Share Email Print
cover

Proceedings Paper

Artificial intelligence for the CTA Observatory scheduler
Author(s): Josep Colomé; Pau Colomer; Jordi Campreciós; Thierry Coiffard; Emma de Oña; Giovanna Pedaletti; Diego F. Torres; Alvaro Garcia-Piquer
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

The Cherenkov Telescope Array (CTA) project will be the next generation ground-based very high energy gamma-ray instrument. The success of the precursor projects (i.e., HESS, MAGIC, VERITAS) motivated the construction of this large infrastructure that is included in the roadmap of the ESFRI projects since 2008. CTA is planned to start the construction phase in 2015 and will consist of two arrays of Cherenkov telescopes operated as a proposal-driven open observatory. Two sites are foreseen at the southern and northern hemispheres. The CTA observatory will handle several observation modes and will have to operate tens of telescopes with a highly efficient and reliable control. Thus, the CTA planning tool is a key element in the control layer for the optimization of the observatory time. The main purpose of the scheduler for CTA is the allocation of multiple tasks to one single array or to multiple sub-arrays of telescopes, while maximizing the scientific return of the facility and minimizing the operational costs. The scheduler considers long- and short-term varying conditions to optimize the prioritization of tasks. A short-term scheduler provides the system with the capability to adapt, in almost real-time, the selected task to the varying execution constraints (i.e., Targets of Opportunity, health or status of the system components, environment conditions). The scheduling procedure ensures that long-term planning decisions are correctly transferred to the short-term prioritization process for a suitable selection of the next task to execute on the array. In this contribution we present the constraints to CTA task scheduling that helped classifying it as a Flexible Job-Shop Problem case and finding its optimal solution based on Artificial Intelligence techniques. We describe the scheduler prototype that uses a Guarded Discrete Stochastic Neural Network (GDSN), for an easy representation of the possible long- and short-term planning solutions, and Constraint Propagation techniques. A simulation platform, an analysis tool and different test case scenarios for CTA were developed to test the performance of the scheduler and are also described.

Paper Details

Date Published: 6 August 2014
PDF: 15 pages
Proc. SPIE 9149, Observatory Operations: Strategies, Processes, and Systems V, 91490H (6 August 2014); doi: 10.1117/12.2057090
Show Author Affiliations
Josep Colomé, Institut de Ciències de l'Espai (Spain)
Pau Colomer, GTD Sistemas de Información (Spain)
Jordi Campreciós, Institut de Ciències de l’Espai (Spain)
Thierry Coiffard, GTD Sistemas de Información (Spain)
Emma de Oña, Institut de Ciències de l'Espai (Spain)
Giovanna Pedaletti, Institut de Ciències de l'Espai (Spain)
Diego F. Torres, Institut de Ciències de l'Espai (Spain)
Alvaro Garcia-Piquer, Institut de Ciències de l'Espai (Spain)


Published in SPIE Proceedings Vol. 9149:
Observatory Operations: Strategies, Processes, and Systems V
Alison B. Peck; Chris R. Benn; Robert L. Seaman, Editor(s)

© SPIE. Terms of Use
Back to Top