Inversion and fast optimization using computational intelligence with applications to geoacoustics.

摘要

With a sufficiently complex underwater acoustic model, one may produce an arbitrarily accurate reconstruction of acoustic energy propagation in any specified underwater environment. Problems arise, however, when these acoustic emulations are required in a timely manner. When many realizations of the acoustic model are required over a short period of time, model complexity prohibits any kind of fast execution of such an algorithm. Two approaches may be applied to increasing the speed of any such iterative technique: first, one may attempt to simplify or speed up the model. Second, one may attempt to reduce the number of times the complex model must be executed. In this dissertation, we take both approaches for two distinct, unsolved problems in the area of geoacoustics: inversion of acoustic models for bottom parameter acquisition, and sonobuoy placement for optimal sonar coverage of a desired area, and we will see both may be phrased as optimization problems. The primary focus of this paper, however, is specifically on the use of computational intelligence to increase the execution time of these optimization algorithms, including a very remarkable greedy algorithm for the placement of sonobuoys, which executes in time orders of magnitude lower than with direct optimization techniques.