Sustaining struvite production from wastewater through machine learning based modelling and process validation

The looming scarcity of phosphorus rock and intensification of its extraction for fertilizing applications has triggered the researchers to work upon a potential alternative such as struvite precipitation from wastewaters. Struvite production at commercial scale requires the support of novel prediction tools to smoothen the planning and execution processes. The present work aims at predicting the struvite recovery using several machine learning algorithms such as linear regression model, polynomial regression model, random forest regression model and eXtreme Gradient Boosting (XGB) regression model. Datasets for ten significant process parameters such as pH, temperature, concentrations of phosphate, ammonium and magnesium, stirring speed, reaction and retention time, drying temperature and time of various wastewater sources were collected for predicting the recovery. To minimize the loss function, extensive grid search hyperparameter tuning was performed to optimize the model. XGB was found to be the most robust method for prediction of nutrient recovery as struvite. The highest regression coefficient (R2) of 0.9683 and 0.9483 were achieved for phosphate and ammonium recoveries, respectively. The key influencing factors on target output were studied using SHapley Additive exPlanations (SHAP) plots that depicts the interactive effect of each of the input parameters on phosphate and ammonium recovery. Experimental validation was carried out to further support the model predictions.