Machine Learning Prediction of Foaming in Anaerobic Co-Digestion from Six Key Process Parameters

Foaming in co-digested anaerobic digesters can reduce biogas production, leading to economic loss. However, the underlying causes of foaming are not completely understood. This study investigated a field-scale mesophilic digester system that experienced intermittent foaming, employing experimental and modeling methods over a 16-month period. Samples were collected during both foaming and non-foaming events and were thoroughly characterized for methane (CH4) yields and different physical and chemical concentrations, including volatile solids (VS), metals, total phosphorus (TP), total chemical oxygen demand (TCOD), total volatile fatty acids (TVFAs), and total alkalinity (TALK). Machine learning techniques were applied to predict foaming events with several algorithms tested to optimize prediction accuracy. The results showed that digester liquid and effluent samples collected from foaming events had significantly lower (p < 0.05) average CH4 yields (77 and 45 mL CH4 g VS−1) than during non-foaming events (150 and 83 mL CH4 g VS−1). Recursive feature modeling identified six key parameters (1. Fe(II):S; 2. Fe(II):TP; 3. TCOD; 4. Fe; 5. TVFA:TALK; and 6. Cu) associated with digester foaming. Among the tested machine learning models, the support vector machine (SVM) algorithm achieved the highest recognition accuracy of 87%. This study demonstrates that the interactions of multiple chemical and physical process parameters are an important consideration for predicting anaerobic digester foaming.