Title :
A parallel method based on the: hopfield neural network for the solution of the finite elements method equations
Author :
Rekanos, Joannis T. ; Tsiboukis, Theodoros D.
Author_Institution :
Alistotle University of Thessaloniki
Abstract :
The Hopfield neural network is applied to the solution of linear systems of equitions obtainid by the Finite Element Analysis. The selection of the form of the neurons´ activation function is very crucial for the success of the proposed method. A method for the selection of the parameters describing a sigmoid activation function is proposed. Finally the hopfield neural network is applied to an eddy current problem where the linear system of equations is complex.
Keywords :
Character generation; Eddy currents; Electrostatics; Equations; Finite element methods; Hopfield neural networks; Neural networks; Neurons; Optimization methods; Symmetric matrices;
Conference_Titel :
Applied Electromagnetism, 1996. Trans Black Sea Region Symposium on
Conference_Location :
Hellas, Greece
Print_ISBN :
0-7803-3763-8
DOI :
10.1109/AEM.1996.873065
