A Data-Driven Approach for Integrated Design and Dynamic Optimization Under Water Demand Uncertainty of Renewable Electrodialysis Systems

This work proposes a modeling framework based on artificial neural networks (ANN) for the integrated design and dynamic optimization of renewable electrodialysis (ED) systems considering water demand uncertainty, using a first-principles ED model as the data source for the development of the ANN. The optimization goal is to identify the optimal photovoltaic (PV) and battery (BAT) capacities and the optimal time-varying ED voltage and flow profiles during the batch process, considering an uncertain distribution of potential water demand for each batch over an annual operating horizon. This is achieved by minimizing the annual capital and operating costs of the renewable ED-PV-BAT system. The ANN model demonstrated excellent predictive capabilities that closely matched the data generated by the ED model, with ± 3.5–9.6% and ± 2.0–4.8% error margins in the prediction intervals at a 95% confidence level. The optimal design resulting from dynamic optimization exhibited a lower cost than the design attained from the steady-state optimization, as the batch time and energy consumption were 50% and 17% lower, respectively. For this design, the energy consumption and nitrate concentration predicted by the ANN were only 0.31% and 1.2% different from the ED model predictions, without any effects on the predicted costs and batch times.