Optimizing TAN Monitoring in Various Anaerobic Digestion Systems: A Machine Learning Approach with XGBoost and AdaBoost

Controlling Total Ammonia Nitrogen (TAN) concentrations is critical for ensuring stabilization in anaerobic digestion (AD) systems. This study investigates the performance of the XGBoost and AdaBoost algorithms for TAN estimation employing operational data from several AD systems. After adjusting hyperparameters and evaluating independent test sets, XGBoost ($R^{2}=0.94, R M S E=0.22$) outperformed AdaBoost ($R^{2}=0.91, R M S E=0.3$) in terms of prediction accuracy. The SHAP (SHapley Additive ExPlanations) evaluation highlighted diverse feature significance patterns: XGBoost selected organic loading rate (OLR) as a highly relevant parameter $(\mathrm{SHAP}=0.766)$, whereas AdaBoost depended mostly on temperature ($\mathrm{SHAP}=0.811$). These results supply important insights into model selection in various AD operating contexts. The study illustrates how comprehensible machine learning may improve process monitoring and decision-making in biogas facilities, resulting in more stable and effective AD processes.