Design and application of soft sensors in rural sewage treatment facilities

Abstract

Recently, with the growing demand for water quality monitoring, soft measurement sensors have drawn public attention, which can overcome the drawbacks of high cost and long time needed in traditional measurement methods. In this study, a machine learning-based soft monitoring sensor was developed to simultaneously monitor four water quality indicators including COD, NH4+-N, NO3--N, PO43--P. Firstly, specialized experimental equipment and calibration methods were developed to generate a matching dataset that collected over 94,000 data points. Secondly, five models including Multiple Linear Regression, Ridge Regression, AdaBoost, Decision Tree Regression, and Bagging Regression were constructed and compared. The learning accuracy of the models ranged from 0.8860 to 0.9999, among which the predicted value of Bagging Regression is highly fit to the true value. Subsequently, the fuzzy grade method was adopted to reduce the prediction error and strike a balance between efficiency and accuracy. Finally, the designed soft sensor was used for real-time monitoring at three monitoring points in Changzhou, China from September to October 2020, and the results proved the feasibility of the soft sensor in practical application. This study provided a fast and accurate method for water quality measurement, which is of great significance for the management of rural sewage treatment facilities. HIGHLIGHTS The soft sensor based on machine learning is applied to the water quality monitoring of rural sewage treatment facilities.; Designing of laboratory-level devices to obtain datasets for model training.; Fuzzy classification method is introduced to analyze and process data to reduce errors and obtain comprehensive water quality evaluation results.;

Keywords

• bagging regression
• fuzzy grading
• machine learning
• process control
• water quality monitoring