A bidirectional dictionary LASSO regression method for online water quality detection in wastewater treatment plants

Abstract

During the online water quality detection of wastewater treatment plants, the organic ingredients hidden in suspended particles are usually ignored, which significantly degrades the online detection accuracy, especially in high turbidity environments. To tackle this problem, in this paper, a novel online detection method is proposed, which can effectively avoid the physical and chemical interferences caused by suspended particles. First, UV–vis spectroscopy is innovatively utilized in the oxidative digestion process to continuously monitor the substance transformation during the reaction. Then, based on dictionary learning and LASSO regression, a novel machine learning method, bidirectional dictionary LASSO regression (BD-LASSO), is developed. BD-LASSO can bidirectionally implement the dictionary learning from both spectrum and feature aspects, ensuring the information extraction at the feature level and therefore improving the detection accuracy and speed. Based on the experimental results, our method can more effectively eliminate the interference brought by turbidity with a much higher detection accuracy (relative error <10%) and a far shorter detection time (within 5 min).