中国安全科学学报 ›› 2021, Vol. 31 ›› Issue (12): 153-159.doi: 10.16265/j.cnki.issn 1003-3033.2021.12.020

• 应急技术与管理 • 上一篇    下一篇

城市应急避难场所管理体系构建与应用*

王海飙1 副教授, 陈海超1, 李蕊2, 于贺1, 赵天怡2   

  1. 1 东北林业大学 土木工程学院,黑龙江 哈尔滨 150040;
    2 东北林业大学 园林学院,黑龙江 哈尔滨 150040
  • 收稿日期:2021-09-05 修回日期:2021-11-10 出版日期:2021-12-28 发布日期:2022-06-28
  • 作者简介:王海飙 (1973—),男,辽宁抚顺人,博士,副教授,主要从事城市安全与工程防震减灾方面的研究。E-mail:whbcumt@163.com。
  • 基金资助:
    中国地震局建筑物破坏机理与防御重点实验室开放基金资助(FZ201105);国家级大学生创新训练项目(202010225086)。

Construction and application of management system of urban emergency shelters

WANG Haibiao1, CHEN Haichao1, LI Rui2, YU He1, ZHAO Tianyi2   

  1. 1 School of Civil Engineering, Northeast Forestry University, Harbin Heilongjiang 150040, China;
    2 School of Landscape Architecture, Northeast Forestry University, Harbin Heilongjiang150040, China
  • Received:2021-09-05 Revised:2021-11-10 Online:2021-12-28 Published:2022-06-28

摘要: 为构建更科学的城市应急避难场所管理体系,根据有效面积等7项指标,使用ArcGIS量化采集成都市36处应急避难场所的数据,并采用Ward法进行系统聚类,结果显示分3类较适宜;分析这3类避难场所的特征与区别,并使用主成分分析法进行指标降维,绘制3类避难场所的特征散点图。研究结果表明:3类应急避难场所分别在城市防震减灾中起枢纽、核心与补充作用,其共同构成橄榄型结构的城市应急避难体系。为应用该体系,通过BP神经网络算法训练得到非线性分类模型,训练集损失函数迭代约800次后收敛,训练集和测试集的分类准确率均达到100%。

关键词: 应急避难场所, Ward法, BP神经网络, 分类, 主成分

Abstract: In order to develop a more scientific management system of urban emergency shelters, ArcGIS was used to collect quantitative data of 36 shelters in Chengdu according to 7 indicators such as effective area. Then, ward method was applied to cluster them systematically, and 3 types of them were showed to be suitable according to results. Finally, characteristics and differences of these 3 types were analyzed, and dimensions of indexes were reduced by using principal component method so as to draw a characteristic scatter plot of these shelters. The results show that these 3 types of emergency shelters play a role of hub, core and supplement in urban earthquake disaster reduction, and they constitute an olive-shaped urban emergency shelter system.In order to apply this system, nonlinear classification model is obtained by BP neural network algorithm. The training set loss function converges after about 800 iterations, and classification accuracy of both training set and test set can reach as high as 100%.

Key words: emergency shelters, Ward method, BP neural network, classification, principal component

中图分类号: