Multiple source tracking and identifications in urban regions with unstable wind flows: Particle swarm optimization methodologies and their benchmark solutions.

The prompt and accurate identification of transient pollution sources in urban regions is imperative for effective pollutant control and the mitigation to public health risks. Depending on sparse concentration distributions, multiple pollution sources identification is inherently complex and difficult, particularly when the pollutant sources exhibit time-varying release rates and are located in urban environments subject to dynamic wind flows. In this research, two niche-based Particle Swarm Optimization (PSO) algorithms, Niche-DPSO (niche pso for dynamic systems) and RSNM-DPSO (ranged subgroup pso with nelder mead for dynamic systems), were simultaneously proposed, aiming to localize multiple pollution sources under the background of unsteady wind flow fields. The MUST experiment, characterized by its complex layout replete with densely arranged obstacles, was adopted as the application scenario. Computational Fluid Dynamics (CFD) was employed to simulate the release of pollutant concentration at (−25.8 m, 0 m, 1.6 m) and (24.2 m, 41.6 m, 1.6 m), with flow velocities ranges from 1.8 to 7.6 m/s. Representative performance-influencing factors, including niche radius, population size, sensor response time, and sensor errors, were systematically evaluated. The results suggested that both Niche-DPSO and RSNM-DPSO algorithms were generally effective in positioning multiple pollution sources. Comparative analysis revealed that Niche-DPSO consistently outperformed RSNM-DPSO in terms of localization success rates, achieving a peak performance of approximately 98 %. Furthermore, Niche-DPSO could maintain its robust performance even under time-spatial-varying environmental and operational conditions, albeit at the expense of requiring more search steps and longer CPU processing time. • Two specialized algorithms were proposed for locating multiple pollution sources. • A CFD model was established and verified for application in the MUST experiment. • The algorithms' performance was rigorously evaluated in the simulation scenario. • Key performance factors were identified, and a sensitivity analysis was conducted. [ABSTRACT FROM AUTHOR]