Machine learning-driven prediction and control system for practical application of sulfur-based autotrophic...

Area of Science:

• Environmental Engineering
• Biotechnology
• Water Treatment

Background:

• Sulfur-based autotrophic denitrification (SADeN) is a sustainable nitrogen removal method.
• Its performance is sensitive to environmental fluctuations, challenging conventional control.
• Existing methods lead to unstable operations and excessive chemical usage.

Purpose of the Study:

• To develop an intelligent system (SADeN-PaCS) for predicting and controlling SADeN biofilters.
• To enable cross-season prediction, interpretable diagnosis, and adaptive regulation.
• To enhance the resilience and cost-efficiency of advanced nitrogen removal.

Main Methods:

• Developed an Artificial Neural Network (ANN) for predicting effluent nitrate (NO₃⁻-Neff).
• Employed model interpretability to identify key factors like water temperature (WT) and specific filler nitrate loading (SFNL).
• Implemented a dual-mode regulation strategy and an optimized chemical dosing model for extreme conditions.

Main Results:

• ANN achieved high prediction accuracy (R² = 0.94) for NO₃⁻-Neff.
• Identified WT as critical for resilience and SFNL as a key controllable variable.
• Dual-mode strategy optimized SFNL based on WT, reducing chemical use by 59.90% and stabilizing effluent quality.

Conclusions:

• Pioneered a diagnostic-driven control paradigm for SADeN processes.
• Leveraged model interpretability for adaptive, multi-level operational strategies.
• Achieved enhanced resilience and cost-efficiency in advanced nitrogen removal.