《计算机应用》唯一官方网站 ›› 2026, Vol. 46 ›› Issue (4): 1309-1322.DOI: 10.11772/j.issn.1001-9081.2025040491

• 前沿与综合应用 • 上一篇    

人群疏散计算方法的演进与变迁综述

吴夏煜, 张洪()   

  1. 成都大学 计算机学院,成都 610100
  • 收稿日期:2025-05-02 修回日期:2026-01-20 接受日期:2026-01-22 发布日期:2026-01-26 出版日期:2026-04-10
  • 通讯作者: 张洪
  • 作者简介:吴夏煜(2000—),女,四川成都人,硕士研究生,主要研究方向:人群流动算法
  • 基金资助:
    室内空间布局优化与安全保障四川省高校重点实验室重点项目(SNKJ1202502)

Review of evolution and changes in crowd evacuation calculation methods

Xiayu WU, Hong ZHANG()   

  1. College of Computer Science,Chengdu University,Chengdu Sichuan 610100,China
  • Received:2025-05-02 Revised:2026-01-20 Accepted:2026-01-22 Online:2026-01-26 Published:2026-04-10
  • Contact: Hong ZHANG
  • About author:WU Xiayu, born in 2000, M. S. candidate. Her research interests include crowd flow algorithms.
  • Supported by:
    Key Project of Key Laboratory of Interior Layout Optimization and Security, Institutions of Higher Education of Sichuan Province(SNKJ1202502)

摘要:

人群疏散建模正从经典物理模拟向数据与行为双驱动智能系统演进,引发经典模型持续改良与新兴智能技术不断涌现的双重趋势。然而,现有研究未能系统性地整合这两个趋势并揭示它们内在演进逻辑的分析框架,导致研究者难以全面评估和恰当选择现有方法。为了应对此挑战,提出一种三阶段递进式分析框架。第一阶段,梳理物理环境、突发事件及主观心理等多重因素驱动模型向更高真实性演进的方式;第二阶段,系统性地评估以社会力模型(SFM)和元胞自动机(CA)为代表的经典物理模型,以及它们在融合多因素情况下的性能改进;第三阶段,聚焦前沿智能方法,并按照“态势感知?行为预测?决策优化”的闭环逻辑对人工智能(AI)技术在疏散领域的应用进行功能性解构与归纳。该框架清晰地揭示了该领域的核心技术矛盾,即在效率与真实性之间的权衡,并阐明下一代智能疏散模型不断演进的根本动力。所提框架为理解人群疏散建模的复杂技术全景提供了一个结构化、高层次的逻辑视图,同时也为该领域的研究人员在评估、选择和创新疏散模型时提供一套系统性的参照基准。

关键词: 人群疏散, 深度强化学习, 元胞自动机模型, 社会力模型, 行为建模

Abstract:

Crowd evacuation modeling is evolving from classic physical simulation to data- and behavior-driven intelligent systems, leading to dual trends: continuous improvement of classic models and emergence of new intelligent technologies. However, the existing research lacks an analytical framework that can integrate these two trends systematically and reveal their inherent evolutionary logic, making it difficult for researchers to evaluate comprehensively and select appropriately the existing methods. To address this challenge, a three-stage progressive analytical framework was proposed. In the first stage, the evolution of models driven by multiple factors, including physical environment, emergencies, and subjective psychology, towards higher realism was sorted out. In the second stage, the classic physical models, represented by Social Force Models (SFMs) and Cellular Automata (CA), and their performance improvements under the integration of multiple factors were evaluated systematically. In the third stage, the cutting-edge intelligent methods were focused on, and the application of Artificial Intelligence (AI) technology in the field of evacuation was deconstructed and induced functionally according to the closed-loop logic of "situation awareness-behavior prediction-decision optimization". This framework reveals the core technological contradiction in this field clearly, that is the trade-off between efficiency and realism, and clarifies the fundamental driving force for the continuous evolution of next-generation intelligent evacuation models. The proposed analytical framework provides a structured, high-level logical view for understanding the complex technology panorama of crowd evacuation modeling, and offers a systematic reference benchmark for researchers in the field when evaluating, selecting, and innovating evacuation models.

Key words: crowd evacuation, Deep Reinforcement Learning (DRL), Cellular Automata (CA) model, Social Force Model (SFM), behavioral modeling

中图分类号: