EIA 2004 |
Society for Environmental Information Sciences
Environmental Informatics Archives
ISSN 1811-0231 /
ISEIS Publication Series Number P002
2004 ISEIS. All
Paper EIA04-064, Volume 2
(2004), Pages 641-652
Short Term Prediction in Nitrogen Removal Processes Using Least Squares Support Vector Machine with NARX Model
Y. H. Yang1*, A. A. Guergachi2 and G. N. Khan1
1. Electrical and Computer Engineering, Ryerson University, Toronto, Ontario, M5B 2K3 Canada. *Corresponding author: firstname.lastname@example.org.
2. School of Information Technology Management, Faculty of Business, Ryerson University, Toronto, ON, M5B 2K3 Canada.
In order to meet the more stringent environmental regulations, it is necessary to investigate the adaptive and optimal control strategies for the biological wastewater treatment processes. Nitrogen removal is one of the essential concerns in wastewater treatment. Because of the complicated activities of microbial metabolism involved, nitrogen removal is a nonlinear, dynamic, and time variant complex process. The mechanistic models for nitrogen removal are complicated and still uncertain to some extent. A new machine learning approach, Support Vector Machine (SVM) was proposed as black-box modeling technique to be used to model the biological wastewater treatment processes. LS-SVM, a simplified formulation of SVM, was applied in this study to predict the concentration of nitrate & nitrite (NO) in the Mixed Liquor (ML) of wastewater treatment plant. Nonlinear AutoRegressive model with Exogenous inputs (NARX model) can be employed with LS-SVM to extract useful information and improve the prediction performance. In this paper, the premium wastewater treatment plant simulation and optimization software, GPS-X, was used to create virtual plant layout and simulated data.
Keywords: Wastewater treatment, Nitrogen removal, NARX model, Support Vector Machines, Least-Square SVMs, GPS-X
Full Text (PDF)