User

Password

Recupero Password

Path: Home » Indice Pubblicazioni » Convegni ATI - Accesso riservato soci CTI » CA - 61 - Perugia 2006 » Sequential genetic/gradient progressi...

Convegni ATI - Accesso riservato soci CTI

Sequential genetic/gradient progressive optimization. Part I

Pubblicazione


Autore: L.A. Catalano, A. Dadone, V.S.E. Daloiso, R. Amirante

Collana: CA - 61 - Perugia 2006

Note:
This paper proposes a sequential (genetic/gradient) progressive optimization strategy, which is capable of optimizing a multimodal objective function derived from the solution of discretized governing equations, in a computational time comparable to that required by a number of gradient optimizations equal to the number of local optima found. The proposed method performs an initial global search on coarse grids and with partially converged solutions, to identify the basins of attraction of the better optima. A multimodal test function derived from the solution of a partial differential equation, is used in the first part of the paper to demonstrate that a genetic algorithm with standard operators is simply needed, and that its global-search capability is not inhibited by the use of coarse grids and of partially converged solutions. The same test is also used to develop a technique for switching from the genetic to the gradient optimization and to select one or more initial designs for one or more final applications of a progressive gradient technique. In this second phase, the solution accuracy is progressively improved while approaching the optimum, by both refining the grid and increasing the solution convergence. Since the initial global search is based on rather inaccurate design solutions and is stopped after few generations, a multistart gradient optimization is preferred in the proposed numerical computations, to assure the detection of the global optimum among the better local optima.


Per scaricare il Documento devi essere loggato!



Viale Elvezia, 12 - 20154 Milano
Tel. +39 02 266.265.1
Fax +39 02 266.265.50
P.IVA 11494010157

Social CTI

Seguici!

Twitter LinkedIn

Disclaimer - Privacy