Robust calibration of numerical models based on relative regret

Classical methods of calibration usually imply the minimisation an objective function with respect to some control parameters. This measures error between observations and results obtained by a numerical model. In presence uncontrollable additional parameters that we model as random inputs, becomes variable, notions robustness have be introduced for such optimisation problem. this paper, are going present how take into account those uncertainties defining relative-regret. quantity allow us compare value its best performance achievable given realisation By controlling relative-regret using probabilistic constraint, can then define new family estimators, whose inputs adjusted.