基于PSO-CNN-XGBoost水下柱形装药峰值超压预测

doi:10.12382/bgxb.2023.0743

摘要/Abstract

摘要：

为探索水下柱形装药结构、爆距等参数与水下柱形装药峰值超压的关系,将装药样本数据视为二维数据,建立粒子群优化(Particle Swarm Optimization, PSO)算法、一维卷积神经网络(1D Convolutional Neural Network,1DCNN)和极端梯度提升(Extreme Gradient Boosting, XGBoost)的水下柱形装药峰值超压融合预测算法。采用相关性分析与数据可视化方法,分析装药结构参数、爆距与峰值超压之间的关联关系。设计1DCNN深度网络挖掘不同长径比、爆距等参数与峰值超压之间的纵向时序关系。运用XGBoost算法寻找装药结构参数、爆距与峰值超压之间的横向非线性关系,提升小样本数据的预测精度。使用PSO算法优化1DCNN和XGBoost的超参数,获得最优算法结构。研究结果表明,在包含10种智能算法的对比实验中,PSO-CNN-XGBoost水下柱形装药峰值超压预测算法在精度、稳定性、拟合程度上均高于其他模型。

关键词: 水下柱形装药, 长径比, 爆距, 峰值超压, 粒子群优化算法, 一维卷积神经网络, 极端梯度提升

Abstract:

In order to explore the relationships among the peak overpressure of underwater cylindrical charge and the parameters of its structure and blast distance, the sample data of the charge is regarded as two-dimensional data.A peak overpressure fusion prediction algorithm of underwater cylindrical charge is established. The fusion prediction algorithm is based on particle swarm optimization(PSO)algorithm,1D convolutional neural network(1DCNN) and Extreme Gradient Boosting(XGBoost). Correlation analysis and data visualization methods are used to analyze the correlation among charge structure parameters, blast distance and peak overpressure. 1DCNN deep network is designed to mine the longitudinal temporal relationship among the parameters, including aspect ratio, blast distance, etc., and peak overpressure. XGBoost algorithm is applied to find the lateral nonlinear relationships among charge structure parameters, blast distance and peak overpressure, so as to improve the prediction accuracy of small sample data. PSO algorithm is used to optimize the hyperparameters of 1DCNN and XGBoost, and gain the optimal algorithm structure. In the comparison experiments involving ten intelligent algorithms, the accuracy, stability and fitting degree of PSO-CNN-XGBoost underwater cylindrical charge peak overpressure prediction algorithm are higher than those of other algorithms.

Key words: underwater cylindrical charge, aspect ratio, blast distance, peak overpressure, particle swarm optimization algorithm, 1D convolutional neural network, extreme gradient boosting

中图分类号:

O383⁺.1

刘芳, 李士伟, 卢熹, 郭策安. 基于PSO-CNN-XGBoost水下柱形装药峰值超压预测[J]. 兵工学报, 2024, 45(5): 1602-1612.

LIU Fang, LI Shiwei, LU Xi, GUO Ce’an. Prediction of Peak Overpressure of Underwater Cylindrical Charge Based on PSO-CNN-XGBoost[J]. Acta Armamentarii, 2024, 45(5): 1602-1612.

