Research On Key Problems Of Online Water Ouality Anomaly Detection Based On Ultraviolet Spectrum

The safety issue of water quality is related to the national economy strategies and the people's livelihood.Water pollution incidents not only destroy the local water ecosystem,but also seriously affect the security of drinking water in water distribution system.The traditional water quality anomaly detection methods,such as chemical analysis method,is time-consuming and may cause the secondary pollution.Using ultraviolet(UV)absorption spectrum for water quality anomaly detection has the advantages of in-situ measurements,rapid analysis and no reagents.In order to improve the ability of detecting and recognizing the water pollution incidents,the thesis carries out the research on online water quality anomaly detection based on UV absorption spectrum,and discusses on how to overcome the influence of noise,baseline drift,background fluctuation and other disturbances such as sudden change in operational state.The main works and innovation points of this thesis are summarized as follows:(1)Considering the noise interference and baseline drift,a method of water quality anomaly detection based on the UV full spectrum was proposed.First of all,it uses Savitzky-Golay(S-G)convolution smooth method for UV spectrum filtering,mean centering to remove error points and spectral scattering,and asymmetric least squares method for spectral baseline correction.Secondly,the principal component analysis(PCA)is applied to reduce the spectral dimension and extract feature.A set of normal water spectrum is trained to build the PCA model after being optimized by eliminating the error samples circularly,and the outliers(sequence)are detected by the Q statistics in the residual subspace.Finally,the probability of water quality outliers is determined by analyzing the characteristics of the outliers in the time sequence.The performance of the proposed method was evaluated with data from contaminant injection experiments.The results showed that the method is effective to eliminate the noise and baseline drift by comparing the effect of different pretreatment methods.(2)The anomaly detection method is optimized in order to solve the influence of long-term water quality background fluctuation.Firstly,this thesis analyzes the reason for the fluctuation of water quality background,discusses the change of the raw water quality and the variation of water delivery flow in the water distribution system for the effect of the slowly-varying fluctuation characteristics of the UV spectrum data.Then the sliding window PCA method based on this characteristics is proposed.The proposed method was analyzed and verified by using the data from contaminants injection experiments.Comparing with the conventional PCA method,the sliding window PCA method can adapt to the slow changing of water background in long-term monitoring,and optimize the "best" window length continually according to the results of different window length under objective function.(3)Aiming at the problem of abnormal water quality caused by sudden change in operational state,the classification of different water anomalies based on UV spectrum is studied.The technical scheme of classification is studied by characteristics analysis of the UV spectrum data in the sudden change in operational state.An water quality knowledge database is set up to match the water anomaly.Considering the large size and high correlation between wavelengths of UV spectrum,the thesis adopts PCA and Fisher linear discriminant analysis method for feature extraction and classification to distinguish the water pollution events and the anomaly caused by sudden change in operational state.Focusing on various types of interference in the process of water quality anomaly detection based on the UV absorption spectrum,the anomaly detection method is studied and improved.Based on the results of this thesis,the software of water quality early warning system is developed and applied.