Learning-Enhanced Smart Predictive Digital Twins for Drinking Water Supply Optimization

Learning-Enhanced Smart Predictive Digital Twins for Drinking Water Supply Optimization A. Andrade-Campos*, M. Brás*, S. Mota*, A. L. Sousa*, T. Pereira*, M. Alão*,**, A. Reis*,** *Department of Mechanical Engineering, TEMA - Centre for Mechanical Technology and Automation, LASI - Intelligent Systems Associate Laboratory, University of Aveiro, Campus Universitário de Santiago, 3810-193 Aveiro, Portugal; g[email protected] **Scubic, Parque da Ciência e Inovação, Edifício Principal, 3830-352 Ílhavo Abstract: Water supply systems face increasing costs, aging infrastructure, water losses, and climate challenges, demanding smarter management solutions. This work presents a predictive digital twin framework integrating machine learning (ML) for adaptive hydraulic modelling and short-term water demand forecasting. Instead of traditional EPANET simulations, differential ML models improve scalability and efficiency. A novel duty-cycle-based pump scheduling optimization, combined with model predictive control (MPC), enhances energy efficiency. The framework also incorporates realtime leakage detection and uncertainty analysis to ensure robust decision-making under varying conditions. By integrating ML, MPC, and optimization, this digital twin improves operational efficiency, sustainability, and resilience in drinking water management. Keywords: Water supply system digital twin; predictive modelling and control; machine learning. Water supply systems face rising costs, aging infrastructure, water losses, and climate challenges, making it harder for utilities to ensure reliable, sustainable, and costeffective operations (vd. Figure 1). Efficient management of drinking water supply systems requires accurate modelling, real-time monitoring, and adaptive control strategies (Coelho and Andrade-Campos 2014). Smart predictive digital twins have emerged as a powerful solution, integrating data-driven models with optimization and control techniques to enhance operational efficiency. This communication presents a novel digital twin framework that incorporates differential machine learning (ML) models to replace traditional EPANET-based hydraulic simulations and recurrent-ML techniques to forecast short-term water demands time-series. By leveraging machine learning, hydraulic predictions become more adaptive, computationally efficient, and scalable for large networks. A key component of the proposed framework (vd. Figure 2) is an advanced pump scheduling optimization method that departs from conventional metaheuristic and computer-inefficient approaches. Instead, it introduces duty-cycle-based optimization variables (Brás et al. 2024), combined with model predictive control (MPC), to improve energy efficiency and operational reliability (Reis et al. 2025). Additionally, the digital twin integrates real-time water leakage detection, utilizing hydraulic models to identify anomalies and minimize losses. Water demand forecasting techniques are also embedded to enhance short-term operational planning. Uncertainty analysis is a critical aspect of the proposed predictive digital twin, ensuring robustness in decision-making under varying demand conditions and unexpected system disturbances. The integration of MPC with uncertainty analysis further strengthens the adaptability of the control strategy, allowing for real-time adjustments to operational constraints. By combining these advancements, the proposed digital twin framework provides a comprehensive and intelligent approach to drinking water supply management, optimizing performance while enhancing sustainability and resilience. ACKNOWLEDGEMENTS This research was supported by the FEDER and Regional Operational Program of the Center Region (CENTRO2030) within project I-ReTiS-LeaksD&Op nº 17304 (CENTRO2030-FEDER-01177300) and through the Portuguese Foundation for Science and Technology (FCT), supported by the Recovery and Resilience Plan (PRR), within project I-ReTiS-Leaks (2024.07270.IACDC). This paper was also supported by the project UID 00481 Centre for Mechanical Technology and Automation (TEMA). Figure 1.1 Water supply systems face a plethora of challenges that compromise their efficiency and sustainability. Among these, escalating energy costs, aging infrastructure, significant water losses, rising customer expectations, difficulties in staff knowledge retention, and the adverse impacts of climate change stand out as critical issues. The result is a scenario where water utilities struggle to maintain service reliability, meet environmental standards, and ensure economic viability. Figure 1.2 A novel smart predictive digital twin technology is presented in this paper, contributing to the advancement of digital twin technology for water supply systems by integrating physics-informed machine learning, predictive optimization, and real-time analytics into a cohesive framework. REFERENCES Coelho, B., Andrade-Campos, A. 2014. Efficiency achievement in water supply systems - A review, Renewable and Sustainable Energy Reviews, 30:59-84, doi:10.1016/j.rser.2013.09.010. Reis, A. L., Andrade-Campos, A., Matos, P., Hengeller Antunes, C., Lopes, M. 2025. An energy and cost efficiency Model Predictive Control framework to optimize Water Supply Systems operation, Applied Energy 384(6): 125478, doi: 10.1016/j.apenergy.2025.125478. Brás, M., Moura, A., Andrade-Campos, A. 2024. Cost efficiency in water supply systems: An applied review on optimization models for the pump scheduling problem, European Journal of Operational Research 323(1):1-19, doi: 10.1016/j.ejor.2024.07.039.