Äîêóìåíò âçÿò èç êýøà ïîèñêîâîé ìàøèíû. Àäðåñ îðèãèíàëüíîãî äîêóìåíòà : http://www.stsci.edu/~miller/papers-and-meetings/95-nmo/millerg1.ps Äàòà èçìåíåíèÿ: Thu Nov 22 00:04:36 2007 Äàòà èíäåêñèðîâàíèÿ: Sat Dec 22 18:53:25 2007 Êîäèðîâêà: Ïîèñêîâûå ñëîâà: ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï ï

A System for Long­Term Scheduling of Ground­Based
Observatories
Glenn E. Miller
Space Telescope Science Institute
Abstract.
Most ground­based observatories grant blocks of consecutive nights
(or half nights) to observers, taking into account a variety of constraints
and restrictions. Creating this schedule for an observing semester typ­
ically takes several days of a senior astronomer's time. This paper de­
scribes an adaptation of the Spike scheduling system to the long­term
scheduling of ground­based observatories. The system accounts for the
relevant factors such as moon phase, position of targets, and observer's
date preferences. In addition it accounts for observatory­level factors
such as minimizing telescope and instrument changes, priorities, and re­
serving time for engineering operations. The system provides both batch
and interactive modes of operation. A variety of automated scheduling
strategies can be run and the best schedules selected from them. Graphi­
cal displays present the timeline of observations and allow for observations
to be adjusted manually.
1. Introduction
The Spike system (Johnston and Miller 1994, Johnston 1995) was developed by
the Space Telescope Science Institute to support planning and scheduling for
NASA's Hubble Space Telescope (Miller 1995). Generality was a prime design
criterion and Spike has been applied to several scheduling problems including
other spacecraft, ground­based observatories and ``job shop'' scheduling prob­
lems (Johnston and Minton 1994). Chavan, Johnston, and Albrecht (1995) de­
scribe scheduling tools for ESO's Very Large Telescope (VLT), including recent
extensions of Spike for long­ and short­term ground­based telescope scheduling.
In this article I describe the long­term scheduler in more detail and use the
Canada­France­Hawaii telescope (CFHT) schedule as an example.
The general problem is to schedule blocks of time (nights and half­nights)
obeying the proposer's scientific constraints, the observatory's constraints, and
constraints on the overall program such as institutional fractions.
2. Scheduling Data
An observing program is defined by the following items:
ffl Program ID, principal investigator name, institutional affiliation and coun­
try
1

ffl Instrument (which implies the telescope configuration, e.g. cassegrain or
coude)
ffl Amount of time required (integral or half day)
ffl Moon preference ­ this can be expressed as the number of days from new
moon, or more simply as one of ``bright'', ``dark'' or ``no constraint''.
ffl Target visibility preference ­ this can be expressed as preferences on partic­
ular months, or a range in right ascension and declination for the targets.
ffl Time Constraints ­ this expresses time intervals where the observations
can occur and can either be expressed as dates to be included or excluded
from consideration.
ffl Relative constraints between program, e.g. program 2 after program 1 or
group two programs within a time interval.
ffl Priority assigned by observatory or time allocation authority.
For CFHT scheduling, the input data was provided in a tabular text file with
each program defined in a row and each column corresponding to a parameter.
A simple function converted this information to the generic input format used
by the Spike scheduler.
In addition to the program information, it is necessary to define the start
and end dates of the scheduling interval (typically a 6 month semester) and any
reserved times where no observations are made (e.g. major holidays or engineer­
ing times). An additional input is a description of the telescope instrumentation
(e.g. instruments and associated telescope foci) and constraints on instrument
changes (e.g. some instrument changes may require a half­day or more of ``down
time'' or a limit on the total number of upper­end changes in a semester).
3. Usage
Once the programs are loaded into the system, the user operates the system
via a graphical interface (see Figure 1). The large panel at the bottom shows
the schedule timeline, labelled with the dates and program ids. Clicking on
a program will display its preferences and conflicts in this panel, as well as
further details in the two panels on the upper left. ``Preference'' (or ``suitability'')
indicates the desirability of scheduling a program at a particular time, while
``conflicts'' shows if any constraints have been violated (see Johnston and Miller
1994 for more details). The middle panel shows the results of various scheduling
runs. This includes measures of the quality of each schedule such as the mean
and total preference, number of unscheduled programs, number of constraint
violations, and total gaps in schedule. Clicking on a schedule restores it to the
lower display. The system also produces a postscript display of the telescope
schedule. The two panels in the upper right summarize the schedule status and
present a menu of commands.
The system encourages ``mixed­initiative'' scheduling where the user can
make manual scheduling decisions but also use automated scheduling strategies.
2

Figure 1. Long­term scheduler user interface
3

For example, clicking on a time in the lower panel of Figure 1 schedules the
selected program at that time. This choice can be ``locked'' so that runs of the
automated strategies do not reschedule this program. This mix of scheduling
modes is important since a fully automatic system cannot make the tradeoffs
nor constraint relaxations that are typically required while scheduling a telescope
(e.g. relaxing a programs's moon requirements or granting the program slightly
more or less time).
Generating a schedule is fast ­ with the 65 programs in the CFHT schedule,
the automated strategies run in tens of seconds on a Sparcstation 2.
4. Scheduling Strategy to Minimize Instrumentation Changes
Although quick instrument changes are often a design goal, it is quite common for
there to be significant operational costs in going from one telescope/instrument
configuration to another. In constructing a long­term scheduler for ESO tele­
scopes, Johnston, Chavan and Albrecht (1995) employed a cost function for
installation of the IRSPEC since a one day instrument change time is needed
in this case. Since the scheduler minimizes the amount of unused time and
maximizes the number of observations scheduled, this cost function drives the
schedule towards a minimum number of instrument changes. In the case of
CFHT, certain instrument changes require a telescope upper­end change (e.g.
from cassegrain to coude). It is desirable to minimize the number of upper­
end changes to the degree consistent with other constraints and there is also
a firm limit of 11 upper­end changes in a semester. Unlike ESO case, there is
no observing time lost due to a change so a different technique was employed.
The scheduling strategy was modified to select observations on the basis of the
upper­end and instrument being used.
Acknowledgments. Mark Johnston (STScI) developed the core extensions
to Spike for ground­based scheduling and Johnston and Maurizio Chavan (ESO)
developed the most recent extensions which were used for this work. John
Glaspey (CFHT) provided the scheduling data and constraints for CFHT along
with substantial encouragement.
References
Chavan, A.M., Johnston, M.D. and Albrecht, M.A. 1995, this volume.
Johnston, M.D. 1995, this volume.
Johnston, M. and Miller, G. 1994 in Intelligent Scheduling, ed. M. Fox and M.
Zweben, (San Francisco: Morgan­Kaufmann), pp 391­422.
Johnston, M. and Minton, S.. 1994 in Intelligent Scheduling, ed. M. Fox and
M. Zweben, (San Francisco: Morgan­Kaufmann), pp 257­289.
Miller, G.E. 1995, this volume.
4