Urban swarms: Swarm intelligence for emergency evacuation planning

Abstract

The relevance of this research is determined by the fact that in the conditions of modern urbanization, the safety of large public facilities increasingly depends on the ability of management systems to quickly and evenly distribute human flows during emergencies. Standard algorithms are focused on thel shortest paths and tend to neglect exit throughput capacity, causing congestion, which in turn critically increases evacuation time. The object of research is the process of emergency evacuation in large-scale, multi-exit buildings, modeled as a capacity-constrained network flow problem. The aim of the article is to  develop and validate a swarm intelligence model that allows for the effective distribution of human flows between the exits, while taking into account their throughput capacity and reducing the total evacuation time. To achieve this goal, the following tasks were solved: a nonlinear objective function was constructed with penalties for congestion; the parameters of swarm algorithms were adapted for searching in congested networks; an empirical assessment of evacuation scenarios was carried out and a compromise between computational speed and stability of solutions was determined. The following methods were used in the work: graph theory methods to describe the structure of buildings; mathematical programming apparatus to construct a nonlinear objective function; and swarm intelligence methods to search for the global optimum in a multidimensional solution space. The scientific novelty lies in the proposal of a novel objective function that includes strong nonlinear, quadratic penalties for overloading and exit capacities violations. To effectively optimize this high-penalty environment, the basic parameters of the standard particle swarm optimization and ant colony optimization algorithms were changed, resulting in the Modified Particle Swarm Optimization and Enhanced Ant Colony Optimization variants. The simulation results for different types of buildings confirmed that the proposed approaches reduce cumulative evacuation delays by up to fifty-five percent compared to classical methods of finding the shortest path. The practical significance of the results lies in supplementing existing engineering methods for evacuation planning – which are based on the shortest-path principle – with a load-balancing mechanism that accounts for exit capacities. This eliminates the issues of localized congestion and the overloading of specific exits during crowd evacuations, while also providing an algorithmic foundation for integrating the developed models into BIM systems to enable the automatic generation of safe routes in real time. Conclusions. The analysis demonstrates that the proposed swarm intelligence model effectively distributes crowd flows among exits, taking their throughput capacities into account, and significantly reduces cumulative evacuation delays compared to the baseline shortest-path method; this confirms the validity of the developed approach for practical evacuation planning.