂系统微观行为与宏观动态的建模与仿真方法。 Hybrid simulation is a modelling approach that organically integrates two or more paradigms — typically agent-based modelling (ABM), discrete-event simulation (DES), and system dynamics (SD) — within a single model. Each paradigm operates at the abstraction level it handles best, exchanging information through explicit coupling interfaces, so that micro-level behaviors and macro-level dynamics of a complex system are represented in a unified way.
为什么需要它Why it is needed
仿真建模长期面对一个权衡：DES 精于刻画排队、流程与随机离散事件，却难以表达累积与反馈；SD 擅长存量流量与反馈回路，却把个体异质性抹平；ABM 能呈现个体交互涌现出的宏观模式，但单独使用时宏观结构往往靠不住。现实问题——一支供应链、一家医院、一个应急体系——从来不会只按一种范式运转。混合仿真的思路是把两难拆开：把「说不清」的部分交给物理，把「算得清」的部分交给数值，用实时耦合换取整体的真实性。 Simulation modelling has long faced a trade-off: DES excels at queues, processes, and stochastic events but struggles to express accumulation and feedback; SD excels at stocks, flows, and feedback loops but smooths away individual heterogeneity; ABM reveals macro patterns emerging from individual interactions, yet on its own its macro structure is often unreliable. Real problems — a supply chain, a hospital, an emergency system — never run on a single paradigm. The hybrid answer: decompose the dilemma, let each part speak its native language, and couple them in real time.
中文名称Naming
文献中常见「混合仿真」「混合建模与仿真」「多方法仿真」（multi-method simulation）等称呼；早期脉络里有「离散-连续混合仿真」（combined discrete-continuous simulation）。本站统一使用「混合仿真」，并注明英文 Hybrid Simulation。 The literature also uses "combined discrete-continuous simulation" and "multi-method simulation". This site consistently uses "hybrid simulation".
耦合是怎么发生的How the coupling works
以一支供应链为例：上游用 ABM 刻画各主体的策略与学习，中游订单履行用 DES 模拟排队与随机到达，下游市场用 SD 捕捉价格与库存的反馈累积。三者之间传递的是明确的变量——智能体的订单进入 DES 队列，DES 的产出改变 SD 的存量，SD 的价格又反过来影响智能体决策。接口设计是混合仿真真正的技术含量所在。 Take a supply chain: upstream firm strategies and learning are modelled with ABM; midstream order fulfilment with DES (queues and stochastic arrivals); downstream market feedback with SD (price and inventory accumulation). The paradigms exchange explicit variables — agent orders enter DES queues, DES outputs update SD stocks, and SD prices feed back into agent decisions. Interface design is where the real craft of hybrid simulation lies.
Hybrid simulation integrates two or more modelling paradigms—typically agent-based modelling (ABM), discrete-event simulation (DES), and system dynamics (SD)—within a single model, letting each paradigm operate at its natural level of abstraction so that micro-level behaviour and macro-level dynamics of a complex system are captured in one coherent whole.
它不是什么，它是什么What it is not — and what it is
理解一个方法最有效的方式，是看清它与相邻概念的边界。The most effective way to understand a method is to see its boundaries against neighbouring concepts.
一个装了三种模板的大软件——混合是建模方法论，不是软件功能。 A big piece of software with three templates — hybrid is a modelling methodology, not a software feature.
一种「跨范式融合」的建模方法论：范式选择、接口设计与时间尺度对齐都有讲究。 A cross-paradigm modelling methodology: paradigm choice, interface design, and time-scale alignment all demand care.
把三种范式各建一个模型再人工对结果。 Building three separate models and comparing their outputs by hand.
同一模型内部实时耦合：范式之间在运行中持续交换信息，共同推进同一个实验。 Real-time coupling inside one model: paradigms continuously exchange information as the same experiment runs.
为了「用上最新工具」的技术堆砌——混合本身有成本。 Tool-stacking for its own sake — hybridisation has real costs.
当单一范式的假设明显失真时，用可控的复杂度换取对系统的完整刻画。 When single-paradigm assumptions clearly distort reality, trading controlled complexity for a complete representation of the system.
一个常被混淆的对照是数字孪生：孪生强调与真实系统持续双向连接的虚拟映射，是「一种系统形态」；混合仿真强调多范式在模型内部的融合，是「一种建模方法」。孪生模型的内核完全可以由混合仿真构建——两者是方法与形态的关系，而非同义词。 A frequently confused contrast is the digital twin: a twin emphasises a continuously connected virtual mirror of a real system — a system shape; hybrid simulation emphasises fusing paradigms inside a model — a modelling method. A twin's core model can well be built with hybrid simulation; the two are method and form, not synonyms.
