基于爆炸现场痕迹反演爆源参数方法及应用
发布时间：2018-08-31 17:43
【摘要】：爆源参数反演在爆炸现场痕迹分析中是重点关注的问题。为了从爆炸现场痕迹数据中获得爆源参数信息（爆源药量、爆源埋深/悬空高度），本文以爆炸产生的痕迹数据为基础，采用数据驱动的前沿技术（广义神经网络GRNN、粒子群PSO、支持向量机SVM、以及非线性非稳态信号分析算法HHT/EMD）建立了基于爆炸现场痕迹反演爆源参数的方法。最终将这些方法应用于爆源参数反演系统软件中，并得到初步应用。本文主要研究内容和主要成果如下： 一、基于非线性回归的数学理论，将广义神经网络引入到对土介质炸坑痕迹反演爆源参数的模型构建中：以炸坑直径与深度、土壤类型作为GRNN的输入层，爆源药量与埋深作为GRNN的输出层，进行模型训练与构建，并进行了实验验证。与传统经验公式反演出的结果进行了比较，粘土、沙土、沙粘混合土壤介质中的炸坑痕迹反演爆源药量和爆源埋深的精度有明显提高：基于GRNN的炸坑痕迹反演得到的药量及埋深相对误差在30%以内；建立的算法对药量及埋深的反演误差平均值分别为：15.41%，，16.93%，与传统经验公式相比精度均有明显提升。 二、建立了分维数值与爆炸地震波幅值衰减指数的关系，并针对近年来对爆炸地震波振动信号进行能量和频谱特征分析常用的数据处理方法EMD/HHT进行了改进。通过PSO-SVM组合模型的应用与优化，有效地解决端点效应，提高信号分解的准确度和可信度；并利用改进的EMD/HHT方法找到了爆炸地震波信号的局部时间尺度本征信息，反演了微差爆破中的微差延期时间。同时，基于GRNN建立了根据爆炸震动参数反演爆源参数的方法，确定了爆源药量与震动峰值速度、爆心距的隐式非线性关系，并将反演结果与实测结果进行对比分析，反演药量的误差平均值由萨道夫斯基经验公式的51.57%提高到了13.32%。 三、构建了基于爆炸现场的玻璃痕迹反演爆源参数的方法。根据建筑常用的平板玻璃的几何特征量（玻璃长、宽、厚度）和爆心距，利用神经网络建立爆源药量反演算法，并进行了实验验证，结果表明爆源药量反演误差平均值从萨道夫斯基经验公式的51.15%提高到了19.3%。同时，对反演中玻璃破坏的临界压力数据库表，利用PSO-SVM的数据延拓方法对小厚度尺寸玻璃所能承受临界超压值进行了补充，为后期玻璃破坏临界超压数据库的使用提供了数据保障。 四、基于单类痕迹反演，开展了基于爆炸现场多类痕迹系统地反演爆源参数的探索研究，初步建立了综合爆炸现场三类主要痕迹因素（炸坑痕迹、玻璃痕迹、爆炸震动记录）反演爆源参数的方法，包括：炸坑—玻璃综合反演爆源参数、炸坑—爆炸震动综合反演爆源参数和炸坑—玻璃—爆炸震动综合反演爆源参数的方法。分别将三因素反演效果与各自单因素反演结果进行对比，结果表明，综合三因素反演精度高于单因素反演。 五、本文将上述三种单因素反演方法以及三种多因素反演方法应用到软件“爆源参数反演系统”的核心模块“爆源特征反演分析”中。该软件是当前和未来爆炸案件侦破的重要手段，为调查人员及相关专家等提供了爆炸现场分析支持工具。 本文研究工作的主要学术贡献在于将数据驱动理念与技术引入到由爆炸现场的各类痕迹数据获取爆源参数信息的过程中，提出了多种爆源参数反演方法，并经检验与验证比传统方法具有更高的精度，对于爆炸现场分析具有实用意义，将为爆炸案件的分析与侦破工作提供技术支持。
[Abstract]:Inversion of explosive source parameters is an important problem in the analysis of explosive traces. In order to obtain the information of explosive source parameters (explosive charge, explosive source buried depth / suspended height) from the explosive traces data, this paper uses the data-driven frontier technology (generalized neural network GRNN, particle swarm optimization, support direction) based on the traces data generated by explosion. SVM and non-linear unsteady-state signal analysis algorithm HHT/EMD are used to retrieve the parameters of explosive source based on the traces of explosion site.
