Novel approach for AI-based N2O emission reduction in biological wastewater treatment relying on genetic algorithms and neural networks.

Abstract

The potential of measurement-based control strategies for achieving lower N₂O emissions in biological wastewater treatment is limited due to strong temporal variations in N₂O emissions and a lack of measurement data regarding influencing parameters. To address this issue, a novel artificial intelligence (AI)-based process optimization method for minimizing N₂O emissions was developed, relying on a genetic algorithm to automatically determine the control settings associated with minimum N₂O emissions for an individual operating situation. The genetic algorithm employs a validated prediction model to evaluate the effect of individual control parameter sets on N₂O emissions and other operating targets. For this purpose, neural networks were trained using data generated with a mechanistic model. This approach is beneficial in practical applications as prediction networks could be successfully trained even if only limited data is available. The developed method also includes a classification algorithm to check the reliability of the AI-suggested control strategy. Two modeling studies confirm that the practical application of the developed methodology holds the potential for a considerable reduction in emissions (43% or 1,588 t CO₂e/a) while still achieving the required effluent quality. Operational settings are identified in less than 2 minutes so that the approach can be applied on a large scale.