Precision Source Parameter Inversion for Typical Air Pollutant Emissions at Microscale: An Integrated PSO-NM Algorithm and Gaussian Dispersion Model Approach
Accurate identification of pollutant emission source parameters is critical for effective pollution response. This study evaluates the performance of genetic algorithm (GA), Nelder-Mead simplex (NM), particle swarm optimization (PSO), and their coupled variants on multi-dimensional, multi-extremum benchmark functions, and develops a source parameter inversion technique integrating PSO-NM with a Gaussian dispersion model. Validation via sulfur hexafluoride (SF6) single-point and multi-point release experiments demonstrates that PSO-NM achieves mean values closest to theoretical optima on Shubert, Hartmann, and Shekel functions, with superior stability and precision. In single-point source experiments, the relative deviation of source strength (Q) inversion ranges from -27.1% to 38.5%, with positional errors below 10 m, indicating robust convergence and repeatability. Multi-point source inversion exhibits stability across two scenarios but with reduced accuracy compared to single-point cases. When source strength is unknown, inversion accuracy for low-release sources (relative deviation 37.3%-70.4%) surpasses that for high-release sources; when position is unknown, positional deviations generally remain below 50 m, with low-release sources yielding better x0 deviations (-1.6 to 8.2 m) but slightly worse y0, z0, and distance parameters. Inversion errors primarily stem from meteorological non-stationarity, inter-source interference, algorithmic local optima, low-concentration measurement noise, and model assumptions. Future improvements may incorporate real-time meteorological correction and source-specific constraints to enhance accuracy and robustness in complex scenarios. The findings provide technical support for precise source tracing, monitoring, and refined management of pollutant emissions at microscale in industrial parks and enterprises.