基于LLM的洪涝灾害下居民需求仿真模拟与时序特征

刘梅, 袁景怡, 陈泽雨, 尉方, 方东平

清华大学学报(自然科学版) ›› 2026, Vol. 66 ›› Issue (11) : 2395-2406.

PDF(2249 KB)
PDF(2249 KB)
清华大学学报(自然科学版) ›› 2026, Vol. 66 ›› Issue (11) : 2395-2406. DOI: 10.16511/j.cnki.qhdxxb.2026.27.061
建设管理

基于LLM的洪涝灾害下居民需求仿真模拟与时序特征

  • 刘梅1, 袁景怡1, 陈泽雨2, 尉方3, 方东平4
作者信息 +

Simulation of resident needs and their temporal characteristics in flood disasters based on large language models

  • LIU Mei1, YUAN Jingyi1, CHEN Zeyu2, WEI Fang3, FANG Dongping4
Author information +
文章历史 +

摘要

近年来,频发的洪涝灾害使中国城市经济社会的可持续发展面临严峻考验。灾害应对中,公众感知与需求直接影响资源配置效能,因此该文旨在揭示灾害演进过程中异质性居民需求的时序性变化特征。首先,依托Maslow需求层次理论构建居民需求指标体系,然后在洪涝情景模拟基础上运用大语言模型开展异质性主体建模,最后模拟不同居民交互的认知行为决策过程和需求的时序性变化。结果表明:受人格特质(基于大五人格模型(big five personality traits))、生理功能状态(健康、生理功能障碍)与环境风险共同影响,异质性主体的需求层级呈差异化递进。尽责性(conscientiousness)与神经质(neuroticism)个体倾向于保守避险与反复安全确认,其生存与安全需求在降雨减弱后仍长期保持高权重,层级递进滞后;开放性(openness)与外向性(extraversion)个体环境适应力较强,较早产生社交协作与精神慰藉需求;生理功能障碍主体则表现出响应滞后性与高度外部依赖特征。该研究丰富了洪涝灾害中异质性居民需求的动态识别方法,为人本化应急管理提供科学依据。

Abstract

[Objective] Urban flooding is characterized by a dynamically evolving risk environment, accompanied by temporal changes in resident perceptions, behavioral decisions, and needs. Because individuals differ from one another, their needs may vary across disaster stages and follow distinct temporal trajectories. Characterizing these changes can inform need identification and emergency resource allocation during disaster response. Against this background, this study aimed to reveal the temporal evolution of heterogeneous resident needs throughout an urban flood disaster. It examined changes and transitions among survival, safety, social, and psychological needs across resident groups, with particular attention to the roles of personality traits, physiological conditions, and environmental risk. The study aimed to provide a quantitative basis for differentiated need identification and emergency resource allocation. [Methods] Drawing on Maslow’s hierarchy of needs theory, a four-level indicator system of resident needs was established through literature review, policy analysis, and semi-structured interviews. A dynamic flood scenario was then constructed for the plain area of Fangshan District, Beijing, in accordance with local Beijing standards; the scenario adopted a 150-year return period and covered a 14-day flood process. The Horton infiltration model generated a continuous time series of inundation depth and the corresponding environmental risk levels. A large language model (LLM)-driven multi-agent model was subsequently developed to represent heterogeneous residents characterized by Big Five personality traits and differing physiological conditions. GPT-4o simulated the cognitive process of perception, recognition, judgment, and decision-making, and the model incorporated agent interactions and historical records into subsequent need decisions. The simulation included 150 resident agents, used a 1-h time step over 336 h, and was repeated independently 30 times. Jensen-Shannon divergence and mean absolute error were used to evaluate variability across the repeated simulations.[Results] Across the 30 simulation runs, the mean Jensen-Shannon divergence was 0.004 3 and the overall mean absolute error was 0.027 3. Need-transition patterns differed among personality groups. For conscientious residents, the self-transition probabilities of survival and safety needs were 60.4% and 56.0%, respectively. For agreeable and extraverted residents, the self-transition probabilities of social needs were 79.9% and 83.7%, respectively. For open residents, the self-transition probability of psychological needs was 78.3%, and the transition probabilities from safety needs and from social needs to psychological needs were 38.1% and 42.1%, respectively. For neurotic residents, the transition probabilities from social and psychological needs back to survival needs were 32.6% and 33.7%, respectively. Physiological conditions were likewise associated with differences in need trajectories. Among residents with functional impairments, safety needs accounted for 40% during the early stage and 23% during the recovery stage, approximately 1.2 and 1.4 times the corresponding proportions among residents without functional impairments. Environmental risk was associated with changes in need composition. As risk increased from low to medium-high, safety response intensity rose from 17% to 60%, whereas social and psychological response intensities fell from 40% to 14% and from 23% to 4%, respectively. As inundation depth decreased, social and psychological needs increased.[Conclusions] Resident needs during urban flooding follow heterogeneous temporal trajectories across disaster stages. In the early stage, when environmental risk is elevated, needs concentrate on survival and safety. As inundation risk recedes, the trajectories diverge across groups that differ in personality traits and physiological conditions. These findings indicate that need assessment should account for both disaster progression and resident heterogeneity. The LLM-driven multi-agent framework provides a quantitative approach for characterizing temporal changes in resident needs and can inform differentiated need identification and stage-specific emergency resource allocation.

