Search

To overcome the strong nonlinearity and dynamic characteristics, the online fault detection was investigated based on principal component analysis ( PCA) in a wastewater treatment process. The kernel principal component analysis ( KPCA) was constructed by introducing kernel functions in PCA, and the dynamic kernel principal component analysis ( DKPCA) was proposed by embedding dynamic models to achieve online fault detection in the wastewater treatment process. According to the data from the treatment process of a paper mail wastewater, the bias faults, the drift faults and the precision degradation faults were established and simulated. The results show that under the condition of bias faults, compared with PCA and KPCA, the squared prediction error of D KPCA is respectively improved by 96. 96% and 87. 87%, and the detection sensitivity in drift fault is also improved. The validity of the DKPCA method is verified in online fault detection of wastewater treatment. [ABSTRACT FROM AUTHOR]

The integrating method of kernel partial least squares with Gaussian process regression (KPLS-GPR) was used to predict the effluent chemical oxygen demand (COD) and effluent suspended solids (SS) in papermaking wastewater treatment processes (WWTP). The latent variables from KPLS were used to handle the high dimensionality and the complex collinearity problem of WWTP data. The GPR model was used to develop the regression between the latent variables and the output variables. Based on the WWTP data set from a paper mill, the simulation experiment was conducted. The soft sensors of artificial neural network (ANN), PLS-ANN and KPLS-ANN were proposed for the comparison. The results show that the latent variables of KPLS can be used to improve the prediction results of conventional models obviously, and KPLS-GPR achieves the best prediction performance. For the prediction of effluent COD and effluent SS, the values of determination coefficient for KPLS-GPR are 0.575 and 0.610, respectively, which are respectively improved by 36.90% and 43.87% in comparison with those of the conventional counterparts. [ABSTRACT FROM AUTHOR]