CSC at NASA Ames Abstract

Scheduling observations by coordinated fleets of Earth Observing Satellites (EOS) involves large search spaces, complex constraints and poorly understood bottlenecks; conditions where stochastic algorithms are often effective. However, there are many such algorithms and the best one to use is not obvious. Here we compare multiple variants of the genetic algorithm, hill climbing, simulated annealing, squeaky wheel optimization and iterated sampling on ten realisticallysized EOS scheduling problems. Schedules are represented by a permutation (non-temperal ordering) of the observation requests. A simple, greedy, deterministic scheduler assigns times and resources to each observation request in the order indicated by the permutation, discarding those that violate the constraints created by previously scheduled observations. Simulated annealing performs best and random mutation outperforms a more ’intelligent ’ mutator. Furthermore, the best mutator, by a small margin, was a novel approach we call ’temperature-dependent swap ’ that makes large changes in the early stages of the search and smaller changes towards the end