Ajuste evolutivo de um modelo de autômatos celulares para simulação de dinâmica de evacuação de pedestres

Abstract

Emergency situations in enclosed spaces require a large number of people to evacuate in a short time, and observing this behavior directly is unfeasible for ethical and cost reasons. Computational simulation fills this gap by allowing such situations to be reproduced and studied, and by supporting the planning of safety measures. Among the available tech- niques, Cellular Automata (CA) stand out for being lightweight and fast, which makes it feasible to dynamically explore many situations and variables. Models of this kind, however, depend on behavioral parameters that can rarely be measured directly, which turns their configuration into a search problem over a large, non-linear space, one for which Genetic Algorithms (GA) are an established alternative. This work proposes a GA built to adjust the parameters of a pedestrian-dynamics CA-based model that combines two models from the literature, one introducing the greedy movement choice and another introducing the concepts of confusion field and signage. The GA’s parameter adjustment is carried out through the analysis of spatial and temporal evacuation metrics: density, dispersion, flow and total time. Due to the absence of real data, the algorithm’s efficiency in reproducing the dynamics observed in data artificially generated with the CA model itself was evaluated, along with the model’s behavior in normative pedestrian-dynamics tests. Additionally, the same evolutionary mechanism was reused to optimize those same metrics through the placement of emergency signs on the already calibrated model. Ex- perimental results showed that the model calibrated by the GA was able to reproduce the reference spatial behavior across the regimes examined, just as sign optimization reduced evacuation time, with the gain depending on the environment’s topology.