Water Quality Anomaly Detection: An Improved Empirical Wavelet Transform Method

Water quality anomaly detection, as an important part of water environmental protection system, can be divided into two parts: feature extraction and anomaly detection. The current detection methods suffer from the problems of over-decomposition and incomplete denoising in the feature extraction stage as well as low detection accuracy in anomaly detection. Hence, this study proposes an improved Empirical Wavelet Transform (EWT) method to accomplish the water quality anomaly detection. (1) The EWT spectral segmentation is optimized with the mutual information method for the over-decomposition of the water quality monitoring data. (2) An adaptive threshold function is constructed by using the multi-scale fuzzy entropy method in order to enable the independent denoising of the various components, which were decomposed by the optimized EWT, and the ARIMA-Isolation Forest anomaly detection framework is also given. Finally, the proposed method is analyzed with the Songhua River basin site water quality monitoring data set. The results show that the proposed method could improve SNR by more than three times, reduce MSE by 89%, RMSE by 67%, and MAE by 76% in feature extraction for water quality anomaly monitoring, which indicates that the method can effectively reduce the noise interference and thus improve the accuracy of anomaly detection.