1. Solving Emden–Fowler Equations Using Improved Extreme Learning Machine Algorithm Based on Block Legendre Basis Neural Network.
- Author
-
Yang, Yunlei, Wu, Yang, Hou, Muzhou, Luo, Jianshu, and Xie, Xiaoliang
- Subjects
MACHINE learning ,NONLINEAR differential equations ,INITIAL value problems ,ALGEBRAIC equations ,WEIGHT training ,ORDINARY differential equations - Abstract
In this paper, a numerical method to solve second order nonlinear ordinary differential equations with singular initial value problems was proposed. Here, block Legendre basis neural network (B-LBNN) was developed to create approximate solution and its derivatives. We use Legendre polynomials to expand the input pattern and eliminate hidden layer. By constructing block subnetwork and transforming variables, the problem of Emden–Fowler type equations was converted to solving the algebraic equation systems. The improved extreme learning machine algorithm was used to train network weights. A series of homogeneous or nonhomogeneous Emden–fowler equations were test to validate the accuracy and efficiency of B-LBNN model. Experimental results showed that the proposed algorithm provides a new tool for this equations and can outperform other approaches in literature in terms of accuracy. [ABSTRACT FROM AUTHOR]
- Published
- 2023
- Full Text
- View/download PDF