Prof Iain Steele, Dr Robert Smith, Dr Chris Copperwheat
A Machine Learning Approach to Telescope Scheduling
The optimal scheduling of observations for both individual and arrays of telescopes is of increasing importance in astronomy. With increasing robotisation of telescope operations and the emergence of larger (and more expensive to operate) facilities such as LSST, maximising the scientific return becomes critical. Telescope scheduling is a complex problem, especially due to the dynamic nature of the atmospheric conditions and the desire to respond rapidly to targets of opportunity. Traditional approaches to this problem rely on using a time invariant “dispatch scheduling” approach which computes a weighting function to decide what to do next without any significant longer term planning. Recent advances in machine learning have however raised the possibility of dynamically changing the scheduling approach based on the current and predicted conditions. This is analogous to how a human observer may (consciously or subconsciously) adapt their observing approach (for example by doing observations with shorter exposure times in changeable conditions so that a long observation is not wasted if the conditions change during execution). The project will use machine learning techniques to analyse the schedules of conventionally and robotically operated telescopes and to develop an adaptive scheduling system that can be baselined against other approaches. It is intended that the work done will form the basis of the scheduling system for the new 4.0 metre robotic telescope being planned by LJMU.