Аннотация:The paper develops a method for detecting extremely close visual double stars based on the use of astrometric catalogues in combination with machine learning (ML) methods. The low probability of random occurrence of such close neighbours allows us to speak about the physical coupling of the detected stars. These stars can therefore be considered binary. A computational experiment was carried out on the example of the Hipparcos mission catalogue and the Pan-STARRS (PS1) catalogue by applying the suggested method. It has shown that the reliability of predicting a stellar multiplicity reaches 90--95%. We note the prospects and effectiveness of creating a proprietary research platform --- Cognotron.
