(635g) Data-Driven Modeling and Optimization of an Industrial Scale Reverse Osmosis Desalination Plant

Population growth and globally depleting water sources contribute to immense stresses for water supply systems [1]. Water desalination technologies can lower these stresses, by utilizing saline water sources to generate water that can be used for various applications [2, 3]. Reverse osmosis (RO) superseded thermal separation technologies as the desalination industry leader due to its high energy efficiency, low space requirements, as well as process compactness [4], accounting for 68% of the globally produced desalinated water volume in 2019. Nevertheless, RO is still an energy and cost intensive process [5, 6], while optimizing the operation of RO systems can be computationally expensive due to the complexity of the interconnected membrane models. Data-driven surrogate models can help in overcoming this challenge, while still capturing the process behavior accurately.

Therefore, this work focuses on the optimal operation of desalination plants integrated with renewable energy sources through surrogate modeling. Firstly, a data-driven surrogate model for capturing the behavior of an industrial scale RO plant is developed. A neural network (NN) with rectified linear units (ReLU) is used to approximate collected data from the H2Oaks RO desalination plant in South-Central Texas. The data consists of process parameters per RO stage, as well as the energy consumption of the pumping system. Consequently, various possible surrogate model structures are investigated, and the performance compared. The developed NN is then transformed into a mixed-integer linear programming (MILP) formulation [7], and used for the derivation of minimal cost for the operation of the RO plant while analyzing energy-water trade-offs, thus enabling techno-economic feasibility analyses. The derived surrogate model can also be applied to the H2Oaks process control, as well as to investigate future investment decisions.

[1] R. Cory Allen, Yaling Nie, Styliani Avraamidou, Efstratios N. Pistikopoulos. Infrastructure Planning and Operational Scheduling for Power Generating Systems: An Energy-Water Nexus Approach, Computer Aided Chemical Engineering, 47, 233-238, 2019. https://doi.org/10.1016/B978-0-12-818597-1.50037-0

[2] Marcello Di Martino, Styliani Avraamidou, Julie Cook, Efstratios N. Pistikopoulos. An Optimization Framework for the Design of Reverse Osmosis Desalination Plants under Food-Energy-Water Nexus Considerations, Desalination, 503, 2021. https://doi.org/10.1016/j.desal.2021.114937

[3] Marcello Di Martino, Styliani Avraamidou, Efstratios N. Pistikopoulos. Superstructure Optimization for the Design of a Desalination Plant to Tackle the Water Scarcity in Texas (USA), Computer Aided Chemical Engineering, 48, 763-768, 2020. https://doi.org/10.1016/B978-0-12-823377-1.50128-2

[4] Muhammad Qasim, Mohamed Badrelzaman, Noora N. Darwish, Naif A. Darwish, Nidal Hilal. Reverse osmosis desalination: a state-of-the-art review, Desalination, 459, 59–104, 2019. https://doi.org/10.1016/j.desal.2019.02.008

[5] Edward Jones, Manzoor Qadir, Michelle T.H. van Vliet, Vladimir Smakhtin, Seong-mu Kang. The state of desalination and brine production: A global outlook, Science of the Total Environment, 657, 1343–1356, 2019. https://doi.org/10.1016/j.scitotenv.2018.12.076

[6] Jad R. Ziolkowska, Reuben Reyes. Prospects for Desalination in the United States - Experiences From California, Florida and Texas, Competition for Water Resources, 2017, pp. 298–316, https://doi.org/10.1016/B978-0-12-803237-4.00017-3.

[7] Bjarne Grimstad and Henrik Andersson. Relu networks as surrogate models in mixed-integer linear programs, Computers & Chemical Engineering, 131, 2019. https://doi.org/10.1016/j.compchemeng.2019.106580