A machine learning method for predicting telescope cycle time applied to the Cerro Murphy Observatory

Abstract

Telescope cycle time estimation is one of the basic issues of observational astronomy. There are not many tools that help to calulate the cycle time for multiple telescopes with multiple instruments. This work presents a new tool for determing the observation time; it was applied at the Cerro Murphy Observatory (OCM) but can be used at any other observatory. The Machine Learning (ML) method was implied, resulting in a fully automatic software module that works without any user intervention. We propose a polynomial multiple regression method and demonstrate all steps to build a reliable ML model like data collecting, data cleaning, model training and error evaluation in relation to the implementation in the observatory software. The method was designed to work for different telescopes with several instruments. Accuracy analysis and the assessment of model errors were based on real data from telescopes, proving the usefulness of the presented method. Error evaluation shows that for 84.2 % of nights, the prediction error in operation time prediction does not exceed 2 %. Converted into a 10-hour observation night, 2 % corresponds to an error of no more than 12 minutes. The described model is already working at the OCM and optimizes the efficiency of the observations.