Optimization for an Uncertainty Reduction Method Applied to Forest Fires Spread Prediction

Abstract

Forest fires prediction represents a great computational and mathematical challenge. The complexity lies both in the definition of mathematical models for describing the physical phenomenon and in the impossibility of measuring in real time all the parameters that determine the fire behaviour. ESSIM (Evolutionary Statistical System with Island Model) is an uncertainty reduction method that uses Statistic, High Performance Computing and Evolutionary Strategies in order to guide the search towards better solutions. ESSIM has been implemented with two different search strategies: the method ESSIM-EA uses Evolutionary Algorithms as optimization engine, whilst ESSIM-DE uses the Differential Evolution algorithm. ESSIM-EA has shown to obtain good quality of predictions, while ESSIM-DE obtains better response times. This article presents an alternative to improve the quality of solutions reached by ESSIM-DE, based on the analysis of the relationship between the evolutionary strategy convergence speed and the population distribution at the beginning of each prediction step.