Dynamic KPCA for Feature Extraction of Wastewater Treatment Process

Abstract

The feature extraction method is an effective tool to understand the behavior of plug-flow wastewater treatment process (PF-WWTP). However, it is a challenge to extract feature components due to PF-WWTP subjected to the time-varying system with dataset mismatch. To solve this problem, in this paper, an adaptive feature extraction method (AFEM) based on dynamic kernel principal component analysis (KPCA) is proposed to improve the feature extraction accuracy. First, a data adjustment method is proposed to adapt datasets of process variables to the different hydraulic residence time. Then, the matching datasets can be used to observe the dynamics of metabolism within PF-WWTP. Second, a dynamic KPCA algorithm based on iterative calculation is introduced to obtain the contribution of feature components for process variables. This algorithm can update the order of feature components online following with the time-varying flow-rates of PF-WWTP. Third, an error-oriented self-adaptive mechanism is designed to determine the dimension of feature components for process variables. This mechanism not only performs preferable feature extraction without giving thresholds but also ensures its realtime accuracy. Finally, AFEM is compared with some existing feature extraction methods through experiments. The results show that the proposed AFEM can accurately extract feature components for PF-WWTP.