一种基于支持向量机的海底声学参数快速统计反演方法
A fast algorithm for estimating posterior probability distributions of unknown parameters based on support vector machine in matched-field statistical inversion
-
摘要: 匹配场统计反演海底声参数的根本目的是求解未知参数的后验概率分布(PPD)。针对现有各种求解参数PPD的数值方法如穷举搜索、Markov Chain Monte Carlo采样、最近邻域插值近似算法普遍存在计算速度慢、时间长、难以满足实际应用的问题,本文提出了一种基于支持向量机的快速求解参数PPD的新算法。该算法利用了支持向量机强大的小样本学习能力,通过训练学习拟合未知海底声参数和后验概率之间存在的函数关系,从而在求解参数PPD时简化了利用声场传播模型计算后验概率的复杂过程,减少了计算时间。数值仿真算例和海上实验数据的处理结果验证了该算法在低维匹配场统计反演海底声参数问题中的有效性。Abstract: The goal of matched-field statistical inversion is to derive the posterior probability distributions (PPD) of unknown seafioor parameters from the measured ocean acoustic data.However,traditional approaches to estimate the PPD such as exhaustive searching,Markov Chain Monte Carlo sampling and nearest neighborhood interpolation approximate algorithm are all computationally slow and not suitable for the practical applications.To remedy their drawback, this paper proposed a new fast algorithm based on support vector machine.Our proposed algorithm simplifies the heavy and complex procedure for calculating the posterior probability by forward model and requires less computational time. The numerical and experimental examples validate the proposed algorithm for low-dimensional matched-field inversion.