图/表 20

参考文献 22

[1]	聂源, 蒋建伟, 门建兵. 考虑环境温、湿度的球形装药爆炸冲击波参数计算模型[J]. 爆炸与冲击, 2018, 38(4):735-742.
	NIE Y, JIANG J W, MEN J B. Calculation models for parameters of spherical charge blasting shock wave considering ambient temperature and humidity[J]. Explosion and Shock Waves, 2018, 38(4):735-742. (in Chinese)
[2]	王树山, 梁策, 高源, 等. 深水爆炸二次压力波超压峰值的工程模型[J]. 兵工学报, 2022, 43(10):2508-2516.
	WANG S S, LIANG C, GAO Y, et al. Engineering model for calculating secondary pressure wave overpressure peak in deep water explosion[J]. Acta Armamentarii, 2022, 43(10):2508-2516. (in Chinese) doi: 10.12382/bgxb.2021.0560
[3]	徐豫新, 蔡子雷, 吴巍, 等. 药毁伤效能评估技术研究现状与发展趋势[J]. 北京理工大学学报, 2021, 41(6):569-578.
	XU Y X. CAI Z L, WU W, et al. Current research and development of ammunition damage effect assessment technology[J]. Transactions of Beijing Institute of Technology, 2021, 41(6):569-578. (in Chinese)
[4]	杨秉妍, 范瑞军, 江自生, 等. 活性元对低附带毁伤弹药的近场超压增强效应[J]. 高压物理学报, 2022, 36(6):164-172.
	YANG B Y, FAN R J, JIANG Z S, et al. Effect of near-field overpressure enhancement of reactive material on low collateral damage ammunition[J]. Chinese Journal of High Pressure Physics, 2022, 36(6): 164-172. (in Chinese)
[5]	YANG G D, FAN Y, WANG G H, et al. Experimental and numerical investigation on dynamic behavior of RC orifice targets subjected to underwater explosions[J]. Engineering Structures, 2023, 279:115541.
[6]	MA X J, KONG D R, SHI Y C. Experimental and numerical investigation of blast loads induced by moving charge explosion[J]. Structures, 2023, 47: 2037-2049.
[7]	COLE R H. Underwaterexplosions[M]. NJ, US: Princeton University Press, 1948.
[8]	WANG L Q, KONG D R. Research on the distribution characteristics of explosive shock waves at different altitudes[J]. Defence Technology, 2023,24:340-348.
[9]	ZHANG A M, LI S M, CUI P, et al. A unified theory for bubble dynamics[J]. Physics of Fluids, 2023, 35(3):033323.
[10]	LI M K, ZHANG A M, MING F R, et al. A coupled smoothed particle hydrodynamics-finite volume method for three-dimensional modeling of bubble dynamics[J]. Physics of Fluids, 2023, 35(5): 056117.
[11]	张之凡, 李海龙, 张桂勇, 等. 聚能装药水下爆炸冲击波和侵彻体载荷作用时序研究[J]. 爆炸与冲击, 2023, 43(10):3-14.
	ZHANG Z F, LI H L, ZHANG G Y, et al. Action time sequence of underwater explosion shock wavesand shaped charge projectiles[J]. Explosion and Shock Waves, 2023, 43(10):3-14. (in Chinese)
[12]	黄洪, 卢熹, 王健. 柱形装药水下爆炸近场冲击波数值仿真[J]. 水下无人系统学报, 2021, 29(4):471-476.
	HUANG H, LU X, WANG J. Near field shock wave numerical simulation of cylindrical charge underwater explosion[J]. Journal of Unmanned Undersea Systems, 2021, 29(4): 471-476. (in Chinese)
[13]	马腾, 王金相, 刘亮涛, 等. 不同长径比柱形装药水下爆炸冲击波演化规律[J]. 振动与冲击, 2022, 41(8):149-157,222.
	MA T, WANG J X, LIU L T, et al. Shock wave evolution of cylindrical charge with different slender ratios[J]. Journal of Vibration and Shock, 2022, 41(8):149-157,222. (in Chinese)
[14]	LIU H B, SHI X M, CHEN X J, et al. Bayesian inference of ammunition consumption based on normal-inverse gamma distribution[J]. Discrete Dynamics in Nature and Society, 2022, 2022:6365712.
[15]	裘镓荣, 曾鹏飞, 邵伟平, 等. 基于PSO-LSSVM弹药装配质量预测方法[J]. 兵工学报, 2022, 43(9):2379-2387.
	QIU J R, ZENG P F, SHAO W P, et al. PSO-LSSVM-based ammunition assembly quality prediction method[J]. Acta Armamentarii, 2022, 43(9):2379-2387. (in Chinese) doi: 10.12382/bgxb.2021.0867
[16]	MA H R, SUN H, LI C S. Penetration overload prediction method based on a deep neural network with multiple inputs[J]. Applied Sciences, 2023, 13(4):2351.
[17]	李广宁, 史宪铭, 陈磊, 等. 基于LSTM神经网络的弹药消耗预测[J]. 火力与指挥控制, 2022, 47(6):75-80.
	LI G N, SHI X M, CHEN L, et al. Prediction of ammunition consumption based on long short term memory neural network[J]. Fire Control & Command Control, 2022, 47(6):75-80. (in Chinese)
[18]	DOBRATZ B M, CRAWFORD P C. LLNL explosive handbook: properties of chemical explosives and explosive simulants[M]. Livermore, CA, US: Lawrence Livermore National Laboratory,1985:21-29.
[19]	OLIVEIRA M A, MONTEIRO A V, FILHO J. A new structural health monitoring strategy based on PZT sensors and convolutional neural network[J]. Sensors, 2018, 18(9):2955.
[20]	WANG K, MA C, QIAO Y H, et al. A hybrid deep learning model with 1DCNN-LSTM-Attention networks for short-term traffic flow prediction[J]. Physica A:Statistical Mechanics and its Applications, 2021, 583:126293.
[21]	CHEN T Q, GUESTRIN C. Xgboost:a scalable tree boosting system[C]//Proceedings of the 22nd International Conference on Knowledge Discovery and Data Mining. NewYork, NY, US:ACM, 2016:785-794.
[22]	邓青, 薛青, 罗佳. 基于支持向量机的坦克驾驶模拟训练结果分析[J]. 兵工学报, 2019, 40(9):1953-1960. doi: 10.3969/j.issn.1000-1093.2019.09.021
	DENG Q, XUE Q, LUO J. Analysis of tank driving simulation training results based on support vector machine[J]. Acta Armamentarii, 2019, 40(9):1953-1960. (in Chinese) doi: 10.3969/j.issn.1000-1093.2019.09.021

