Neural networks in air turbulence modelling

Local and integral (corrective, correcting Abramovich-Sekundov and Spalart-Allmaras models) neural network models of turbulent viscosity are proposed, showing intermediate accuracy results between one-parameter and two-parameter models. A new algorithm for controlling the numerical calculation error in some typical problems of aerodynamics has been proposed. Its effectiveness on a number of problems is shown. The algorithm is based on the application of particle indicators and uses the particle-in-cell method. The principles of data compression, representing physical quantities varying depending on some parameter, by constructing cluster-neural network descriptions, are proposed.