Page 84 - 《爆炸与冲击》2026年第2期
P. 84
第 46 卷 第 2 期 爆 炸 与 冲 击 Vol. 46, No. 2
2026 年 2 月 EXPLOSION AND SHOCK WAVES Feb., 2026
DOI:10.11883/bzycj-2024-0471
城市建筑外爆威力场与毁伤效应
数智仿真模型及应用 *
1
1
彭江舟 ,潘刘娟 ,高光发 ,王祉乔 ,胡 杰 ,吴威涛 ,王明洋 ,何 勇 1
1
1
3
1
2
(1. 南京理工大学机械工程学院,江苏 南京 210094;
2. 南京理工大学安全科学与工程学院,江苏 南京 210094;
3. 陆军工程大学爆炸冲击与防灾减灾全国重点实验室,江苏 南京 210007)
摘要: 为准确预测建筑外爆威力场,并解决传统经验公式中未能充分考虑环境因素的复杂性而导致的精度受
限、数值仿真在处理大规模城市场景时效率低下的难题,构建了一种基于图神经网络(graph neural network, GNN)的爆
炸威力场预测模型,直接利用建筑的几何特征,对其表面的爆炸峰值超压、峰值冲量及冲击波到达时间等三维物理场
的进行预测。与数值仿真结果的对比验证表明,本文模型展现出了卓越的预测性能:对不同几何结构的单体建筑表面
超压参数的预测均方误差为 0.97%;对复杂几何建筑、建筑群落建筑表面超压参数的平均预测误差为 3.17%;当应用于
实际城市区域时,平均预测误差为 1.29%;物理场单次预测耗时不超过 0.6 s,与数值仿真相比速度提升 3~4 个数量
级。基于模型的高精度预测,不仅可以重构建筑表面任意位置的超压时程曲线,还能准确评估结构的毁伤程度。
关键词: 爆炸威力场;毁伤评估;爆炸冲击;数智仿真;图神经网络
中图分类号: O389 国标学科代码: 13035 文献标志码: A
A digital intelligence simulation model for explosion power field and
urban building damage effect and its application
1
1
1
2
1
PENG Jiangzhou , PAN Liujuan , GAO Guangfa , WANG Zhiqiao , HU Jie ,
3
1
WU Weitao , WANG Mingyang , HE Yong 1
(1. School of Mechanical Engineering, Nanjing University of Science and Technology, Nanjing 210094, Jiangsu, China;
2. School of Safety Science and Engineering, Nanjing University of Science and Technology, Nanjing 210094, Jiangsu, China;
3. State Key Laboratory of Explosion & Impact and Disaster Prevention & Mitigation,
Army Engineering University of PLA, Nanjing 210007, Jiangsu, China)
Abstract: To accurately predict the explosion power fields in buildings, solving the failure of traditional empirical formulas
often failing to account for complex environmental factor due to their inability to account for complex environmental factors,
and that of numerical simulations inefficient for large-scale urban scenarios and do not meet the needs of rapid damage
assessment. Addressing this challenge, an innovative prediction model for explosion power fields based on graph neural
networks (GNN) was constructed using an end-to-end strategy. This model enabled rapid and precise forecasting of three-
dimensional physical fields, including peak overpressure, peak impulse, and shock-wave arrival times on building surfaces.
Compared with numerical simulations, the proposed GNN model demonstrated excellent predictive performance: it achieved a
mean square error of 0.97% for predicting surface overpressure parameters of single buildings with varying geometries, and an
average prediction error of 3.17% for complex geometric buildings and building communities. When applied to real-world
* 收稿日期: 2024-12-02;修回日期: 2025-03-19
基金项目: 国家重点研发计划(2021YFC3100705)
第一作者: 彭江舟(1996- ),男,博士,博士后,pengjz@njust.edu.cn
通信作者: 何 勇(1964- ),男,博士,教授,he1964@mail.njust.edu.cn
022202-1

