Most current prestack AVA joint inversion methods are based on the exact Zoeppritz equation and its various approximations. However, these equations only reflect the relation between reflection coefficients, incidence...Most current prestack AVA joint inversion methods are based on the exact Zoeppritz equation and its various approximations. However, these equations only reflect the relation between reflection coefficients, incidence angles, and elastic parameters on either side of the interface, which means that wave-propagation effects, such as spherical spreading, attenuation, transmission loss, multiples, and event mismatching of P-and S-waves, are not considered and cannot accurately describe the true propagation characteristics of seismic waves. Conventional AVA inversion methods require that these wave-propagation effects have been fully corrected or attenuated before inversion but these requirements can hardly be satisfied in practice. Using a one-dimensional(1 D) earth model, the reflectivity method can simulate the full wavefield response of seismic waves. Therefore, we propose a nonlinear multicomponent prestack AVA joint inversion method based on the vectorized reflectivity method, which uses a fast nondominated sorting genetic algorithm(NSGA II) to optimize the nonlinear multiobjective function to estimate multiple parameters, such as P-wave velocity, S-wave velocity, and density. This approach is robust because it can simultaneously cope with more than one objective function without introducing weight coefficients. Model tests prove the effectiveness of the proposed inversion method. Based on the inversion results, we find that the nonlinear prestack AVA joint inversion using the reflectivity method yields more accurate inversion results than the inversion by using the exact Zoeppritz equation when the wave-propagation effects of transmission loss and internal multiples are not completely corrected.展开更多
掺铋光纤放大器有助于将光纤通信系统拓展至新的传输波段。然而,其增益和噪声性能存在相互制约的关系,提升增益往往会导致噪声性能的恶化,反之亦然。因此,提出一种结合反向传播神经网络(BPNN)和带精英保留策略的快速非支配排序遗传算法(...掺铋光纤放大器有助于将光纤通信系统拓展至新的传输波段。然而,其增益和噪声性能存在相互制约的关系,提升增益往往会导致噪声性能的恶化,反之亦然。因此,提出一种结合反向传播神经网络(BPNN)和带精英保留策略的快速非支配排序遗传算法(NSGA-Ⅱ)的多目标优化方法,通过对两级掺铋光纤放大器结构进行设计,实现了增益和噪声性能的同时优化。使用经过训练的BPNN对增益和噪声系数预测的均方根误差分别为0.191和0.084,具有较高预测精度。以高增益和低噪声系数为目标,使用NSGA-Ⅱ算法进行优化,得到包含500个解的Pareto最优解集。优化后,放大器所能实现的平均增益范围为15~37 d B,相应的平均噪声系数范围为4.95~5.31 d B。利用BPNN代替求解耦合微分方程来评价个体适应度,使得优化时间较传统方法由106s左右降低为80 s左右,大幅提升了优化效率。所提方法也为其他掺杂光纤放大器的高效率、多目标结构优化设计提供了一种新的思路。展开更多
基金supported by the National Science and Technology Major Project(No.2016ZX05003-003)
文摘Most current prestack AVA joint inversion methods are based on the exact Zoeppritz equation and its various approximations. However, these equations only reflect the relation between reflection coefficients, incidence angles, and elastic parameters on either side of the interface, which means that wave-propagation effects, such as spherical spreading, attenuation, transmission loss, multiples, and event mismatching of P-and S-waves, are not considered and cannot accurately describe the true propagation characteristics of seismic waves. Conventional AVA inversion methods require that these wave-propagation effects have been fully corrected or attenuated before inversion but these requirements can hardly be satisfied in practice. Using a one-dimensional(1 D) earth model, the reflectivity method can simulate the full wavefield response of seismic waves. Therefore, we propose a nonlinear multicomponent prestack AVA joint inversion method based on the vectorized reflectivity method, which uses a fast nondominated sorting genetic algorithm(NSGA II) to optimize the nonlinear multiobjective function to estimate multiple parameters, such as P-wave velocity, S-wave velocity, and density. This approach is robust because it can simultaneously cope with more than one objective function without introducing weight coefficients. Model tests prove the effectiveness of the proposed inversion method. Based on the inversion results, we find that the nonlinear prestack AVA joint inversion using the reflectivity method yields more accurate inversion results than the inversion by using the exact Zoeppritz equation when the wave-propagation effects of transmission loss and internal multiples are not completely corrected.
文摘掺铋光纤放大器有助于将光纤通信系统拓展至新的传输波段。然而,其增益和噪声性能存在相互制约的关系,提升增益往往会导致噪声性能的恶化,反之亦然。因此,提出一种结合反向传播神经网络(BPNN)和带精英保留策略的快速非支配排序遗传算法(NSGA-Ⅱ)的多目标优化方法,通过对两级掺铋光纤放大器结构进行设计,实现了增益和噪声性能的同时优化。使用经过训练的BPNN对增益和噪声系数预测的均方根误差分别为0.191和0.084,具有较高预测精度。以高增益和低噪声系数为目标,使用NSGA-Ⅱ算法进行优化,得到包含500个解的Pareto最优解集。优化后,放大器所能实现的平均增益范围为15~37 d B,相应的平均噪声系数范围为4.95~5.31 d B。利用BPNN代替求解耦合微分方程来评价个体适应度,使得优化时间较传统方法由106s左右降低为80 s左右,大幅提升了优化效率。所提方法也为其他掺杂光纤放大器的高效率、多目标结构优化设计提供了一种新的思路。
