Geoacoustic inversion hybrid model

Abstract

Simulated annealing (SA) or genetic algorithms are accepted optimization methods for geoacoustic inversion to obtain seabed parameters from measured acoustic data. These techniques permit optimizations where numerous local minima are present, but are computationally intensive. Gradient methods, such as the Levenberg–Marquardt (LM) approach, are extremely efficient at finding local minima, but typically are unable to locate the global minimum. To improve the computation speed and efficiency of SA, a ‘‘hybrid’’ algorithm combining SA and LM has been developed. The hybrid approach applies a LM algorithm after each temperature cycle of SA if an improved cost function value has been obtained during the cycle. The hybrid technique takes advantage of the fact that SA often examines a point that is in the proximity of the global minimum when the temperature is high. Factors of 3–8 increases in speed over SA alone are obtained with simulated data. Results will be presented from the application of the hybrid model to experimental data collected during the Area Characterization Test-I in the Gulf of Mexico. [Work supported by ONR.]