论文部分内容阅读
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。
Matching Field Statistics The fundamental purpose of inverse submarine acoustic parameters is to solve the posterior probability distribution (PPD) of unknown parameters. In view of the existing numerical methods for solving PPD such as exhaustive search, Markov Chain Monte Carlo sampling, nearest neighbor interpolation approximation algorithms generally have the problems of slow calculation speed and long time duration, and are difficult to meet the practical application. In this paper, A New Algorithm for Solving Parameter PPD Fast with Support Vector Machine. The algorithm makes use of the strong ability of SVM for small sample learning. By training and fitting the function relationship between the unknown underwater acoustic parameters and the posterior probability, the algorithm simplifies the calculation of the posterior probability by using the sound field propagation model when solving the parameter PPD The complexity of the process, reducing the calculation time. The results of numerical simulation and experimental data at sea validate the effectiveness of this algorithm in inverse acoustic parameters inversion in low-dimensional matched fields.