Flexibility of Nested Models for the Energy Efficiency of Large-Scale Membrane Bioreactors – FlexMIEE

The FlexMIEE project addresses critical challenges in the optimization of large-scale Membrane Bioreactors (MBRs), with a strong focus on reducing energy consumption while adhering to the environmental constraints imposed on modern wastewater treatment plants (WWTPs). A key challenge in MBR modeling lies in the hierarchical and multi-scale nature of the underlying processes, where complex interactions between biological processes, membrane filtration, and energy consumption demand advanced modeling and optimization approaches. This project introduces an innovative hybrid methodology, combining mechanistic models with artificial intelligence (AI) techniques to enhance predictive accuracy, control strategies, and operational adaptability. To achieve this, FlexMIEE will develop robust and flexible numerical tools capable of managing inherent operational uncertainties and optimizing MBR performance under variable pollutant loads and membrane fouling phenomena. A major methodological advancement of the project involves integrating sensitivity analysis for nested models, incorporating advanced uncertainty representations through Symbolic Data Analysis (SDA), which enables handling variables as intervals or probability densities. This approach will improve uncertainty quantification related to operating conditions, membrane performance, and energy consumption, while providing interpretable and actionable indicators for decision-making. Furthermore, to enhance the prediction of critical variables, the project will explore Physics-Informed Neural Networks (PINNs) and stacked ensemble models, enabling effective integration of experimental data, mechanistic models, and machine learning. For real-time MBR optimization, FlexMIEE will implement Nonlinear Model Predictive Control (NMPC) to jointly optimize operational costs and effluent quality, ensuring compliance with environmental regulations. A critical challenge is state estimation, as many essential variables (e.g., substrate, biomass concentrations) cannot be directly measured in WWTPs. To address this, FlexMIEE will leverage advanced state estimation techniques for NMPC, including Reinforcement Learning (RL), to infer hidden variables and improve closed-loop control robustness. Reinforcement Learning will also be applied to autonomous aeration control, developing a self-adaptive strategy that dynamically adjusts operational parameters in response to process variations. The project’s industrial validation will rely on large-scale real-world data collection, supported by a partnership with the SIAAP (Syndicat Interdépartemental pour l’Assainissement de l’Agglomération Parisienne), a leading WWTP operator. From a scientific perspective, a core objective is defining sensitivity metrics for nested models, accounting for structural uncertainties and developing tools to analyze their propagation across sub-models. The originality of FlexMIEE stems from its applied research framework, aligned with concrete environmental needs, positioning it as a pioneering initiative in MBR optimization in the era of smart data processing and advanced control. This initiative was launched following the recruitment of Mr. Ouaret R., the project lead and research coordinator at LGC.