Parameter estimation and calibration of simulation models as a non-linear optimization problem

Abstract Parameters' estimation of complex simulation programs is proposed as a non-linear optimization problem. Either the Maximum Likelihood or the Generalized Least Square criterion, subject to linear or non-linear constraints, can be applied. The statistical properties of the estimated parameters are calculated, discussed and fully detailed. A powerful Reduced-Gradient optimization algorithm is applied. The method is developed for general and wide range use. A case study is implemented on a cotton simulation model. Comparing this method to a more traditional estimation method leads to the following major conclusions: (a) a much higher statistical and computational efficiency is achieved; (b) higher dimensional problems can be handled; and (c) ease of use is obtained.