Simplex simulated annealing: A hybrid approach to geoacoustic inversion

The problem of matched-field inversion of acoustic fields for ocean-bottom geoacoustic properties has received considerable attention in recent years. In order to avoid the large number of local minima typical of multidimensional geoacoustic parameter spaces, many inversions apply global optimization methods, such as simulated annealing (SA), to minimize the mismatch between measured and modeled fields. However, the random manner in which SA searches the parameter space can be quite inefficient, particularly near convergence and/or in cases where the minima represent narrow valleys not aligned with the parameter axes. In addition, standard SA has no memory, so a good solution may be discarded at the early stages and never revisited. In order to improve the efficiency and effectiveness of SA, a local inversion component based on the downhill simplex method within a global approach to geoacoustic inversion has been incorporated. The simplex method is sensitive to local gradients, and provides efficient perturbations at all stages which are independent of the coordinate axes. Also, the simplex of models carried through the inversion retains good models and provides a memory for the algorithm. The hybrid approach of simplex simulated annealing will be described and illustrated for geoacoustic inversion. [For Underwater Acoustics Best Student Paper Award.]