参数	数值
ρ₀/(g·cm^-3)	1.55
Q/(m·s^-1)	6930
p_C-J/GPa	21
A/GPa	373.77
B/GPa	3.747
R₁	4.15
R₂	0.9
ω	0.35
E	0.07

参数	数值
ρ₀/(g·cm^-3)	1.55
Q/(m·s^-1)	6930
p_C-J/GPa	21
A/GPa	373.77
B/GPa	3.747
R₁	4.15
R₂	0.9
ω	0.35
E	0.07

序号	长径比	长度/ mm	直径/ mm	爆距/ mm	爆径比	峰值超压/ MPa
1	0.5	12.7	25.4	30	11.8	29.15
2	0.5	12.7	25.4	40	15.7	19.81
	︙	︙	︙	︙	︙	︙
28	1	20.2	20.2	50	24.8	21.35
29	1	20.2	20.2	60	29.8	16.73
30	1	20.2	20.2	70	34.7	13.60
	︙	︙	︙	︙	︙	︙
219	10	93.6	9.4	120	128.2	17.66
220	10	93.6	9.4	130	138.9	16.17

序号	长径比	长度/ mm	直径/ mm	爆距/ mm	爆径比	峰值超压/ MPa
1	0.5	12.7	25.4	30	11.8	29.15
2	0.5	12.7	25.4	40	15.7	19.81
	︙	︙	︙	︙	︙	︙
28	1	20.2	20.2	50	24.8	21.35
29	1	20.2	20.2	60	29.8	16.73
30	1	20.2	20.2	70	34.7	13.60
	︙	︙	︙	︙	︙	︙
219	10	93.6	9.4	120	128.2	17.66
220	10	93.6	9.4	130	138.9	16.17

爆距/ cm	仿真/ MPa	经验公式^[9]/ MPa	实弹实验^[11]/ MPa	误差1/ %	误差2/ %
50	21.35	19.71	19.97	8.3	6.9
60	16.73	15.71	16.45	6.5	1.7
70	13.60	12.59	13.95	8.0	2.5