(452f) Hybrid Modeling and Predictive Control for an Open Water System with Varying Parameters

Low-lying countries must continuously pump excess water into the sea to keep their land dry and carefully manage the adjacent open waters [1]. In these water systems, the distribution of water is affected by hydraulic structures, such as pumps, weirs, and gates. Additionally, the operation of the open water systems can be easily affected by meteorological factors since they operate in an open environmental context [2]. It is crucial to maintain the water levels of the open water system within a safe range to ensure safety. Furthermore, the management of water distribution is closely linked to energy consumption and specific requirements related to agriculture, navigation, and flood mitigation [1][3][4]. In this research, we consider an open water system that consists of 14 branches with varying backwater areas and four connected external rivers. The hydraulic structures within this system include weirs, gates, and pumps.

The complex structure of the open water system and the disturbances caused by meteorological factors present challenges in achieving energy-efficient process operations while maintaining the water levels within safe areas. These considerations highlight the importance of developing and implementing advanced control methods to ensure the consistent and efficient operation of the open water system. Model predictive control (MPC) is one of the most widely recognized advanced control methods [5]. In the existing literature, MPC designs have been proposed for open water systems [1],[7][8]. To develop effective MPC designs, a high-fidelity model is necessary to accurately characterize the dynamics of the process. Building a first-principles model using mass and energy balance equations typically requires substantial time and resources, and may face difficulties due to the unavailability of some key parameters [9], such as the varying backwater areas in the open water system, which are difficult and expensive to measure in real time. On the other hand, a data-driven model can be developed using process data. However, relying solely on input-output data can lead to a model that lacks robustness across a wide range of operational conditions [10]. Hybrid modeling is an alternative modeling framework that handles the complex structure and dynamics of the process by integrating available first-principles knowledge with data information. Hybrid models are promising to provide high prediction accuracy, strong generalization, and good interpretability [11][12]. Based on an accurate dynamic model, zone MPC [13], which tracks the desired zone area instead of a set-point, can be developed to ensure safe operation and minimize operational energy consumption.

In this work, we present a complete design for an open water system that incorporates varying parameters. Given the varying parameters of the open water system can be challenging to measure in real time, a neural network is employed and integrated with the existing first-principles model to form an accurate dynamic model for the open water system. A layered control architecture is developed based on the hybrid model to achieve multiple operational objectives for the open water system. The first layer utilizes a zone MPC to obtain the optimal control actions, including the discharge references, to control the water levels in the branches within the safe range. The second layer addresses a scheduling problem to minimize the energy consumption by optimizing the control inputs for the pumps and gates with respect to the optimal discharge references. The effectiveness and performance of the proposed framework are evaluated and compared with the layered control architecture based on a first-principles model with approximated constant parameters.

References

• Horváth, B. van Esch, T. Vreeken, T. Piovesan, J. Talsma, and I. Pothof. Potential of model predictive control of a polder water system including pumps, weirs and gates. Journal of Process Control, 119:128–140, 2022.
• M. Maestre, L. Raso, P. J. van Overloop, and B. De Schutter. Distributed tree-based model predictive control on an open water system. American Control Conference, 1985–1990, 2012.
• Castelletti, A. Ficchì, A. Cominola, P. Segovia, M. Giuliani, W. Wu, S. Lucia, C. Ocampo-Martinez, B. De Schutter, and J. M. Maestre. Model predictive control of water resources systems: A review and research agenda. Annual Reviews in Control, 55:442–465, 2023.
• P. Becker, C. J. Jagtenberg, K. Horváth, A. Mitchell, and J. A. Rodríguez‐Sarasty. Optimization methods in water system operation. Wiley Interdisciplinary Reviews: Water, 11(6):e1756, 2024.
• Morari and L. J. Lee. Model predictive control: past, present and future. Computers & Chemical Engineering, 23(4-5):667–682, 1999.
• E. Garcia, D. M. Prett, and M. Morari. Model predictive control: Theory and practice—A survey. Automatica, 25(3):335–348, 1989.
• K. Pour, P. Segovia, E. Duviella, and V. Puig. A two-layer control architecture for operational management and hydroelectricity production maximization in inland waterways using model predictive control. Control Engineering Practice, 124:105172, 2022.
• Segovia, L. Rajaoarisoa, F. Nejjari, E. Duviella, and V. Puig. Model predictive control and moving horizon estimation for water level regulation in inland waterways. Journal of Process Control, 76:1–4, 2019.
• Shah, M. Z. Sheriff, M. S. Bangi, C. Kravaris, J. S. Kwon, C. Botre, and J. Hirota. Deep neural network-based hybrid modeling and experimental validation for an industry-scale fermentation process: Identification of time-varying dependencies among parameters. Chemical Engineering Journal, 441:135643, 2022.
• Narasingam and J. S. Kwon. Development of local dynamic mode decomposition with control: Application to model predictive control of hydraulic fracturing. Computers & Chemical Engineering, 106:501–511, 2017.
• Shah, M. Z. Sheriff, M. S. F. Bangi, C. Kravaris, J. S. I. Kwon, C. Botre, and J. Hirota. Deep neural network-based hybrid modeling and experimental validation for an industry-scale fermentation process: Identification of time-varying dependencies among parameters. Chemical Engineering Journal, 441:135643, 2022.
• S. F. Bangi, and J. S. Kwon. Deep hybrid modeling of chemical process: Application to hydraulic fracturing. Computers & Chemical Engineering, 134:106696, 2020.
• Liu, Y. Mao, and J. Liu. Model-predictive control with generalized zone tracking. IEEE Transactions on Automatic Control, 64(11):4698–4704, 2019.
• Liu and J. Liu. Economic model predictive control with zone tracking. Mathematics, 6(5):65, 2018.