Development of a wide-range soft sensor for predicting wastewater BOD₅ using an eXtreme gradient boosting (XGBoost)...

Area of Science:

• Environmental Science
• Machine Learning Applications
• Wastewater Engineering

Background:

• Accurate wastewater monitoring is crucial for treatment process calibration.
• Existing hardware sensors have limited linear ranges, failing at extreme pollutant concentrations.
• Conventional soft sensors struggle with highly skewed data distributions common in wastewater.

Purpose of the Study:

• To develop a novel soft sensor for detecting extremely high pollutant concentrations in wastewater.
• To utilize eXtreme Gradient Boosting (XGBoost) machine learning for measuring wastewater organics, specifically 5-day biochemical oxygen demand (BOD5).
• To validate the performance of the XGBoost soft sensor on real-world wastewater treatment plant data.

Main Methods:

• Development of a soft sensor model using XGBoost machine learning algorithm.
• Application of XGBoost's sparsity awareness for handling sparse input matrices.
• Testing and validation of the soft sensor on influent and effluent BOD5 data from two wastewater treatment plants.

Main Results:

• The XGBoost-based soft sensor demonstrated superior performance in detecting extreme BOD5 values compared to conventional soft sensors.
• The model effectively handled highly skewed data distributions.
• The soft sensor proved capable of functioning with sparse input data, addressing hardware sensor limitations.

Conclusions:

• XGBoost provides a robust solution for developing soft sensors capable of measuring extreme pollutant levels in wastewater.
• This approach enhances the reliability of wastewater monitoring, enabling better calibration of treatment processes.
• The developed soft sensor overcomes key limitations of existing hardware and conventional soft sensor technologies.