A Procedure of Global Optimization and its Application to Estimate Parameters in Interaction Spatial Models

¹ Universidade do Vale do Itaja – UNIVALI, Rodovia SC 407, Km 04, CEP 88122-000, São José/SC, Brazil
² Universidade Federal de Santa Catarina – UFSC, Depto de Engenharia de Produção e Sistemas, 88010-970, Florianópolis/SC, Brazil
³ LMR – INSA – Rouen, Avenue de l'Université BP 8, Saint-Etienne du Rouvray, FR-76801, France

Corresponding author: eduardo.souza@insa-rouen.fr

Received: 15 November 2009
Accepted: 30 March 2010

Abstract

This work concerns the calibration of trip distribution models issued from Transportation Planning. The optimal estimators are the global minima of a non-convex function. A mathematical result establishes that the solution is the limit of a sequence of means of convenient random variables. A population-based global optimization involving this representation and random perturbations of a local descent method is applied. Experiments are performed on classical test functions. In the application to the models of trip distribution, 4 sets of data, including real data are used. The results show that the use of the representation of the solution leads to a significant improvement and increases the robustness.