Research On Water Quality Online Anomaly Detection Techniques Based On Ultraviolet-Visible Spectra In Water Distribution Systems

Safety of municipal water distribution network should be closely attended by the public since it provides drinking water for almost every citizen in the country. Traditional water quality anomaly detection involves a series of process that consumes plenty of time and could possibly cause secondary pollution as well. Ultraviolet-visible spectra analysis has such merits as convenient operation, no chemical reagent adding, excellent repeatability and short detection period, etc., which makes it suitable for online and large scale detection. Therefore, the research on UV-Vis spectra application in network water anomaly detection was undertook in this work.This work mainly includes:1.This work undertook researches on anomaly detection based on specific spectral absorption peaks, with the consideration of circumstances that the contaminant was identified beforehand. In this part, the specific spectral absorption peaks were regarded as the indicator for anomaly detection, which was combined with the analysis of time-series prediction algorithms. Potassium biphthalate and potassium ferricyanide were selected for online water quality anomaly detection test, and the spectral data was input into the model. It achieved satisfying detection results.2. Considering the anonymous contaminant situations, however, a spectral data dimension reduction based approach was proposed. It utilized principal component analysis and clustering analysis to form an anomaly detection model. The model reduced the data dimension of the online collected spectra and extracted features. And clustering analysis determined the water quality situation accordingly afterwards. A proper experiment was designed and the model was tested on basis of that, which proved the effectiveness of this model.3. The approach above has some limitations when applied to low density contamination. Therefore, statistical model was introduced for a new approach, which is based on spectra statistical feature extraction. It utilized probabilistic principal component analysis for spectral data dimension reduction and online multivariate control chart for real time analysis, which then determined water anomaly. The experiment data was analysed and parameters were determined through analysis as well. Finally, satisfying detection results were achieved.Overall, this work utilised ultra-violet and visible spectrum for water quality detection in municipal distribution network. It discussed the issue from various perspectives and formed the optimised broad spectrum detection approach, with the combination of traditional spectral analysis. Besides, the theories were verified through the experiments on the tiny distribution network platform. It verified the usability and reliability of the online water quality detection model, and accumulated sufficient data in support of further analysis as well.