Multi-Objective Optimization Under Uncertainty of Satellite Systems Via Simulated Annealing

Abstract

A method for performing multi-objective optimization under uncertainty in conceptuallevel multidisciplinary design is presented. The method is applied to a ship tracking and environmental protection satellite design problem seeking to optimize three outputs: coverage time, resolution, and total cost. Uncertainties are quantified and propagated using Monte Carlo simulation. Optimization is then performed via a multi-objective simulated annealing algorithm on each Monte Carlo sample. The single solution is selected as the best solution (according to a weighted sum of the outputs) from a tail-sample Pareto set. A composite solution is obtained as a composite of a subset of best solutions. This subset of best solutions consists of a user-defined number of solutions from all Pareto sets above a certain confidence level. A baseline solution, a deterministic multi-objective simulated annealing solution, the single solution, and the composite solution are compared. The optimizationbased solutions all provide better solutions than the baseline system. The composite solution provides the best solution but a greater computational expense than the deterministic solution. A comparison of multi-spectral imager based systems is also made. The composite solution is again found to be the best solution especially under uncertainty where the deterministic and single-best solutions suffer from dramatic increases in the total cost at higher confidence-levels.