考虑洪涝灾害风险的小样本应急避难所选址模型  

Few-shot model for identifying emergency shelter locations considering flood disaster risk

作　　者：李娜 吴怀石 LI Na;WU Huaishi(School of Economics and Management,Tianjin Chengjian University,Tianjin 300384,China)

机构地区：[1]天津城建大学经济与管理学院,天津300384

出　　处：《安全与环境学报》2025年第3期1090-1101,共12页Journal of Safety and Environment

基　　金：天津市哲学社会科学规划项目(TJGL23-004)。

摘　　要：探索应急避难所的选址合理性,对于城市防灾减灾工作和城市居民生命财产的保护具有重要意义。从洪涝灾害的视角出发,构建以致灾因子、孕灾环境、承灾体和防灾减灾能力为选址要素的洪涝灾害指标体系,并采用信息量模型,在选址要素的基础上通过地图代数计算洪涝灾害风险指数;然后,通过元学习算法,设计一种适用于小样本应急避难所选址的组合元模型。结果显示:对比天津市蓟州区现有避难所分布与优化方案,候选点的平均洪涝灾害风险指数分别下降61.52%,39.13%和24.95%;以洇溜镇为例,发现元模型候选点具有较为集中的避难资源且更加符合避难需求。此外,为验证模型的整体有效性,通过对比其他数据集,展现了元模型在小样本应急避难所选址场景中的性能优势。To enhance the spatial arrangement of emergency shelters in response to flood disasters,this study focused on 79 emergency shelters located in Jizhou District,Tianjin.Initially,we developed an indicator system for flood disasters,incorporating site selection factors such as disaster-causing elements,high-risk environments,vulnerable populations,and disaster prevention and mitigation capabilities.An information value model was employed to assess the information values of various indicators across different levels.Subsequently,we utilized ArcGIS 10.2 to apply reclassification and weighted overlay tools for integrating the various layers.This process generated a flood disaster risk index and a corresponding risk distribution map,which facilitated the identification of shelters requiring site optimization.Furthermore,we developed a combined meta-model tailored for few-shot emergency shelter site selection,leveraging meta-learning algorithms to enhance decision-making efficiency.This model utilizes inner and outer loops connected by an intermediate representation model,enabling rapid iterative optimization.This approach effectively adapts to the complexities of varying shelter distributions,resulting in an optimized meta-testing solution.The results indicate that the distribution of emergency shelters in Jizhou District is suboptimal,with 44.30%of shelters situated in high-risk areas.Furthermore,the meta-model outperforms the basic model in training performance,demonstrating its strong generalization capabilities.In comparison to the current distribution of shelters in Jizhou District,the proposed locations from all models are more widely dispersed and better aligned with shelter needs.This results in average flood risk indices decreasing by 61.52%,39.13%,and 24.95%,respectively.Among the key contributing factors,proximity to rivers is vital for the planning of emergency shelters.The proposed locations in Yinliu Town offer abundant shelter resources,showing enhancements in distance from rivers,as well as improvements

关 键 词：公共安全 应急避难所 选址优化 洪涝灾害风险指数 元模型 小样本 

分 类 号：X915.5[环境科学与工程—安全科学]