Firstly, based on the mathematical theory of nonlinear regression, the generalized neural network is introduced into the model construction of inversion of blasting source parameters for the blasting pit traces in soil medium: the diameter and depth of the blasting pit, the soil type as the input layer of GRNN, the explosive charge and the buried depth as the output layer of GRNN, the model training and construction are carried out, and the experimental verification is carried out. Comparing the inversion results of the empirical formulas, the precision of inversion of explosive charge and depth of explosive source in clay, sandy soil and sand-clay mixed soil has been improved obviously: the relative error of explosive charge and depth obtained by inversion of explosive crater trace based on GRNN is less than 30%; the inversion error of explosive quantity and depth is equal to that of the established algorithm. The mean value is 15.41%, 16.93%, respectively, and the accuracy is obviously improved compared with the traditional empirical formula.
Secondly, the relationship between the fractal dimension and the amplitude attenuation index of explosive seismic wave is established, and the EMD/HHT data processing method, which is commonly used to analyze the energy and spectrum characteristics of explosive seismic wave vibration signals in recent years, is improved. The end effect is effectively solved and the accuracy of signal decomposition is improved by the application and optimization of PSO-SVM combined model. The local time scale intrinsic information of explosive seismic wave signal is found by using the improved EMD/HHT method, and the millisecond delay time in millisecond blasting is inverted. The inversion results are compared with the measured ones. The average error of the inversion is increased from 51.57% to 13.32%.
Thirdly, a method of inversion of explosive source parameters based on glass traces in explosion site is constructed. According to the geometric characteristics (length, width and thickness of glass) and the distance between explosive centers, the inversion algorithm of explosive source charge is established by using neural network. The experimental results show that the average inversion error of explosive source charge is from Sadovsky. At the same time, the data continuation method of PSO-SVM is used to supplement the critical overpressure value of glass with small thickness, which provides a data guarantee for the use of the critical overpressure database of glass failure in later period.
Fourthly, based on the inversion of single trace, the exploration and study of inversion of blasting source parameters are carried out systematically based on multi-trace in the explosion site. The inversion methods of blasting source parameters for three main trace factors (crater trace, glass trace, blasting vibration record) in the comprehensive explosion site are preliminarily established, including: crater-glass comprehensive inversion of blasting source parameters, blasting crater. The three-factor inversion results are compared with the single-factor inversion results. The results show that the inversion accuracy of the three-factor inversion is higher than that of the single-factor inversion.
Fifthly, the above three single factor inversion methods and three multi-factor inversion methods are applied to the core module of the software "Explosion Source Parameter Inversion System", "Explosion Source Characteristic Inversion Analysis". The software is an important means for the detection of current and future explosion cases, and provides an explosion site analysis branch for investigators and related experts. Hold tools.
The main academic contribution of this research work lies in introducing the concept and technology of data driving into the process of obtaining the information of explosive source parameters from various trace data of explosion site, and putting forward several inversion methods of explosive source parameters, which are proved to be more accurate than the traditional methods and have practical significance for explosion site analysis. It will provide technical support for the analysis and detection of explosion cases.
【学位授予单位】：北京理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：X82
本文编号：2215704
[Abstract]:Inversion of explosive source parameters is an important problem in the analysis of explosive traces. In order to obtain the information of explosive source parameters (explosive charge, explosive source buried depth / suspended height) from the explosive traces data, this paper uses the data-driven frontier technology (generalized neural network GRNN, particle swarm optimization, support direction) based on the traces data generated by explosion. SVM and non-linear unsteady-state signal analysis algorithm HHT/EMD are used to retrieve the parameters of explosive source based on the traces of explosion site.
Firstly, based on the mathematical theory of nonlinear regression, the generalized neural network is introduced into the model construction of inversion of blasting source parameters for the blasting pit traces in soil medium: the diameter and depth of the blasting pit, the soil type as the input layer of GRNN, the explosive charge and the buried depth as the output layer of GRNN, the model training and construction are carried out, and the experimental verification is carried out. Comparing the inversion results of the empirical formulas, the precision of inversion of explosive charge and depth of explosive source in clay, sandy soil and sand-clay mixed soil has been improved obviously: the relative error of explosive charge and depth obtained by inversion of explosive crater trace based on GRNN is less than 30%; the inversion error of explosive quantity and depth is equal to that of the established algorithm. The mean value is 15.41%, 16.93%, respectively, and the accuracy is obviously improved compared with the traditional empirical formula.
