Prediction and Optimization of Ammonia Removal in Direct Aeration Process Based on Wastewater Properties: An Integrated Experimental and Machine Learning Approach

Alkali-free direct aeration is an effective and economic approach for ammonia recovery from wastewater by increasing the pH through CO2 stripping. However, optimizing operating parameters is challenging due to its unknown adaptability to different wastewaters, leading to extensive trial-and-error experiments. To address this issue, a machine learning framework was developed to predict the ammonia removal efficiency, identifying the initial concentration ratio of dissolved inorganic carbon and total ammonia nitrogen (DICi/TANi) as a critical feature. Experimental results showed that increasing DICi/TANi from 0.28 to 1.15 intensified CO2 stripping and pH rise, facilitating continuous NH3 (aq) existence and boosting the TAN removal efficiency from 34.33% to 96.03%. From the insight of ammonia mass transfer, a higher aeration temperature and gas-liquid ratio promoted ammonia removal, with more significant enhancements under higher DICi/TANi due to their interactions. Machine learning methods were adopted to capture these interactions and make predictions. The eXtreme gradient boosting (XGBoost) algorithm accurately predicted TAN removal efficiency (R 2 = 0.981) with robust generalization. Finally, we proposed a conceptual closed-loop control framework based on the XGBoost model to achieve dynamic and real-time process optimization. Overall, the offered critical indicator and machine learning-assisted prediction model have significant guidance in the industrial application of direct aeration to recover ammonia from diverse wastewater.