Smart aeration strategy for electroplating wastewater: Insights from a dynamic interaction networks based interpretable optimization

It is a challenge to achieve adaptive aeration control for electroplating wastewater, as existing studies, which are largely focused on municipal plants and reliant on static or opaque models, fail to link dynamic interactions with interpretable optimization. We developed a data-driven framework that integrates sliding-window correlation networks, SHAP-enhanced XGBoost models, and SHAP-guided Bayesian optimization. This framework identifies key process variables and effluent-specific control targets, underscoring the need for differentiated aeration strategies rather than a single plant-wide setting. The network analysis showed that early windows had dense positive links between flow related variables such as Flow_H and Flow_F and effluent indicators, whereas later windows became sparse with many isolated nodes, indicating a weakened aeration-effluent coupling and the instability of fixed control relationships. XGBoost models for COD_out, NH₃-N_out, TN_out and TP_out achieved high predictive accuracy with R² values between 0.951 and 0.998, and SHAP consistently identified Flow_H and Flow_F together with influent loads as key predictors, confirming that zone specific aeration flows are major operational levers for effluent quality. Using these interpretable models, Bayesian optimization combined with response surface analysis found that higher Flow_F minimized predicted COD and NH₃-N, while phosphorus removal (TP_out) benefited from elevated Flow_H and Flow_M, underscoring the need for differentiated aeration strategies tailored to individual effluent targets rather than a single plant wide setting. Overall, this framework offers a transparent, data-driven approach for adaptive and potentially energy-efficient aeration management in electroplating WWTPs and may help operators maintain reliable effluent-quality compliance while lowering the energy and greenhouse-gas footprint of metal-rich industrial discharges.