Digital Twins for Climate-Conscious WRRFs: Integrating Data and Knowledge for Process Optimization and Decision Support

Introduction Water resource recovery facilities (WRRFs) face increasing complexity as climate objectives push operators to balance multiple goals beyond effluent quality standards, including energy optimization, cost reduction, and greenhouse gas (GHG) mitigation. [1]. Nitrous oxide (N2O), contributing up to 80% of a WRRF's carbon footprint, adds significant challenges due to its production under varied operating conditions. Strategies like low dissolved oxygen (DO) levels and increased biogas-focused COD diversion can stimulate N2O formation [2]. Due to its high global warming potential (273-fold that of CO2), N2O mitigation has emerged as a priority influencing operation in WRRFs. The complexity of managing multiple operational objectives requires advanced digital tools to aid operators in understanding and solving these challenges [3]. To address these challenges, advanced digital tools, such as digital twins (DTs), offer integrated solutions combining process models, machine learning (ML), and control algorithms to aid decision-making [4]. The present contribution illustrates the use of a DT solution (TwinPlant, DHI A/S) to support decisions and optimize the operation of WRRFs. We show a full-scale case study presenting the implementation of TwinPlant with an additional climate footprint layer (TwiN2Ops) utilizing calibrated N2O model, accounting for wide range of modern process KPIs representing effluent quality, chemical costs, energy consumption, and total carbon footprint. We also show a proof-of-concept of integrating domain knowledge with ML models towards soft sensors with decision support capabilities. Methodology The operational DT was implemented with a dedicated carbon footprint layer. The NDHA process model including all known N2O possible pathways [5] was implemented in WEST (DHI A/S) extending the ASM2d model [6]. The model was first calibrated with experiments carried out at Marselisborg pilot WRRF (Aarhus, Denmark) and validated at Bjergmarken full-scale WRRF (Roskilde, Denmark). Calibration included heterotrophic and ammonia oxidizing bacteria kinetics under controlled pilot conditions. The full-scale model for Bjergmarken was first used to identify adequate N2O mitigation measures (carbon dosing, load equalization, aeration control adjustment) through offline scenario analysis. Subsequently, the full-scale model was included in the DT in operation, providing real-time and 48-hour forecasting N2O emission alongside conventional parameters (e.g., NH4, NO3, PO4, MLSS) and KPIs. Findings Controlled pilot testing with real wastewater was conducted at Marselisborg WRRF to calibrate the NDHA model using systematic calibration procedure to minimize identifiability issues. The heterotrophic bacteria kinetics were estimated using anoxic nitrate (NO3-), nitrite (NO2-), and N2O pulses under varying carbon doses. Moreover, aerobic ammonia (NHx) oxidation tests were conducted with NHx pulses, to estimate the ammonia oxidizing bacteria (AOB) kinetics. A selection of the model calibration results is presented in Figure 1. The calibrated model was validated at the full-scale Bjergmarken WRRF. The model could capture short-term N2O fluctuations and identify the operational issues associated with the high N2O. An example of this (N2O spike due to biomass washout under wet weather conditions) is presented in Figure 2. This model was implemented as a carbon footprint layer in the operational DT (Figure 3). The DT dashboard provides a visualization of the relative contribution of each emission source to the facility's overall carbon footprint (Figure 4). N2O mitigation was assessed in the DT (i) through offline scenario evaluations and (ii) using an optimization algorithm to recommend control settings to optimize multiple KPIs accounting for effluent quality, energy efficiency, operational cost, and carbon footprint. This evaluation allowed comparing the full impact of several mitigation scenarios on all KPIs. For instance, as demonstrated in Figure 5, a scenario involving increasing carbon dosing was tested, revealing that it has an overall positive effect on the process by reducing the cost per treated volume of wastewater by up to 25% while also lowering CO2eq emissions by 30% through reduction N2O emissions (while accounting for CO2 footprint of the chemical use). The carbon dosing also reduced effluent TN, NH4, and TP levels, while the additional cost of chemical use was offset by reductions in both power consumption and effluent taxes. A hybrid N2O soft sensor integrating ML and WEST modeling was tested to enhance decision support (results to be shown). Methodological experiences integrating domain knowledge with ML will be shared. Conclusion This contribution demonstrates how DTs that integrate process knowledge, mechanistic and data-driven methods support tailored WRRF operations to optimize conflicting process goals such as climate objectives, regulatory compliance, and operational efficiency.