My master thesis presentation 2012:
The agent-based approach focuses on the modeling and the simulation of complex systems.
A crucial phase of the development process of multi-agent simulation is probably the validation, which is the process of determining the degree to which a model or simulation is an accurate representation of the real-world.
One main issue of the validation phase is the calibration, which is an instantiation phase of the local and global parameters of the simulation. The complexity of calibration is due to the risk of the combinatory explosion, the nonlinear relationship between the parameters, and sometimes the lack of information about the simulated system. Nevertheless, it is an important step given that the global dynamics of the simulated model is governed by parameters to be calibrated.
The approach we suggest is to consider the calibration problem as an optimization problem. We apply directly an optimization algorithm, the genetic algorithm, the model is seen as a black box whose inputs are the values of the parameters to be calibrated and the output is the value of an objective function evaluated after the run of the simulation. The output value represents the degree of plausibility, a measure of the simulation’s validity (accuracy, completeness, quantification of an emerging phenomenon ...) or a measure of distance between the simulation’s results and those obtained by observing the real system.