Intelligent feature extraction for anaerobic/anoxic/aerobic process in municipal wastewater treatment plant

In the municipal wastewater treatment plants (MWWTPs), significant fluctuations in flow rate result in time delays and non-uniform distribution, making it challenging to accurately capture the real dynamics. Feature extraction can extract useful information from complex process data for wastewater treatment process monitoring. However, non-uniform distribution poses a challenge for effective feature extraction. To tackle this issue, a sparse autoencoder feature extraction method (SAFEM) is designed to improve monitoring accuracy. Firstly, the influence of the thrust rate on sample distribution is analyzed in conjunction with MWWTPs inflow fluctuations, and residence time between the two monitoring points is analyzed by hydrodynamic mechanism. Secondly, by constructing a sparse matrix, the distribution of non-zero element is used to reflect the inhomogeneity of the sample distribution, which can effectively capture the high-dimensional sparsity. Thirdly, based on the sparse autoencoder, the loss function is designed for feature extraction, which could better adapt to the change of sample distribution and improve the accuracy of feature extraction in the wastewater treatment process. Finally, the effectiveness of SAFEM is illustrated with experimental studies from a real MWWTP in China. The abnormal condition diagnosis performance of SAFEM shows that SAFEM can extract MWWTPs features accurately.