Anomaly detection of organic pollution in drinking water based on fluorescence spectroscopy and stacked autoencoder

In this paper, a three-dimensional fluorescence anomaly detection method for organic pollutants of drinking water is studied, and a normal water sample reconstruction model based on stacked autoencoder is proposed. The organic contamination events are detected by the residual of the reconstructed output and the original input of samples with the 3σ criterion, and the algorithm is validated on the common organic pollutants such as phenol, salicylic acid and rhodamine B in drinking water. The results show that the algorithm can effectively detect both high-concentration and low- concentration organic pollutants under the premise of ensuring low false alarm rate when affected by fluctuations in water quality background. The method proposed is proved to have the potential to improve the sensitivity and robustness of anomaly detection of organic pollution in drinking water.