Page 42 - 《爆炸与冲击》2026年第11期
P. 42

第 46 卷    第 11 期                  爆    炸    与    冲    击                      Vol. 46, No. 11
                2026 年 11 月                   EXPLOSION AND SHOCK WAVES                          Nov., 2026

               DOI:10.11883/bzycj-2025-0383


                      基于贝叶斯深度主动学习的三维城市建筑群

                                           爆炸载荷快速预测                         *


                                                                              2
                                                               1
                                                        1
                                                                       1
                                                1
                                   潘美霖 ,张    情 ,邱玖禄 ,田    宙 ,钟    巍 ,冷春江 ,彭卫文         2
                                         1
                           (1. 西北核技术研究所强脉冲辐射环境模拟与效应全国重点实验室,陕西 西安 710024;
                                           2. 中山大学智能工程学院,广东 广州 510275)
                  摘要: 城市爆炸载荷快速预测对于防灾设计、应急救援及灾后重建具有重要意义。针对现有基于深度学习的预
               测模型依赖大量高质量样本、建模成本高且样本利用率低的问题,提出一种基于贝叶斯深度主动学习(Bayesian deep
               active learning, BDAL)的三维城市建筑群爆炸载荷快速预测方法。研究在三维空间构建规则化城市建筑群,设定包含
               能量源当量、起爆距离、建筑尺寸及街道特征的七维参数空间,采用全因子实验设计系统生成参数组合,并利用
               blastFoam  软件开展三维数值模拟,获取关键位置超压峰值数据。基于贝叶斯推断实现模型参数的概率化建模,结合主
               动采样策略量化预测不确定性并优化样本选择,从而提升样本利用效率。测试结果显示,在                                 780  组未经训练的样本
               上,该方法的平均绝对百分比误差为             13.1%,预测区间覆盖真实值的概率为          85.9%,单点预测响应时间小于         20 ms,仅需
               约  50%  的标注数据即可达到与全样本训练的被动式深度学习模型相近的精度。结果表明,该方法可在典型规则化三
               维城市环境中实现高效、低成本的爆炸载荷预测,具有在防灾减灾领域应用的潜力。
                  关键词: 三维典型城市建筑群;爆炸载荷;贝叶斯深度主动学习;快速预测
                  中图分类号: O389; TP183   国标学科代码: 1303520   文献标志码: A

                   Fast prediction of blast loading for three-dimensional urban building
                                clusters based on Bayesian deep active learning


                                                                         1
                           1
                                                                                         2
                                        1
                                                  1
                                                             1
                 PAN Meilin , ZHANG Qing , QIU Jiulu , TIAN Zhou , ZHONG Wei , LENG Chunjiang , PENG Weiwen 2
                        (1. National Key Laboratory of Intense Pulsed Radiation Simulation and Effect, Northwest Institute of
                                          Nuclear Technology, Xi’an 710024, Shaanxi, China;
                     2. School of Intelligent Systems Engineering, Sun Yat-sen University, Guangzhou 510275, Guangdong, China)
               Abstract:   To  address  the  high  cost  and  low  sample  efficiency  of  deep  learning-based  blast  loading  prediction  in  urban
               environments,  a  Bayesian  deep  active  learning  (BDAL)  method  is  proposed.  The  objective  is  to  significantly  reduce  the
               dependency  on  large-scale,  high-fidelity  numerical  simulation  data  while  maintaining  prediction  accuracy  and  providing
               reliable  uncertainty  quantification.  A  three-dimensional  typical  urban  building  cluster  consisting  of  a  3×3  regular  array  of
               cuboid buildings was constructed. A seven-dimensional parameter space was defined, including explosive charge equivalence
               (1 000, 2 000, 3 000 kt), detonation distance (1 000, 2 000, 3 000 m), building length (10, 20, 30 m), building width (20, 40 m),
               building height (75, 100 m), street length (50, 75, 100 m), and street width (50, 75 m). A full factorial experimental design was
               employed,  generating  648  parameter  combinations.  For  each  combination,  the  open-source  computational  fluid  dynamics
               (CFD)  software  blastFoam  was  used  to  perform  three-dimensional  numerical  simulations  of  blast  wave  propagation.  The
               background mesh size was set to 30 m based on grid sensitivity analysis, and adaptive mesh refinement (AMR) with local
               refinement level 2 and dynamic refinement level 1 was applied to capture shock wave details. Peak overpressure values were



                 *   收稿日期: 2025-11-25;修回日期: 2026-04-14
                   第一作者: 潘美霖(1999- ),女,硕士,panml25@mails.tsinghua.edu.cn
                   通信作者: 钟 巍(1986- ),男,博士,研究员,zhongwei@nint.ac.cn


                                                         111503-1
   37   38   39   40   41   42   43   44   45   46   47