Secondly, the relationship between the fractal dimension and the amplitude attenuation index of explosive seismic wave is established, and the EMD/HHT data processing method, which is commonly used to analyze the energy and spectrum characteristics of explosive seismic wave vibration signals in recent years, is improved. The end effect is effectively solved and the accuracy of signal decomposition is improved by the application and optimization of PSO-SVM combined model. The local time scale intrinsic information of explosive seismic wave signal is found by using the improved EMD/HHT method, and the millisecond delay time in millisecond blasting is inverted. The inversion results are compared with the measured ones. The average error of the inversion is increased from 51.57% to 13.32%.
Thirdly, a method of inversion of explosive source parameters based on glass traces in explosion site is constructed. According to the geometric characteristics (length, width and thickness of glass) and the distance between explosive centers, the inversion algorithm of explosive source charge is established by using neural network. The experimental results show that the average inversion error of explosive source charge is from Sadovsky. At the same time, the data continuation method of PSO-SVM is used to supplement the critical overpressure value of glass with small thickness, which provides a data guarantee for the use of the critical overpressure database of glass failure in later period.
Fourthly, based on the inversion of single trace, the exploration and study of inversion of blasting source parameters are carried out systematically based on multi-trace in the explosion site. The inversion methods of blasting source parameters for three main trace factors (crater trace, glass trace, blasting vibration record) in the comprehensive explosion site are preliminarily established, including: crater-glass comprehensive inversion of blasting source parameters, blasting crater. The three-factor inversion results are compared with the single-factor inversion results. The results show that the inversion accuracy of the three-factor inversion is higher than that of the single-factor inversion.
Fifthly, the above three single factor inversion methods and three multi-factor inversion methods are applied to the core module of the software "Explosion Source Parameter Inversion System", "Explosion Source Characteristic Inversion Analysis". The software is an important means for the detection of current and future explosion cases, and provides an explosion site analysis branch for investigators and related experts. Hold tools.
The main academic contribution of this research work lies in introducing the concept and technology of data driving into the process of obtaining the information of explosive source parameters from various trace data of explosion site, and putting forward several inversion methods of explosive source parameters, which are proved to be more accurate than the traditional methods and have practical significance for explosion site analysis. It will provide technical support for the analysis and detection of explosion cases.
【学位授予单位】：北京理工大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：X82
【参考文献】
相关期刊论文 前10条
1 张彦春;罗亚平;王仲琦;王彦吉;;土介质炸点痕迹反演爆源特征的研究[J];安全与环境学报;2012年02期
2 张义平;李夕兵;赵国彦;;爆破微差延时的EMD识别法[J];地下空间与工程学报;2006年03期
3 宇德明,冯长根,徐志胜,曾庆轩,郭新亚;炸药爆炸事故冲击波、热辐射和房屋倒塌的伤害效应[J];兵工学报;1998年01期
4 宇德明,冯长根,徐志胜,曾庆轩,郭新亚;炸药爆炸事故的综合后果模型[J];兵工学报;1998年02期
5 徐全军,张庆明,恽寿榕;爆破地震峰值的神经网络预报模型[J];北京理工大学学报;1998年04期
6 焦永斌;爆破地震安全评定标准初探[J];爆破;1995年03期
7 贾光辉,王志军,张国伟,裴思行,王文龙山西省煤炭工业学校;爆炸过程中的应力波[J];爆破器材;2001年01期
8 王仲琦,宁建国,赵衡阳,恽寿榕;挡墙对远场爆炸效应影响的数值模拟研究[J];爆炸与冲击;2000年01期
9 李夕兵;张义平;刘志祥;左宇军;王卫华;;爆破震动信号的小波分析与HHT变换[J];爆炸与冲击;2005年06期
10 王民寿,郭庆海;用双随机变量回归改进爆破振速回归分析[J];爆炸与冲击;1998年03期
本文编号：2215704
本文链接：https://www.wllwen.com/kejilunwen/anquangongcheng/2215704.html
