Research On Water Quality Multi-scale Feature Extraction And Anomalv Detection Using Time-frequencv Analysis

Safety of water quality is closely related to people’s livelihood, which attracts high attention around the world. Water pollution events usually happen unexpectedly and suddenly, and are prone to threaten people’s health within a short time. Therefore, precise detection and warning of potential water quality abnormity is an essential prerequisite to guarantee people’s health and quality of life. And it is an important issue of common concerns.Considering the volatility properties of drinking water in supply networks embedded in the time series of the parameters of water quality, the time-frequency domain based water quality abnormity detection technology is investigated in this thesis. This paper mainly focuses on extracting the features of water quality index under different scales and realizing feature recognition based on threshold method and energy spectrum analysis method to implement water quality anomaly detection. The contributions of this work are listed as follows:(1) Aimed to utilize time-frequency information hidden within time series data, the wavelet analysis method is introduced to preprocess water quality data and detect the outliers. The anomaly events could be detected with single or multiple indicators of water quality. The detection performance is evaluated by Receiver Operation Characteristics Curve (ROC). In the process, the influence of baseline drift and outliers should be firstly eliminated, and the time series of single and multiple indicators is decomposed with wavelet packet analysis. Then, the detection results could be acquired according to strength distribution of water quality signals in various frequency bands. And finally, the viability of the method is approved by combining the simulation with software-package of Canary, which has been developed by USEPA.(2) After data preprocessing based on wavelet analysis, the water quality time series analysis is studied here based on wavelet packet energy spectrum, in order to fully utilize the energy characteristics of water quality signals within different frequency bands. The energy spectrums of the water quality monitoring data and background data in different time-frequency bands are compared so as to detect the abnormality in water quality signals comprehensively. Algorithm characteristic is discussed and evaluated using multiple parameters of water quality monitoring data.(3) Water quality anomaly detection method based on periodic feature extraction is proposed for periodic characterization of water quality signals in this paper. Periodic pattern analysis for water quality signals is carried out using Fourier spectrum analysis. Additionally, empirical mode decomposition method is used to extract the periodic characteristic components. Water quality anomaly detection is accomplished by comparing the normal pattern and periodic pattern. Finally, the performance of the proposed method is compared with the time series increasing method and the linear prediction method.As mentioned above, the water quality time series are analyzed using the features in frequency domain and time domain, and the energy spectrums for different frequency bands. The anomaly detection methods based on the analysis are proposed and tested, which are proved to be effective, and the study presented provides technical accumulation for water quality anomaly detection.