Automated Parameter Estimation and Sensitivity Analysis: Implementation Issues (1st Report)

The present report establishes the computational issues that will be considered for the implementation of hybrid optimization approaches oriented to automated parameter estimation problems. The proposed hybrid optimization approaches are based on the coupling of the Simultaneous Perturbation Stochastic Approximation (SPSA) approach (a global and derivative free optimization method) with two local and derivative dependent optimization methods: (1) a globalized Newton-Krylov Interior-Point algorithm (NKIP) and, (2) the global Levenberg-Marquardt (GLM) method. The first coupling will imply the generation of a metamodel or surrogate model that will allow to incorporate derivative information on a simpler representation of the original problem. The second type of coupling assumes that there is some derivative information available but its utilization is postponed until the SPSA algorithm has made sufficient progress towards the solution. At that point, the GLM takes over the computation to speedup the parameter estimation process. This preliminary description will set the ground for the implementation of a Matlab prototypes on these ideas.