基于BP神经网络的爆炸用激波管峰值压力预测方法  

A method for predicting peak pressure in an explosion shock tube based on BP neural network

作　　者：陈梓薇 王仲琦[1] 曾令辉[1] CHEN Ziwei;WANG Zhongqi;ZENG Linghui(State Key Laboratory of Explosive Science and Technology,Beijing Institute of Technology,Beijing 100081,China)

机构地区：[1]北京理工大学爆炸科学与技术国家重点实验室,北京100081

出　　处：《爆炸与冲击》2024年第5期130-139,共10页Explosion and Shock Waves

基　　金：国家重点研发计划(2021YFC3001204)。

摘　　要：针对爆炸用激波管缺乏相应的经验公式和数值模拟时效性差的问题,同时为了快速得到激波管内的峰值压力,建立预测爆炸用激波管试验段峰值压力的四层反向传播(back propagation,BP)神经网络。采用数值模拟方法计算激波管试验段峰值压力,计算结果与激波管爆炸试验结果进行对比,平均相对误差为2.69%。证明激波管数值模型的准确性后,将数值模拟得到的195组激波管测得的峰值压力作为输出层,激波管驱动段TNT的药量、药柱的长径比以及爆炸比例距离作为神经网络的输入层。为了加快神经网络迭代速度和提高预测精度,使用自适应矩估计(adaptive moment estimation,ADAM)算法作为神经网络误差梯度下降的优化算法。结果表明,训练好的神经网络得到的预测结果与模拟值基本吻合,预测结果与数值模拟结果的平均相对误差为3.26%。BP神经网络模型能够反映激波管爆炸的峰值压力与影响因素之间的映射关系,采用BP神经网络模型计算时比数值模拟节约了大量运算时间。In response to the problems of the lack of corresponding empirical formulas and the poor timeliness of simulation for the explosive shock tube,and to quickly obtain the peak pressure of the shock tube used in explosions,a four-layer back propagation(BP)neural network was established to predict the peak pressure in the experimental section of the shock tube.After verifying the grid independence,numerical simulation was used to calculate the peak pressure of the test section of the shock tube,and the simulation data were compared with the experimental data of the shock tube explosion,and the average relative error is 2.49%.After proving the accuracy of the numerical simulation values,the 195 sets of peak pressure obtained from the numerical simulation in the shock tube test section were used as the output layer,and the TNT dosage in the shock tube driving section,aspect ratio of the charge column,and explosion proportional distance were used as the input layer for BP neural network training.To speed up the neural network iterations and increase the prediction accuracy,Adam's algorithm was used as an optimization algorithm for neural network error gradient descent.The results show that the predicted results obtained through the trained neural network are basically consistent with the simulated values,and the average relative error between the predicted results and the numerical values is 3.26%.In contrast to the evaluation metrics obtained using multiple regression analysis(mean absolute error(MAE)of 480 and coefficient of determination(R2)of 0.58),the four-layer BP neural network obtains a MAE of 25.4 and an R2 of 0.99 for the validation set. The BP neural network model can reflect the mapping relationship between the peak pressure of the shock tube explosion and the influencing factors, and improve several times compared with the time required for numerical simulation, so it has the value of practical engineering applications.

关 键 词：BP神经网络 激波管 峰值压力 自适应矩估计 

分 类 号：O383[理学—流体力学]