|
|
|
Multi-Objective Optimal Scheduling for Space Science Applications
|
|
|
| We propose to develop and assess a new multi-objective multi-participant optimization application for scheduling space-based
observatories. Our approach will address how to optimize scheduling given a set of competing global objectives for a mission as well as
a large set of operational and scientific constraints. We will initially focus on James Webb Space Telescope (JWST) as it is under
development and is representative of a broad class of science scheduling problems. We will also consider the requirements of the Space
Interferometry Mission (SIM) to ensure and validate the generality of our approach. Multi-objective means retaining and exploiting
information about multiple, often competing objectives, during schedule optimization. This is in contrast to optimization techniques
that require the consolidation of all objectives into a single quantity to be optimized. Consolidation necessarily implies committing in
advance to the relative importance of different objectives, and hides information that can be used to make scheduling more effective.
Multi-participant means architecting and designing the system to allow multiple users to work with the system at one time, generally in
collaborative ways, but possibly independently. This is in contrast to approaches based on a single central scheduling process. By
incorporating both of these capabilities into a single framework we gain more than we would from either one separately. This is
because it is frequently the case that competing objectives arise directly from the various participants, and balancing these objectives
requires the participants to cooperate and compromise. This work will advance the state of the art for observatory scheduling in three
ways: (1) by providing simulation and trade study capabilities that will facilitate key early design choices; (2) by scheduling more
responsively in reaction to interesting events or system capability changes, and (3) by optimizing competing objectives of the mission so
that the overall science schedule is more efficient, and robust. The key elements of our approach are: a) a model-based representation
of objectives, constraints, and preferences, which will provide a basis for schedule optimization and for flexible modification as the
mission progresses b) an architecture to support multi-participant scheduling with both global and per-participant objectives c)
multi-objective optimization algorithms to find and characterize Pareto-optimal solutions d) an experiment plan to systematically
investigate and document our results e) the application of our approach to two distinct observatory missions, as a demonstration of its
generality and applicability. Our approach will leverage our experience and familiarity with science planning and scheduling in past
missions, including HST, Spitzer, Chandra, EUVE, FUSE, etc. from both operational and software perspectives. We will assess and
incorporate available tools and approaches, such as existing scheduling modeling languages, algorithms and technologies (e.g. as
embodied in Spike, ASPEN, and others), science simulators (e.g. JMS, the JWST Mission Simulator), and user interface elements. We
will explicitly consider and design for the future incorporation of constraints, preferences, and objectives not yet defined, by providing
component APIs and extension points. Improvements to scheduling applications have a proven track record for increasing science
return and lowering operations costs for space missions. For example, scheduling improvements allow the Hubble Space Telescope to
routinely achieve a weekly scheduling efficiency higher than thought possible at launch. Our proposal builds on this work and our
success criteria include the demonstration of a simultaneous up to 20% increase in scheduled science activities for JWST along with a
reduction of up to 20% resource consumption that can lead to a prolonged mission lifetime. |
|
|
