A Machine Learning Framework for Enhanced Assessment of Sewer System Operation under Data Constraints and Skewed Distributions

Abstract

In the realm of sewer management, precise machine learning simulations of physicobiochemical processes during sewage transport are essential yet are hindered by skewed distributions and data constraints. To address this issue, the present study introduces an innovative algorithm, the Automatic Synthetic Minority Over-Sampling Technique for Regression with Gaussian Noise (AutoSMOGN), designed to mitigate the adverse effects of skewed data set distributions. The findings reveal that the integration of the AutoSMOGN algorithm with ML models significantly enhances the precision of gaseous H₂S concentration predictions. Of these approaches, ensemble learning models demonstrated superior accuracy in forecasting gaseous H₂S concentrations within sewer environments, achieving the highest coefficient of determination (R²) of 0.80. Furthermore, the study validates the effectiveness of the AutoSMOGN algorithm in addressing skewed distribution, as evidenced by its acceptable predictive performance on a full-sequence data set (R² of 0.52) and when applied to multiple variables, yielding R² values of 0.88 for total nitrogen and 0.66 for total organic carbon, respectively. These results underscore the potential of the AutoSMOGN algorithm to significantly contribute to the development of new control and optimization strategies, thereby enhancing the maintenance and operational efficacy of sewer systems.