Particle swarm optimization for source localization in realistic complex urban environments

Abstract

In this work, we present a method to localize a source in complex urban environments using particle swarm optimization (PSO). Instead of using PSO to minimize the difference between a plume model and measurements as is often done, PSO is run such that each particle is modeled by an unmanned aerial vehicle (UAV) that measures and directly finds the global maximum of the concentration field. Several modifications were made to PSO to allow it to perform successfully in this application. The synthetic data used to test PSO were produced using the 3D building resolving Quick Urban & Industrial Complex Dispersion Modeling System (QUIC), and PSO was implemented in Python. Three different domains were tested: (1) a case with no obstacles, (2) a case with four large obstacles, and (3) a real-world case modeled after the Joint Urban 2003 experiment in Oklahoma City. We found that PSO works well in idealized and real cases. In the Oklahoma City simulation, approximately 90% of the PSO runs with 10 particles make it to within 1% of the maximum domain distance to the source, and approximately 98% of the PSO runs with 50 particles make it to within 1% of the maximum domain distance to the source. However, PSO is not completely immune to local maxima, and there is the possibility of convergence to the wrong point in the domain. The insight from this study can be used to inform first responders or create a tool that can be implemented on UAVs to locate a contaminant source.