Advancing the Comprehension of Iron Ions with Different Valences in Anammox-Based Processes: Insight into the Response and Mechanism Based on Prediction and Classification Machine Learning Models

Abstract

The responses of anammox to Fe 2+ and Fe 3+ have been widely discussed; however, the critical concentration of Fe 2+ and Fe 3+ is not certain due to the different culture environments. A reliable and widely applicable predictive classification system based on the anammox-based system with iron-containing wastewater needs to be established, which can surmount the independence between different experiments and be combined with statistical analysis to elucidate the mechanism of the effect of Fe 2+ and Fe 3+ based on critical concentrations. The results confirmed that 5 mg/ L iron promoted the nitrogen removal process, while higher concentrations inhibited nitrogen removal with a ratio of above 80%. Moreover, the results of Spearman correlation analysis proved that Fe 2+ showed a more obvious effect on the nitrogen removal rate than Fe 3+ . To precisely predict the nitrogen removal performance of anammox, Support Vector Machine, XGBoost, and Random Forest (RF) were compared, and the RF model was confirmed as the preferable model (R 2 = 0.99). According to the classification model, the influence of iron ions on the anammox performance could be successfully traced back to the valent states of iron with an accuracy of 97.7%. Furthermore, a mechanism of the nitrogen removal process in the anammox-based system under iron ion stress was proposed.