基于遗传BP神经网络反演楔体参数的研究  

Research on genetic BP neural network based wedge parameter inversion

作　　者：张雨[1] 贾静[1] 韩庆邦[1] 姜学平[1] 单鸣雷[1] 朱昌平[1] 

机构地区：[1]河海大学物联网工程学院,江苏常州213022

出　　处：《声学技术》2017年第1期1-5,共5页Technical Acoustics

基　　金：国家自然科学基金(11574072;11274091);江苏省重点研发计划(BE2016056);河海大学中央高校基金项目(2015B04714;2015B04614)资助项目

摘　　要：为了获得未知楔体的参数,建立了遗传算法和反向传播(Back Propagation,BP)神经网络结合的反演模型。仿真得到不同角度、密度、杨氏模量下楔体导波的频散曲线。采用反对称第一阶模态相速度数据作为遗传BP神经网络反演模型的输入变量;利用遗传算法改进BP神经网络获得优化的初始权值和阈值,并对BP神经网络进行训练;最后将实测的楔体一阶模态相速度代入训练好的网络进行参数反演。结果表明,通过该反演模型可同时反演出楔体的角度、密度、杨氏模量,并且较单一BP神经网络具有收敛速度快、精度高的优点。In order to obtain the material parameters of an unknown wedge, an inversion model based on back propagation neural network combined with genetic algorithm is established. The wedge wave dispersion curves with different angles, density and young modulus are obtained by simulation. The phase velocity of the first mode in the anti-symmetrical flexural modes is chosen as the inputs of the established model. Genetic algorithm is introduced to get the optimized initial weight and threshold. Then the optimized results are taken to train the BP neural network.The first mode data measured from samples are used as the inputs of the network that has been trained to get the inversion results. It is found that the inversion model can be used to inverse angle, density and young modulus simultaneously. Compared with the single BP neural network, combining genetic algorithm has the advantages in fast convergence speed and high precision.

关 键 词：反演 楔体导波 频散 BP神经网络 遗传算法 

分 类 号：TB551[理学—物理]