关键词

洪涝灾害 / 居民需求 / 大语言模型 / Maslow需求层次理论 / 时序特征

Key words

flood disasters / resident needs / large language models / Maslow hierarchy of needs / temporal characteristics

引用本文

导出引用
刘梅, 袁景怡, 陈泽雨, 尉方, 方东平. 基于LLM的洪涝灾害下居民需求仿真模拟与时序特征[J]. 清华大学学报(自然科学版). 2026, 66(11): 2395-2406 https://doi.org/10.16511/j.cnki.qhdxxb.2026.27.061
LIU Mei, YUAN Jingyi, CHEN Zeyu, WEI Fang, FANG Dongping. Simulation of resident needs and their temporal characteristics in flood disasters based on large language models[J]. Journal of Tsinghua University(Science and Technology). 2026, 66(11): 2395-2406 https://doi.org/10.16511/j.cnki.qhdxxb.2026.27.061
中图分类号: TU998.4   

参考文献

[1] 吕伟, 江惠华, 汪京辉, 等. 洪涝灾害下城市大规模人员疏散分区规划方法[J]. 清华大学学报(自然科学版), 2024, 64(11): 1880-1892. LV W, JIANG H H, WANG J H, et al. Urban large-scale evacuation zoning planning methods for flooding disasters[J]. Journal of Tsinghua University (Science and Technology), 2024, 64(11): 1880-1892. (in Chinese)
[2] 田水承, 聂玉蓉, 陈磊, 等. 基于SEM的矿工大五人格特质和情绪对不安全行为的影响[J]. 中国安全科学学报, 2025, 35(8): 22-32. TIAN S C, NIE Y R, CHEN L, et al. A SEM-based study of influence of miners’ Big Five personality traits and emotions on unsafe behaviors[J]. China Safety Science Journal, 2025, 35(8): 22-32. (in Chinese)
[3] 景国勋, 程梦洁, 蒋方, 等. 大五人格特质对矿工风险感知的影响研究[J]. 安全与环境学报, 2026, 26(5): 1887-1895. JING G X, CHENG M J, JIANG F, et al. Examining the impact of big five personality traits on miners’ risk perception[J]. Journal of Safety and Environment, 2026, 26(5): 1887-1895. (in Chinese)
[4] 王喆, 王志, 况星园, 等. 考虑大五人格特质的应急救援人员派遣模型[J]. 安全与环境学报, 2024, 24(10): 3982-3991. WANG Z, WANG Z, KUANG X Y, et al. An emergency rescue personnel dispatch model integrating the Big Five personality traits characteristics[J]. Journal of Safety and Environment, 2024, 24(10): 3982-3991. (in Chinese)
[5] 宋晓猛, 徐楠涛, 张建云, 等. 中国城市洪涝问题: 现状、成因与挑战[J]. 水科学进展, 2024, 35(3): 357-373. SONG X M, XU N T, ZHANG J Y, et al. Urban flooding in China: Current status, causes and challenges[J]. Advances in Water Science, 2024, 35(3): 357-373. (in Chinese)
[6] ZHOU S T, ZHAI G F. A multi-hazard risk assessment framework for urban disaster prevention planning: A case study of Xiamen, China [J]. Land, 2023, 12(10): 1884.
[7] LI Y T, WANG Y F, GONG J. An integrated metric for rapid and equitable emergency rescue during urban flash flooding events [J]. International Journal of Disaster Risk Reduction, 2025, 118: 105209.
