Oceanographic Experiment Design II: Genetic Algorithms

Abstract

The oceanographic experiment design problem is one of many problems in oceanography requiring nonlinear, constrained, global optimization. Having already revisited a steady-state (time-independent) experiment in acoustic tomography using simulated annealing, the experiment design problem is explored further in terms of time-dependent observational strategies. An example drawn from tracer studies is used to illustrate many of the issues. In particular, the optimization of an objective function, which characterizes the quality of an observational strategy, is carried out using a genetic algorithm (GA). Comparison with simulated annealing (SA) and another (problem specific) heuristic method is carried out. The genetic algorithm is found to be significantly faster than simulated annealing for all cases considered. The simple heuristic method is faster than either GA or SA but fails to find the optimum.