另一个边界是多模型对比：分别建一个 DES 模型和一个 SD 模型再比较结果，是范式对比研究；只有当它们在同一个模型里实时耦合、协同推进同一个实验时，才叫混合仿真。 Another boundary is multi-model comparison: building a DES model and an SD model separately and comparing results is paradigm-comparison research. Only when they are coupled in real time inside one model, jointly advancing the same experiment, is it hybrid simulation.
| 方法Approach | 核心抽象Core abstraction | 最擅长Best at | 盲区Blind spots |
|---|---|---|---|
| 离散事件仿真 DESDiscrete-event (DES) | 实体、队列、事件调度Entities, queues, event scheduling | 流程、排队、资源竞争与随机性Processes, queuing, resource contention, randomness | 累积效应与反馈回路表达力弱Weak at accumulation and feedback loops |
| 系统动力学 SDSystem dynamics (SD) | 存量、流量、反馈回路Stocks, flows, feedback loops | 宏观动态、政策情景与长期演化Macro dynamics, policy scenarios, long-run evolution | 个体异质性与离散随机被抹平Individual heterogeneity and discrete randomness erased |
| 智能体建模 ABMAgent-based (ABM) | 异质个体、交互规则、涌现Heterogeneous agents, interaction rules, emergence | 个体策略、学习适应与涌现现象Individual strategies, adaptation, emergent phenomena | 宏观结构缺乏约束，校准成本高Unconstrained macro structure, high calibration cost |
| 解析 / 优化模型Analytical / optimization | 数学结构、目标与约束Mathematical structure, objectives, constraints | 最优解、灵敏度与可证明的性质Optimal solutions, sensitivity, provable properties | 复杂动态与随机行为难以内嵌Complex dynamics and randomness hard to embed |
| 混合仿真Hybrid simulation | 多范式 + 明确接口Multi-paradigm + explicit interfaces | 微观行为与宏观动态统一刻画Unified micro behaviour and macro dynamics | 接口设计、时间尺度对齐与 V&V 复杂度上升Interface design, time-scale alignment, heavier V&V |
典型工作流程Typical workflow
七步：从问题界定到实验分析。Seven steps: from problem framing to experimental analysis.
问题界定与抽象层级选择Problem framing & abstraction levels
先回答「问题的哪些侧面需要微观精度、哪些侧面宏观动态足矣」。抽象层级的选择决定范式组合，是整个建模中最重要的决策。 First answer: which facets of the problem need micro-level precision, and where macro dynamics suffice. Choosing abstraction levels determines the paradigm mix — the single most consequential modelling decision.
范式组合与混合架构设计Paradigm mix & hybrid architecture
为每个子系统选择范式（如个体策略→ABM、流程执行→DES、市场反馈→SD），并设计三种典型架构：分层式（不同层级各用一种范式）、并行式（同层级分区耦合）、嵌套式（一种范式内嵌于另一种）。 Assign a paradigm to each subsystem (individual strategies → ABM, process execution → DES, market feedback → SD), then choose an architecture: layered (one paradigm per level), parallel (side-by-side coupling at the same level), or nested (one paradigm embedded in another).
接口与时间尺度对齐Interfaces & time-scale alignment
明确范式之间交换哪些变量、以什么方向和频率交换；DES 的事件时间与 SD 的连续步长需要对齐策略，这是混合仿真区别于单范式建模的核心技术环节。 Specify which variables flow between paradigms, in which direction and at what frequency; DES event time and SD continuous steps need an alignment strategy — the core technical element that sets hybrid simulation apart.
数据采集与参数标定Data collection & calibration
不同范式需要不同粒度的数据：ABM 需要个体属性与行为规则依据，DES 需要到达与服务的分布，SD 需要存量初值与结构关系。数据缺口要在建模前显式承认。 Different paradigms need different data granularity: ABM needs individual attributes and behavioural rules, DES needs arrival and service distributions, SD needs stock initialisations and structural relations. Data gaps should be acknowledged before modelling begins.
模型实现Model implementation
选择支持多范式的平台或自建耦合框架；把概念模型转为可运行代码，固化接口契约，保证范式模块可独立替换。 Choose a multi-paradigm platform or build a custom coupling framework; translate the conceptual model into runnable code, freeze interface contracts, and keep paradigm modules independently replaceable.
校核与验证（V&V）Verification & validation (V&V)
校核（verification）确认「模型按设计运行」：先单独校核各范式模块，再校核接口；验证（validation）确认「模型与现实足够一致」：对照历史数据、极限测试与极端情景。混合模型的 V&V 必须覆盖范式交互部分——这是它最容易被糊弄过去的地方。 Verification confirms the model runs as designed: each paradigm module first, then the interfaces. Validation confirms the model matches reality well enough: against historical data, stress tests, and extreme scenarios. A hybrid model's V&V must cover the cross-paradigm interactions — the part most easily glossed over.
