On training Neural Network To Solve Some Class of Boundary Value Problems

Abstract

The aim of this paper is to train feed forward neural network for solving some class of boundary value problems for ordinary differential equations. Using backpropagation training algorithm, where the Levenberg – Marquardt training algorithm is used to train the network .The existence of the proposed solution was proved. The suggested networks have been studied intensively for a few decades and have provided an option for modeling complex systems. Therefore this option was utilized to reduce the computation of solution, and finally the method is demonstrated through illustrative examples.