Author(s):
Pedro Valente 
; António Pereira ; Luis Paulo Reis
Date: 2008
Persistent ID: http://hdl.handle.net/10216/15882
Origin: Repositório Aberto da Universidade do Porto
Subject(s): Ciências tecnológicas; Tecnologia de agentes; Tecnologia do conhecimento; Tecnologia
Description
This paper presents an approach to the calibration of ecological models, using intelligent agents with learning skills and optimization techniques. Model calibration, in complex ecological simulations is tipically performed by comparing observed with predicted data and it reveals as a key phase in the modeling process. It is an interactive process, because after each simulation, the agent acquires more information about variables inter-relations and can predict the importance of parameters into variables results. Agents may be seen, in this context, as self-learning tools that simulate the learning process of the modeler about the simulated system. As in common Metaheuristics, this self-learning process, initially involves analyzing the problem and verifying its inter-relationships. The next stage is the learning process to improve this knowledge using optimization algorithms like Hill-Climbing, Simulated Annealing and Genetic Algorithms. The process ends, when convergence criteria are obtained and thus, a suitable calibration is achieved. Simple experiments have been performed to validate the approach This paper presents an approach to the calibration of ecological models, using intelligent agents with learning skills and optimization techniques. Model calibration, in complex ecological simulations is tipically performed by comparing observed with predicted data and it reveals as a key phase in the modeling process. It is an interactive process, because after each simulation, the agent acquires more information about variables inter-relations and can predict the importance of parameters into variables results. Agents may be seen, in this context, as self-learning tools that simulate the learning process of the modeler about the simulated system. As in common Metaheuristics, this self-learning process, initially involves analyzing the problem and verifying its inter-relationships. The next stage is the learning process to improve this knowledge using optimization algorithms like Hill-Climbing, Simulated Annealing and Genetic Algorithms. The process ends, when convergence criteria are obtained and thus, a suitable calibration is achieved. Simple experiments have been performed to validate the approach