实验设计与结果分析Experiment design & analysis
设计情景与因子（含跨范式耦合变量的因子），运行重复实验，分析输出分布而非单次结果；对结论给出适用边界，而不是越过数据说话。 Design scenarios and factors (including cross-paradigm coupling variables), run replicated experiments, and analyse output distributions rather than single runs; state the boundaries of validity instead of overreaching the data.
应用领域Applications
混合仿真的适用场景有一个共同特征：系统同时存在「个体异质行为」与「宏观反馈动态」。A common signature: the system exhibits both heterogeneous individual behavior and macro-level feedback dynamics.
供应链网络与物流系统Supply networks & logistics systems
企业主体的策略互动用 ABM，仓储运输的排队与订单履行用 DES，市场需求与价格反馈用 SD——一支供应链天然横跨三个抽象层级，是混合仿真最典型的用武之地。库存波动、牛鞭效应、网络韧性等问题都能在统一模型中研究。 Firm-level strategic interaction with ABM, warehousing and transport queues with DES, market demand and price feedback with SD — a supply chain naturally spans three abstraction levels, making it the signature application. Inventory fluctuation, bullwhip effects, and network resilience can all be studied in one unified model.
医院与公共卫生系统Hospitals & public health
患者流用 DES，慢病人群与资源存量用 SD，医患行为用 ABM。运筹学界在此积累了大量混合建模实践。 Patient flows with DES, chronic populations and capacity stocks with SD, clinician and patient behavior with ABM. OR literature offers extensive hybrid modelling practice here.
灾害响应与疏散Disaster response
个体避险行为用 ABM，救援资源调度用 DES，灾情演化用 SD——多层级耦合是应急问题的本性。 Evacuation behavior with ABM, relief resource scheduling with DES, hazard evolution with SD — multi-level coupling is inherent to emergencies.
生产系统与能源市场Production & energy markets
产线执行用 DES，产能投资与价格用 SD，产业链主体博弈用 ABM。 Production lines with DES, capacity investment and prices with SD, supply-chain firm games with ABM.
大语言模型 × 混合仿真LLMs × hybrid simulation
LLM 正被用于辅助范式选择、代码生成、参数校准与模型验证，把建模者从重复劳动中解放出来；更前沿的方向是让 LLM 直接扮演模型中的智能体决策者。混合仿真的「理论优化」阶段正在到来——这也是本站作者当前的研究切入点之一。 LLMs increasingly assist paradigm selection, code generation, calibration, and verification — freeing modellers from repetitive work; the frontier is letting LLMs act as decision-making agents inside the model. Hybrid simulation is entering its "theoretical grounding" phase — one of the author's current research entry points.
延伸阅读Further reading
本页内容的来源与进阶路径，均为真实可查文献。Sources behind this page and paths for going deeper — all real, verifiable references.
- Jia, F. (2026). 混合仿真方法与应用的系统综述与展望（1980–2024）. Archives of Computational Methods in Engineering.——对混合仿真 1980–2024 演进历程的系统综述：三大建模范式、混合架构与核心决策框架的全景梳理。（本站作者所著；DOI 与卷期待刊发信息确认后补入） Jia, F. (2026). A Systematic Review and Prospect of Hybrid Simulation Methods and Applications from 1980 to 2024. Archives of Computational Methods in Engineering. — A systematic review of the field's evolution: three modelling paradigms, hybrid architectures, and core decision frameworks. (Authored by this site's author; DOI and volume to be added upon publication.)
- Brailsford, S. C., Eldabi, T., Kunc, M., Mustafee, N., & Osorio, A. F. (2019). Hybrid simulation modelling in operational research: A state-of-the-art review. European Journal of Operational Research, 278(3), 748–765. doi:10.1016/j.ejor.2019.03.025——运筹学领域混合仿真的权威综述，被引 600+。 Brailsford, S. C., Eldabi, T., Kunc, M., Mustafee, N., & Osorio, A. F. (2019). Hybrid simulation modelling in operational research: A state-of-the-art review. European Journal of Operational Research, 278(3), 748–765. doi:10.1016/j.ejor.2019.03.025 — The authoritative OR review of hybrid simulation.
- Borshchev, A., & Filippov, A. (2004). From System Dynamics and Discrete Event to Practical Agent Based Modeling: Reasons, Techniques, Tools. The 22nd International Conference of the System Dynamics Society.——多方法建模的开山之作，直接回答「为什么要混合、怎么混合」。 Borshchev, A., & Filippov, A. (2004). From System Dynamics and Discrete Event to Practical Agent Based Modeling: Reasons, Techniques, Tools. The 22nd International Conference of the System Dynamics Society. — The founding paper of multi-method modelling: why hybridise, and how.
注：本页定义与流程表述由本站作者整理，术语与上述文献经逐条核对；发现表述问题欢迎指正。第 1 条的刊发信息以期刊官网为准，本站作者主页亦有收录。 Note: definitions and workflow descriptions on this page were prepared by the site author and cross-checked against the references above. Corrections are welcome. Publication details for the first item follow the journal's official records.