A fast predicting neural fuzzy model for on-line estimation of nutrient dynamics in an anoxic/oxic process

In this paper a software sensor based on a fuzzy neural network approach was proposed for real-time estimation of nutrient concentrations. In order to improve the network performance, fuzzy subtractive clustering was used to identify model architecture, extract and optimize fuzzy rule of the model. A split network structure was applied separately for anaerobic and aerobic conditions was employed with dynamic modeling methods such as autoregressive with exogenous inputs and multi-way principal component analysis (MPCA). The proposed methodology was applied to a bench-scale anoxic/oxic process for biological nitrogen removal. The simulative results indicate that the learning ability and generalization of the model performed well and also worked well for normal batch operations corresponding to three data points inside the confidence limit determined by MPCA. Real-time estimation of NO(3)(-), NH(4)(+) and PO(4)(3-) concentration based on fuzzy neural network analysis were successfully carried out with the simple on-line information regarding the anoxic/oxic system.