Online Detecting Water Quality Anomaly from UV/Vis Spectra Using Baseline Correction and Principal Component Analysis Method

In recent years, water quality security issue has aroused widespread social concerns. Ultraviolet and visible absorption spectrum for real-time monitoring of water quality has the advantages of in-situ detection as well as no need for reagents and rapid analysis, which makes it suitable for online detection. However, the ultraviolet and visible spectra are of large size and easily interfered by instrument and the normal water quality fluctuation, which affect the water quality anomaly detection result. In this paper, the baseline correction and principal component analysis method for UV/Vis spectra is proposed to detect abnormal water quality. The asymmetric least squares algorithm is used to correct the baseline and the principal component analysis for UV/Vis spectral matrix is adopted to reduce dimensions and extract features. Afterwards the outliers in the test samples are evaluated according to the Q statistic in the residual subspace. Finally, anomalies warning is updated by calculating the cumulative probability. The performance of the proposed method is evaluated by using data from phenol injection experiments. Results show that the proposed method effectively improves the detection limit of pollutants and has a higher detection rate and lower false alarm rate compared with the result without baseline correction.