The processes of collisions of relativistic nuclei are quite complex and include several stages. The main iportant one is the so-called thermal expansion of the medium, during which the matter is in the stage of local thermodynamic equilibrium. The behavior of such a medium is described by the equations of relativistic hydrodynamics. Due to the fact that the initial configurations of colliding
nuclei are unique for each collision, this process must be described event by event, which requires significant computational resources.
In this work, for the approximate solution of the equations of relativistic hydrodynamics, the ML-approach based on convolutional neural networks is used. The set of Pb-Pb collisions at LHC energy generated within the VISH2+1 package was used as input data. Other types of nuclei were also used for validation at different energies. Several machine learning schemes have been developed, most
of which have shown high efficiency to reconstruct the full space-time evolution of the medium up to the moment of the freezeout. The applicability of the approaches was validated on synthetic data.
The azimuthal flows were calculated and copared with the results of the full modelling