Premium
Data‐driven soft‐sensor modelling for air cooler system pH values based on a fast search pruned‐extreme learning machine
Author(s) -
Ye Yisha,
Ren Jia,
Wu Xuehua,
Ou Guofu,
Jin Haozhe
Publication year - 2016
Publication title -
asia‐pacific journal of chemical engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.348
H-Index - 35
eISSN - 1932-2143
pISSN - 1932-2135
DOI - 10.1002/apj.2064
Subject(s) - soft sensor , pruning , extreme learning machine , benchmark (surveying) , computer science , cluster analysis , soft computing , artificial neural network , measure (data warehouse) , data mining , oil refinery , refinery , artificial intelligence , engineering , geodesy , process (computing) , waste management , agronomy , biology , geography , operating system
Failures caused by acid erosion corrosion occur frequently in air cooler systems because of the use of increasingly low‐quality crude oil with high sulfur, acid and chlorine content. The pH value is one of the key parameters used to evaluate the corrosion risk of an air cooler. However, this value is difficult to measure online because of the severe measurement environment, which includes high temperatures, high pressures and corrosion risks. In this paper, a pH soft‐sensor modelling method for an air cooler system in a refinery production processing is developed. Using modelling and simulation in Aspen together with site data, the correlated influence factors are determined first. Then, a soft‐sensor model based on a fast search pruned‐extreme learning machine is proposed in which the pruning problem of hidden‐layer nodes with random weights is solved by adopting the fast search density peak clustering algorithm. The proposed fast search pruned‐extreme learning machine can improve the model's prediction performance by pruning redundant hidden layer nodes with a simpler structure. The feasibility and efficiency of the developed method are demonstrated by the results in the form of benchmark data and real air cooler system data. Copyright © 2016 Curtin University of Technology and John Wiley & Sons, Ltd.