Evaluating Multi-Objective Evolutionary Scheduling Strategies for the James Webb Space Telescope

Effective scheduling of the James Webb Space Telescope (JWST) requires careful management of the trade-offs among multiple scheduling criteria. Minimizing unscheduled time, angular momentum build-up, and the number of observations that miss their last opportunity to schedule are important objectives that are in competition with each other. Previous studies evaluated alternative multi-objective algorithms for scheduling JWST, including algorithms for breaking down the search into separately assigning and optimizing schedule times and spacecraft roll. This study builds on these results by enriching the JWST scheduling model to include a high fidelity maneuver model between observations, along with realistic science restrictions on how spacecraft roll can be assigned to observations. Additional multi-objective search strategies are evaluated using a parallelized multi-objective algorithm that can exploit available computing resources to reduce overall runtime. We have also incorporated constraints into the problem formulation, to drive the candidate solution set to better sample the Pareto frontier in regions of greater interest. The results not only demonstrate techniques useful for scheduling JWST but also show how multi-objective algorithm features such as parallelism, constraints, and the choice of population size impacts the quality of the schedules generated. Each of the algorithms explored produces a Pareto frontier as output. As a result there is no canonical way of comparing which algorithm performs better. This paper presents tools and techniques for comparing multi-objective algorithms and presents the analogy that this evaluation is essentially a recursive application of multi-objective optimization