基于MCMC算法改进的贝叶斯海底声参数反演方法研究

Research on inversion method of Bayesian bottom acoustic parameters based on improved MCMC algorithm

在半无限弹性海底环境下,针对海底声参数的非线性反演问题,采用贝叶斯反演方法,分析垂直分布水听器所接收的不同深度点上的复声压,以反演海底声速、密度、声速衰减3类海底声参数.为加速反演收敛,对贝叶斯参数反演中传统的MCMC算法进行了改进,采样过程中分别引入多步退火算法和并行回火算法,有效解决MCMC算法所致的收敛缓慢问题.使用数值模拟实验比较MCMC算法与2种改进算法的反演效果,仿真结果表明,多步退火算法和并行回火算法均能有效提高反演效率,且并行回火采样算法可得到精度更高、均方差更小的反演结果,是一种更优的MCMC改进算法.

In a semi-infinite elastic seabed environment,in order to solve the problem of nonlinear inversion of seabed acoustic parameters,the Bayesian inversion method is used to analyze the complex sound pressure at different depth points received by the vertically distributed hydrophone to invert the seabed sound velocity,Density,and sound velocity attenuation three types of bottom acoustic parameters.In order to accelerate the convergence of the inversion,the traditional MCMC algorithm in Bayesian parameter inversion is improved,and the multi-step annealing algorithm and the parallel tempering algorithm are respectively introduced in the sampling process to effectively solve the slow convergence problem caused by the MCMC algorithm.Numerical simulation experiments are used to compare the inversion effects of the MCMC algorithm and the two improved algorithms.The simulation results show that both the multi-step annealing algorithm and the parallel tempering algorithm can effectively improve the inversion efficiency,and the parallel tempering sampling algorithm can obtain higher accuracy.The inversion result with a smaller mean square error is a better MCMC improved algorithm.