Solution to the Problem of Diffusion from a Ring Source taking into Account Meteorological Conditions, using Kolmogorov–Arnold Networks

Abstract

This article examines the issue of pollutant dispersion from a ring source in the atmosphere, taking into account meteorological parameters (wind speed and direction and atmospheric stability class). It proposes a novel approach based on physically informed Kolmogorov–Arnold networks for solving the stationary convection-diffusion equation and compares it to a traditional fully connected neural network in a similar formulation. The study demonstrates that the KAN architecture provides higher accuracy with significantly fewer parameters and better interpretability. The results of the numerical experiments show the dependence of the plume shape on meteorological conditions and provide a convergence analysis.