Intelligent framework for bipolar membrane electrodialysis: AI-based forecasting and multi-objective optimization of electrochemical performance

초록

Sodium sulfate (Na2SO4) in battery wastewater poses ecological risks by disrupting osmoregulation and reducing the reproductive capacity of aquatic species. Bipolar membrane electrodialysis (BMED) is an effective method for directly recovering acids and bases from wastewater containing Na2SO4. This study developed a Temporal Fusion Transformer (TFT) model to predict the acid and base conductivities and electric current in a BMED system, using 5135 time-series data points collected under diverse experimental conditions. The optimized model achieved high predictive performance, with R2 values of 0.959, 0.976, and 0.924 for acid conductivity, base conductivity, and current, respectively. Attention analysis identified the feed conductivity and applied voltage as the most influential control variables. Furthermore, multi-objective optimization based on Non-dominated Sorting Genetic Algorithm-II (NSGA-II) was performed to explore optimal solution (223.6 mS/cm for acid conductivity, 128.8 mS/cm for base conductivity, and 0.265 kWh of energy consumption) between maximize acid and base production while minimizing energy consumption. The generalization capability of the TFT model was evaluated using additional experimental data that were not included in the model development, thereby validating its applicability to previously unseen scenarios. The findings of this study demonstrate the potential of the proposed model as a surrogate for process simulation and control, and suggest a forward-looking framework for future BMED applications, including the potential for integration with other desalination processes.

키워드