ESSIM-EA applied to Wildfire Prediction using Heterogeneous Configuration for Evolutionary Parameters

Abstract

Wildfires devastate thousands forests acres every year around the world. Fire behavior prediction is a useful tool to cooperate in the coordination, mitigation and management of available resources to fight against this type of contingencies. However, the prediction of this phenomenon is usually a difficult task due to the uncertainty in the prediction process. Therefore, several methods of uncertainty reduction have been developed, such as the Evolutionary Statistical System with Island Models based on Evolutionary Algorithms (ESSIM-EA). ESSIMEA focuses its operation on an Evolutionary Parallel Algorithm based on islands, in which the same configuration of evolutionary parameters is used. In this work we present an extension of the ESSIM-EA that allows each island to select an independent configuration of evolutionary parameters. The heterogeneous configuration proposed, at the island level, with the original methodology in three cases of controlled fires has been contrasted. The results show that the proposed ESSIM-EA extension allows to improve the quality of prediction and to reduce processing times.