Enhancing hydrogen sulfide control in urban sewer systems using machine learning models: Development of a new predictive simulation approach by using boosting algorithm.

Sewer networks are important urban infrastructure for transporting sewage to treatment plants, yet the generation of hydrogen sulfide within these systems poses significant challenges. This acidic toxic gas not only emits foul odors but also causes corrosion, necessitating effective control measures. Recent studies have introduced a modelling approach to predict and control the formation of hydrogen sulfide in sewer system. However, the conventional and mathematical models have demonstrated limitations in simulating non-linear data. Meanwhile, advanced (boosting) machine learnings are proving to be effective tools for forecasting complex data, making them particularly suitable for simulating of sulfide concentrations. In this work, we aimed to develop a novel approach to predict hydrogen sulfide formation in sewer systems. This work employed 11 machine learning models (4 boosting algorithms and 7 traditional algorithms) for over 700 datasets to analysis the correlations between the key sewer operational parameters (including pH, dissolved oxygen (DO), temperature, weather conditions, sulfate concentration, and ammonia levels) and hydrogen sulfide production. The results showed that eXtreme Gradient Boosting (XGBoost) has the highest prediction efficiency (R=0.97, RMSE=0.177 mg/L), outperformed other boosting and traditional methods. The newly developed boosting-based model successfully predicted sulfide formation in various sewer networks, validated against literature data (R> 0.9, RMSE of 0.24 mg/L), confirming its effectiveness for simulating hydrogen sulfide in sewer tunnels. The optimal conditions for minimizing total sulfide generation were identified by the XGBoost model. These findings have the potential to improve the control and operation of sewer system in the future.