Simulação multiagentes em áreas atingidas por terremotos com focos de incêndio.

Abstract

The scenario created by an earthquake presents many challenges for rescue teams, such as landslides, disruptions in the supply of water and energy, and others, may occur. Disruption of gas pipes can cause fires in multiple locations that can spread to neighboring buildings. This paper presents a development proposal for a Multiagent System composed of intelligent agents able to fight multiple fires in an urban environment simulating a post-earthquake situation. The objective of this system is to reduce the overall fire damage at the end of the simulation. To accomplish this task, the environment is explored by dividing the map into sectors, through K-means clustering, and the decision of which fire each agent will combat is based on ANN (Artificial Neural Networks), which evaluates the following inputs: fire intensity, building material and floor area, number of firefighters fighting the fire, distance between the fire and the closest hydrant, distance between the firefighter and the fire, average inflammability coefficient of the building materials of the fireless neighbors of the fire. To evaluate the proposed strategy, 10 simulation pairs were performed, each pair composed of a simulation using firefighters with ANN and another with simple ones. The ANN identified relevant factors for decision making in regards to the fires to be fought by the firefighters, with an effective environment exploration by the clustering method.