Soft Sensor Monitoring of an Onsite Water Recycling System

Machine learning algorithms can be used as soft sensors to support real-time monitoring of contaminants during the recycled water treatment process. An analysis found that soft sensors for COD and TSS provide robust predictions (R^2 0.96 and 0.99), but E. coli predictions showed lower accuracy with regression algorithms.

Learning Objectives

Learn about how machine learning-based soft sensors can enable real-time monitoring and enhance the performance of onsite wastewater treatment systems (OWTS). Explore the selection for integrated multi-input in-line sensor data to accurately predict key water quality parameters such as COD, TSS, and E. coli. Discussion of comparative effectiveness of several different well-known machine learning algorithms, including partial least square regression, support vector regression, cubist regression, and quantile regression neural network, for real-time water quality estimation. Understand the challenges and opportunities in predicting E. coli in OWTS and the potential for further improvement.

Presenter(s)

Hsiang-Yang Shyu
University of South Florida