Deep Leaning Model Predictive Control for Reduction of N2O Emissions from WWTPs

Abstract:
Nitrous oxide (N2O) is a greenhouse gas (GHG) that has a global warming potential 265 times stronger than CO2. Up to 48% of Danish wastewater treatment plants’ (WWTP) carbon footprint can be related to nitrous N2O emissions – a product of nitrogen removal from wastewater – and 1% of the national CO2 equivalent (CO2e) emissions is caused by N2O emissions from WWTPs. N2O dynamics are known to be nonstationary, nonlinear and time varying – making them challenging to model. The objective of this project is to minimize N2O emission from WWTPs, enabling the realization of a climate neutral water sector. To this end, the project will develop deep learning predictive control algorithms which apply a plant wide optimization strategy with constraints on N2O, total nitrogen emissions and energy consumption. To ensure generalizability of the developed method, various process designs, and side stream processes will be evaluated. By use of online measurements, artificial intelligence (in the form of deep learning) and model predictive control, we can provide digital solutions for the Danish and international WWTPs, which guarantee efficient control of the treatment processes with respect to not only GHG emissions, but also water effluent quality, preserving nature and biodiversity in the surrounding environment. The goal is a 70% reduction in N2O emissions from WWTPs, which should pave the way for a climate neutral water sector in Denmark.

Funding: Innovationsfonden