Development of an ensemble of machine learning algorithms to model aerobic granular sludge reactors

Area of Science:

• Environmental Engineering
• Wastewater Treatment Technologies
• Computational Science

Background:

• Aerobic granular sludge (AGS) is an emerging wastewater treatment technology with limited long-term operational data.
• The scarcity of data hinders the development of robust data-driven predictive models for AGS processes.
• Accurate prediction of AGS performance is crucial for optimizing treatment efficiency and operational stability.

Purpose of the Study:

• To develop and validate a machine learning model for simulating the performance of aerobic granular sludge (AGS) reactors.
• To address the data scarcity challenge in emerging wastewater treatment technologies.
• To provide a predictive tool for optimizing AGS reactor operation under various conditions.

Main Methods:

• Utilized 475 days of operational data from three laboratory-scale AGS reactors.
• Employed RReliefF ranking for input feature selection after multicollinearity reduction.
• Implemented a five-stage model structure with ensemble methods (ANN, SVR, ANFIS) for enhanced prediction accuracy.

Main Results:

• The developed machine learning model accurately predicted MLSS, MLVSS, SVI5, SVI30, granule size, and effluent COD, NH4-N, and PO43-.
• Achieved high prediction performance with average R² of 95.7%, nRMSE of 0.032, and sMAPE of 3.7%.
• Demonstrated the model's capability to simulate AGS process dynamics effectively.

Conclusions:

• Machine learning models can effectively simulate AGS reactor performance, overcoming data limitations.
• The developed ensemble model offers a reliable tool for predicting key operational and effluent parameters.
• This approach supports the advancement and wider adoption of AGS technology in wastewater treatment.