Author(s)

Y. Z. Feng, G. C. Chen, D. Feng & M. Zhuo

Abstract

By analyzing laws of growth and reproduction of microorganisms in the activated sludge (AS) sewage treatment system, this paper proposes a multineural network (NN) COD (chemical oxygen demand) soft-measuring method based on clustering approach for the sewage treatment project. Various reasons which might affect the accuracy of the model are analyzed. Experiments show that the same radiuses of multiple neural network diffusion constant are quite close which means the prediction accuracy is high and the soft-measuring method based on clustering approach is suitable for COD measuring. Keywords: sewage treatment, clustering approach, neural network, microorganisms. 1 Introduction Cybenko [1] has proved theoretically that if there are plenty of training data and without restriction of the size of the network, modelling based on NN can always get a satisfactory model structure. But, in the actual industrial process, people often need to face the limited effective process data and due to the real-time requirements, the network structure also cannot be expanded unlimited. Modeling effect normally relies on the good generalization ability of the network. Paper [2] proposed a robust classification method by combining different models which are based on neural networks with fuzzy combination approach. Paper [3] presented a method to improve model prediction accuracy and robustness by adding different models together. Paper [4] discussed Stacked Neural Network Approach for process modeling. The basic idea of these

Keywords

sewage treatment, clustering approach, neural network, microorganisms