Parameter Estimation in a Three-dimensional Wind Field Model Using Genetic Algorithms ⋆ Eduardo Rodr´ ıguez, Gustavo Montero, Rafael Montenegro, Jos´ e Mar´ ıa Escobar, and Jos´ e Mar´ ıa Gonz´ alez-Yuste University Institute of Intelligent Systems and Numerical Applications in Engineering, University of Las Palmas de Gran Canaria, Edificio Instituto Polivalente, Campus Universitario de Tafira, 35017 Las Palmas de Gran Canaria, Spain barrera@dma.ulpgc.es, {gustavo, rafa}@dma.ulpgc.es, escobar@cic.teleco.ulpgc.es, josem@sinf.ulpgc.es Abstract. The efficiency of a mass consistent model for wind field ad- justment depends on several parameters that arise in various stages of the process. First, those involved in the construction of the initial wind field using horizontal interpolation and vertical extrapolation of the wind measures registered at meteorological stations. On the other hand, the Gauss precision moduli which allow from a strictly horizontal wind ad- justment to a pure vertical one. In general, the values of all of these parameters are based on empirical laws. The main goal of this work is the estimation of these parameters using genetic algorithms, such that the wind velocities observed at the measurement station are regenerated as much as possible by the model. 1 Introduction Mass consistent models are diagnostic models for constructing wind velocity fields from a few experimental measurements. In general, these models are de- fined by the physical laws of an incompressible fluid, by the empirical design of the wind profiles and by the values of velocities measured at the stations. This explains the existence of many parameters in the model. Some of them are clearly bounded and defined, while others are still under discussion and in- terpretation. Our work deals with the latter ones. There are many methods for the resolution of inverse problems involving parameter estimation and they have been largely studied in the literature. Among them, we have chosen a robust and flexible tool: genetic algorithms, which allow to solve linear and non-linear multiparameter optimisation problems. This work has been structured as follows. First, the wind model is summarised in Sect. 2. We remark the studied parameters in Sect. 3. Next, in Sect. 4, the fitness function is established and genetic algorithms are briefly introduced, with their properties and possibilities used in this work. Numerical experiments are shown in Sect. 5 and, finally, conclusions are presented in Sect. 6. ⋆ Partially supported by MCYT, Spain. Grant contract: REN2001-0925-C03-02/CLI P.M.A. Sloot et al. (Eds.): ICCS 2002, LNCS 2329, pp. 950-959, 2002.  Springer-Verlag Berlin Heidelberg 2002