Free nitrous acid prediction in ANAMMOX process using hybrid deep neural network model

Area of Science:

• Environmental microbiology
• Biotechnology
• Chemical engineering

Background:

• Free nitrous acid (FNA) is vital for Anammox process stabilization.
• Direct measurement of FNA is challenging, hindering effective Anammox management.
• Accurate FNA prediction is needed for real-time operational control.

Purpose of the Study:

• To develop a hybrid deep learning model for predicting FNA concentrations in Anammox reactors.
• To optimize model hyperparameters using a multiobjective tree-structured Parzen estimator (MOTPE).
• To evaluate the model's performance against traditional methods for Anammox process monitoring.

Main Methods:

• A hybrid model combining Temporal Convolutional Network (TCN) with an Attention Mechanism (AM) was developed.
• The model, termed MOTPE-TCNA, was optimized using the Multiobjective Tree-structured Parzen Estimator (MOTPE).
• The model was applied to predict FNA in a case study of an Anammox reactor.

Main Results:

• Nitrogen removal rate (NRR) showed a high correlation with FNA concentration, enabling operational status forecasting.
• MOTPE effectively optimized TCN hyperparameters, achieving high prediction accuracy.
• The MOTPE-TCNA model demonstrated superior prediction accuracy (R² = 0.992), outperforming other models by 1.71-11.80%.

Conclusions:

• The MOTPE-TCNA model offers a highly accurate and advantageous deep neural network approach for FNA prediction.
• Accurate FNA prediction facilitates stable operation and easier control of the Anammox process.
• This predictive capability is beneficial for optimizing wastewater treatment efficiency.