[8] 王婷婷, 陈伟炯, 张善杰, 等. 城市雨灾洪涝背景下应急物资动态需求预测[J]. 上海海事大学学报, 2023, 44(1): 73-80. WANG T T, CHEN W J, ZHANG S J, et al. Dynamic demand prediction of emergency materials under background of urban heavy rain and flood[J]. Journal of Shanghai Maritime University, 2023, 44(1): 73-80. (in Chinese)
[9] ZHAO Z Y, ZHOU X S, LIN Z Z, et al. A resident-centric framework for postdisaster infrastructure recovery: Characterizing hierarchical needs and fulfillment cycles to assess urban resilience [J]. Journal of Management in Engineering, 2025, 41(3): 04025003.
[10] 苏伯尼, 黄弘, 张楠. 基于情景模拟的城市内涝动态风险评估方法[J]. 清华大学学报(自然科学版), 2015, 55(6): 684-690. SU B N, HUANG H, ZHANG N. Dynamic urban waterlogging risk assessment method based on scenario simulations[J]. Journal of Tsinghua University (Science and Technology), 2015, 55(6): 684-690. (in Chinese)
[11] 杨宇涵, 殷杰, 王丹丹, 等. 基于ABM的城市暴雨洪涝灾害应急疏散仿真研究——以河南郑州“7·20”特大暴雨洪涝灾害为例[J]. 中国科学: 地球科学, 2023, 53(2): 267-276. YANG Y H, YIN J, WANG D D, et al. ABM-based emergency evacuation modelling during urban pluvial floods: A “7·20” pluvial flood event study in Zhengzhou, Henan Province[J]. Science China Earth Sciences, 2023, 66(2): 282-291. (in Chinese)
[12] YANG Y H, YIN J, FENG W H, et al. Dynamic flood evacuation modelling for coastal cities: A case study of Shanghai[J]. International Journal of Disaster Risk Reduction, 2025, 125: 105591.
[13] 李常敏, 马啸吟, 崔娜. 洪涝灾害下考虑灾民恐慌心理的应急物资调度研究[J]. 安全与环境学报, 2025, 25(9): 3615-3625. LI C M, MA X Y, CUI N. Investigating emergency supply dispatch strategies considering panic psychology during flood disasters [J]. Journal of Safety and Environment, 2025, 25(9): 3615-3625. (in Chinese)
[14] 薛书琦, 龚壮壮. 突发洪涝灾害中考虑心理焦虑影响的多周期应急物资调度研究[J]. 安全与环境学报, 2023, 23(5): 1522- 1529. XUE S Q, GONG Z Z. Multi-period emergency supplies dispatching considering the effect of psychological anxiety in sudden flood disaster[J]. Journal of Safety and Environment, 2023, 23(5): 1522-1529. (in Chinese)
[15] MORADZADEH M, AHMADI M. Unraveling the interplay of human decisions and flood risk: An agent-based modeling approach[J]. International Journal of Disaster Risk Reduction, 2024, 107: 104486.
[16] 柯罗马, 李增逸, 廖江群, 等. 大语言模型模拟区域心理结构的有效性: 人格与幸福感的实证检验[J]. 心理科学, 2025, 48(4): 907-919. KE L M, LI Z Y, LIAO J Q, et al. Effectiveness of large language models in simulating regional psychological structures: An empirical examination of personality and subjective well-being [J]. Journal of Psychological Science, 2025, 48(4): 907-919. (in Chinese)
