Detection Of Water Quality Contamination Events With Uv/vis Spectrum Based On Supervised Learning

Recently reported pollution events concerned with water quality has seriously affected the lives and property of people.The use of Ultraviolet(UV/VIS)spectra,for the in-situ detection of water-quality anomalies,has the several advantages:it is free of reagent,has a low cost and facilitates rapid analysis.For these reasons,it is one of the more popular research trends for urban distribution water systems.Because of the disadvantages of high detection limit and alarm delaying in the present UV/VIS anomaly detection,we make a study of water quality event detection with UV/VIS Spectra based on supervised learning.The main work and innovation points of the thesis are as follows:(1)In order to decrease the detection limit of present unsupervised methods,we use the full spectrum data of specific contaminations and normal water samples to build the supervised model.Partial Least Square Discriminant Analysis(PLS-DA)method is used to extract the feature,and Sequential Bayesian Method is used to find the anomaly sequence and trigger the alarm,pretreatments handle the correction of water background baseline based on orthogonal projections.The experiment results showed that the proposed method is able to decrease the detection limit of specific contamination and make use of the prior spectrum information better than the present method.(2)In order to adapt the long-term fluctuation of spectral baseline for the method,the adaptation of event detection based on supervised learning is studied.Dynamic correction method,updating baseline and adjusting model,based on orthogonal projection is proposed.The experiment results showed that the method can decrease the false alarm efficiently under the condition of the break of the baseline or trend variation.(3)In order to solve the interference of other contaminations,the specificity of the detection is studied based on supervised learning.The index of Variable Importance in Projection(VIP)is used to evaluate quantitatively the similarity of any two different contamination model.The experiment results showed that low similarity among different models of contaminations implies specific detection is effective.By studying the adaptability and specificity of the method and making use of the prior pollution information,the proposed method shows better performance in the water quality event detection.The technique platform of contaminations intrusion events detecting system based on the method was developed.