The article considers the equation of transport and diffusion for mathematical modeling of the process of distribution of harmful impurities in the atmosphere. The paper explores an explicit difference scheme for the transfer and diffusion equation, as well as the architecture of physically informed neural networks (PINNS), their principle of operation and application to solving the transfer and diffusion equation. In the course of a practical study, the wind direction was studied in various conditions and their different effects on the distribution of the mixture were discussed.  The dynamics of the distribution of pollutants in the urban area in accordance with the north-west and north-east winds observed in the city of Ust-Kamenogorsk is shown. As a result of the study, graphs of numerical solutions to the problem were shown, patterns of the distribution of harmful impurities in the atmosphere were determined, making sure that the direction and speed of the wind have a direct effect on the distribution of harmful substances in the atmosphere.