Soft Sensor for Effluent Quality in Onsite Wastewater Treatment
Machine-learning soft sensors predict COD and TSS of an onsite wastewater treatment system from in-line water-quality sensors: support vector regression reached R² = 0.96 for COD and cubist regression R² = 0.99 for TSS.
Soft Sensor Solution
Approach
Multi-input, single-output soft sensors combine parameters that well-established in-line sensors can measure to predict water-quality parameters that otherwise need sampling and laboratory analysis. Four machine-learning algorithms were compared: partial least squares regression (PLS), support vector regression (SVR), cubist regression and a quantile regression neural network (QRNN).
Input Variables
- Turbidity
- Color
- pH
- Ammonium (NH₄⁺)
- Nitrate (NO₃⁻)
- Electrical conductivity
Output Variables
- Chemical oxygen demand (COD)
- Total suspended solids (TSS)
- E. coli
Model Type
- Support vector regression
- Cubist regression
- PLS and QRNN (compared)
Key Performance Indicators
Results
-
On water-quality data from a field-tested onsite wastewater treatment system operated for almost two years (56 data points), SVR predicted COD with R² = 0.96 and cubist regression predicted TSS with R² = 0.99. E. coli could not be predicted adequately (best R² = 0.22). The authors conclude that, although the dataset is small, machine-learning soft sensors can give useful real-time estimates of parameters that are otherwise measured off-line.
Why It Matters
- COD, TSS and E. coli need sampling and laboratory analysis, so they give no real-time view of treatment performance or component failure.
- Real-time COD analysers exist but are not economically viable for onsite systems because of their cost and chemical consumables; a soft sensor on established in-line sensors is a cheaper route to continuous monitoring.
Have a control challenge? Let's talk.
Sources
Shyu, H.-Y., Castro, C. J., Bair, R. A., Lu, Q., Yeh, D. H. — open access.
Pattern Overview
A soft sensor predicts laboratory-measured water-quality parameters (COD, TSS) from parameters that in-line sensors measure continuously, giving operators a real-time estimate between laboratory samples.
When to Use This Pattern
- Effluent quality is monitored through laboratory analysis, and the delay limits operational decisions.
- Online analysers for COD or TSS are too expensive for the installation, or their reagents and maintenance are a burden.
- Reliable in-line sensors (turbidity, pH, conductivity, nutrients) are already installed or cheap to add.
What the cited study shows — and its limits
Different algorithms won for different outputs (SVR for COD, cubist regression for TSS), so candidate models should be compared per output. The dataset was small (56 data points), and E. coli was not predicted adequately. A production soft sensor needs more data, a plan for recalibration, and monitoring of the input sensors themselves.
Related Patterns
Contact
Send a message
Direct contact
Dr. Rafał Noga
Stay Updated
Get insights on Industrial AI, APC, and process optimization delivered to your inbox.