Inversion of Seismo-Acoustic Data using Genetic Algorithms and a Posteriori Distributions
Abstract
Numerical models for sound propagation are required for sonar range assessment. These models are very advanced and practical applications are limited by the available knowledge of the environmental parameters. They can also be used as means to predict ocean and geo-acoustic parameters from acoustic measurements of the sound field. Traditionally this inversion has been carried out manually through iterative forward modelling, where unknown environmental parameters are varied in a systematic fashion until a good fit is obtained between measured and modelled data. Helped by the surge in computer power it now seems feasible to do such an inversion automatically. The current methods minimize an object function, a measure of the misfit between the observed and the computed sound fields based on the estimated environmental parameters. This optimization is complicated by the fact that the object function can have several local minima, and that the total number of parameter combinations can easily be of the order of 10,000. This calls for a global optimization procedure which can jump between the local minima, without doing an exhaustive search. Hereby only a fraction of the possible models are sampled, say lo4. In order to sample so many models infinite time, one forward model run should be done in about 1 s.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 1993
- Accession Number
- AD1119665
Entities
People
- P. Gerstoft
Organizations
- SACLANT ASW Research Centre