Soft Sensors Model Study Based On Improved Neural Network For Wastewater Treatment Parameters

This paper carries out a study on soft sensors model of wastewater treatment process control parameters and water variables ,which enables further control and parameters'on-line monitoring . A improved neural network based soft sensors model is highlighted after the analysis of various soft sensors models. The estimation of some difficultly and expensively measurable parameters such as COD,TN,TP and SVI is gained by this model,too.The result of the soft sensors model preprocessed by Principal Component Analysis(PCA) and Rough Sets(RS) and simulated by Matlab is much better than models without preprocesses. This paper mainly contains: 1.The study of various soft sensors models including the field they can be used in and finally the improved model based neural network. turned out, which is better to implement the measurement,based on the characteristics of wastewater treatment. 2. Detailed analysis of two common neural network models-Back Propagation(BP) neural network and Radial Basis Function(RBF) neural network,which shows their own advantages and disadvantages and the improvements. 3.Because that neural network is easily affected by err and linear datas which are always among the wastewater treatment variables, PCA and RS are used to preprocess the datas.With PCA,it is able to lessen effect of err datars, eliminate datas'linear correlation and reduce dimension of input datas.With RS, it is able to eliminate useless variables and reduce dimension of the datas.They can both simplify neural network structure and get better result of the neural network.Furthermore,RS can reduce the amount of input datas,so there are less instruments used for on-line monitoring. 4.The effectiveness of the model is tested on a municipal wastewater treatment plant in Chongqing.Different methods(BP,PCA+BP,RS+BP,RBF,PCA+RBF and RS+RBF)are presented to set up corresponding models for some difficultly and expensively measurable variables such as COD,TN,TP and SVI.At the end, the best model is chosen out after detailed comparison of the effectiveness of all the models.