Rapid prediction of airborne pollutant transport diffusion based on Reynolds-averaged Navier-Stokes and neural network

A computational fluid dynamics (CFD) model of a post-treatment plant area was constructed, and the Reynolds-Averaged Navier-Stokes (RANS) algorithm was employed to simulate the impact of chimney discharge velocity, chimney exhaust gas temperature and atmospheric inflow velocity. Simulation data were collected, organized and split into training and testing sets with an 8:2 ratio to construct a back propagation neural network (BPNN) prediction model. Particle swarm optimization (PSO) was applied to adaptively optimize the weights and thresholds and accuracy of the BPNN model. Results indicated that increasing the inflow wind speed could enhance the atmospheric dilution rate of pollutants, leading to a reduced plume lift height and a 24.12% increase in pollutant concentration near the ground. An increase in flue gas temperature would increase the plume lift height, driving the migration and diffusion of pollutants, which resulted in a 63.13% decrease in near-ground pollutant concentrations, though it had minimal effect on the dilution rate. Increasing the discharge velocity of pollutants can effectively raise the plume lift height, promoting atmospheric dispersion and improving dilution capabilities but may also cause a 23.98% increase in pollutant concentrations in the downwind area of the plant. Comparative analysis of CFD results and PSO-BPNN model predictions showed that the PSO-BPNN model can predict airborne pollutant concentrations within two minutes with an average error of 5.46% in pollutant concentration, meeting the requirements for both accuracy and real-time performance.