Support vector machines for environmental informatics : application to biological nitrogen removal in wastewater treatment plants

Abstract

In order to meet the more stringent environmental regulations, the adaptive and optimal control strategies should be investigated for the biological nitrogen removal (BNR) processes in wastewater treatment plants. Because of the complex nature of the microbial metabolism involved, the conventional mechanistic models for nitrogen removal are difficult to formulate and the existing ones are still uncertain to some extent. Alternatively, the machine learning methods have been investigated as black-box modelling techniques. A new approach, Support Vector Machine (SVM) was proposed to be used to model the biological nitrogen removal processes in this thesis. Specifically, LS-SVM, a simplified formulation of SVM, was applied to predict the concentration of nitrate & nitrite (NO). The simulation results indicate that the proposed method has better generalization performance in comparison with generalized regression neural network, especially under weather conditions that are quite different from the training weather conditions.

Keywords

Identifiers

DOI

10.32920/ryerson.14646114.v1

Journal

Unknown journal

Year

2021