Title: Recurrent neural network model for reheating furnace based on sequential learning with unscented Kalman filter

Authors: Ying-Xin Liao, Jin-Hua She, Min Wu

Addresses: School of Computer and Information Engineering, Central South University of Forestry and Technology, 498 South Shaoshan Road, Changsha, Hunan, 410004, China. ' School of Computer Science, Tokyo University of Technology, 1404-1 Katakura, Hachioji, Tokyo, 192-0982, Japan. ' School of Information Science and Engineering, Central South University, 154 South Lushan Road, Changsha, Hunan, 410083, China

Abstract: In order to model the dynamics of a walking beam reheating furnace, a multi-input multi-output recurrent neural network (RNN) is constructed based on a sequential learning algorithm. The learning algorithm employs growing and pruning criteria based on the concept of significance of hidden neurons to achieve a compact network. An unscented Kalman filter (UKF) is used to improve the learning accuracy by estimating the parameters of the RNN from incomplete and noisy measurements. Unlike existing methods, this one uses a vector instead of a scalar to denote the width of the allocated neuron so as to precisely represent the probability distributions of different input variables. The effectiveness of the RNN combined with the UKF is compared with that of the RNN with an extended Kalman filter (EKF), and the results show that the former estimates the temperatures of zones of the furnace with a higher precision than the latter.

Keywords: walking beam reheating furnaces; recurrent neural networks; RNN; sequential learning; unscented Kalman filter; UKF; temperature estimation.

DOI: 10.1504/IJAMECHS.2010.033041

International Journal of Advanced Mechatronic Systems, 2010 Vol.2 No.3, pp.164 - 173

Published online: 07 May 2010 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article