Koopman-based model predictive control for denitrification process in wastewater treatment

Effective nitrogen removal in wastewater treatment is critical to meet stringent total nitrogen (TN) discharge standards. Denitrification, the biological conversion of nitrate (NO3--N) to nitrogen gas under anoxic conditions, is highly sensitive to dissolved oxygen (DO), carbon sources, and microbial activity, and conventional strategies such as fixed aeration or Proportional-Integral-Derivative (PID) feedback control often fail to adapt to dynamic influent conditions, causing TN exceedance and energy inefficiency. We propose a Koopman operator-based Model Predictive Control (Koopman-MPC) framework for intelligent aeration control. A data-driven Koopman operator linearizes the nonlinear denitrification dynamics in a latent space, enabling robust multi-step prediction, while a predictive controller optimizes aeration under nitrate, energy, and actuator constraints. Spectral radius regularization ensures model stability and recursive feasibility. Validated on real-world data from a municipal plant in Zhejiang, China, Koopman-MPC achieves a 28.6% TN reduction, 13.7% energy savings, and superior adaptability compared with PID and Long Short-Term Memory (LSTM)-based MPC baselines. A 24-hour continuous test attains 99.2% discharge compliance while maintaining operational stability. By integrating operator learning with control optimization, this study offers a theoretically sound and practically viable solution for advanced nitrogen removal.