[17] 曾秀芹, 陈珂璐. 谁更像“人”? 模型类型与人格设定对大语言模型复刻传播学实验准确率的影响[J]. 新闻与传播评论, 2026, 79(1): 25-39. ZENG X Q, CHEN K L. Which model best simulates human behavior? The impact of model type and personality setting on the accuracy of large language models in replicating communication studies experiments[J]. Journalism & Communication Review, 2026, 79(1): 25-39. (in Chinese)
[18] 牛玥铧, 聂百胜. 人格特征对驾驶人驾驶行为的影响研究[J]. 中国安全科学学报, 2024, 34(2): 117-123. NIU Y H, NIE B S. Study on influence of personality traits on drivers’ driving behavior[J]. China Safety Science Journal, 2024, 34(2): 117-123. (in Chinese)
[19] 苏鑫, 邵薇薇, 刘家宏, 等. 基于情景模拟的洪涝灾害经济损失动态评估[J]. 清华大学学报(自然科学版), 2022, 62(10): 1606-1617. SU X, SHAO W W, LIU J H, et al. Dynamic assessment of economic losses from flood disasters based on scenario simulations [J]. Journal of Tsinghua University (Science and Technology), 2022, 62(10): 1606-1617. (in Chinese)
[20] 王沐林, 吕伟, 杨晓婷, 等. 基于机器学习和数据增强的洪灾人口转移需求预测方法[J]. 清华大学学报(自然科学版), 2026, 66(1): 160-168. WANG M L, LV W, YANG X T, et al. Prediction method based on machine learning and data augmentation for population relocation demand during floods[J]. Journal of Tsinghua University (Science and Technology), 2026, 66(1): 160-168. (in Chinese)
[21] 邵蕊, 邵薇薇, 苏鑫, 等. 基于TELEMAC-2D模型分析不同洪涝情景对城市应急响应时间的影响[J]. 清华大学学报(自然科学版), 2022, 62(1): 60-69. SHAO R, SHAO W W, SU X, et al. Impact of various flood scenarios on urban emergency responses times based on the TELEMAC-2D model[J]. Journal of Tsinghua University (Science and Technology), 2022, 62(1): 60-69. (in Chinese)
[22] 段琴, 罗思阳. 基于智能体建模研究人类情感与认知的演化(英文) [J]. 心理科学, 2025, 48(6): 1314-1332. DUAN Q, LUO S Y. Using agent-based modeling to study the adaptation and evolution of human emotions and cognition[J]. Journal of Psychological Science, 2025, 48(6): 1314-1332.
[23] 朱洋甫, 李美玲, 谭嘉辰, 等. 文本人格检测研究综述[J]. 计算机科学, 2024, 51(12): 209-222. ZHU Y F, LI M L, TAN J C, et al. Study on text-based personality detection-a review[J]. Computer Science, 2024, 51(12): 209-222. (in Chinese)
[24] 颉浩田, 李宁, 王虹虹, 等. 五态人格的潜在剖面分析及其与大五人格的相关性研究[J]. 中华中医药学刊, 2025, 43(12): 226-230. XIE H T, LI N, WANG H H, et al. Relationship between five pattern personalities and big five personality based on latent profile analysis[J]. Chinese Archives of Traditional Chinese Medicine, 2025, 43(12): 226-230. (in Chinese)
[25] FU X R, WANG D, LUAN Q H, et al. Community scale assessment of the effectiveness of designed discharge routes from building roofs for stormwater reduction[J]. Remote Sensing, 2022, 14(13): 2970.

基金

国家自然科学基金青年科学基金项目(72301018);北京市社会科学基金年度规划项目青年项目(23GLC046);山西省基础研究计划资助项目(202404030401044, 202303021222005)

PDF(2249 KB)

Accesses

Citation

Detail

段落导航
相关文章

/

〈 〉