Optimizing the conditions of biowastes hydrothermal treatment and predicting phosphorus fate in the hydrochar and liquid phase using machine learning

Abstract Applications of artificial intelligence (AI)- and machine learning (ML)-based methodologies for predicting optimal conditions in sustainable and effective management of biowastes and natural resources are of great concern. However, the AI-applications for optimizing the hydrothermal treatment (HT) of organic solid biowastes and prediction of nutrients fate during the HT process have not yet been investigated. Therefore, this study explores the application of different ML models (e.g., XGBoost, Decision Tree, and Random Forest) for optimizing HT of swine manure, focusing on the role of calcium (Ca) and iron (Fe) ions in phosphorus (P) distribution in the produced liquid and solid phases (hydrochar). Specifically, we investigated the fate of total P (TPS) in the hydrochar and inorganic P (IPL) in the liquid phase during HT. Experimental validation was conducted alongside the ML predictions, with XGBoost outperforming the other models, showing strong predictive accuracy for TPS (R 2 = 0.77) and IPL (R 2 = 1.0). Key factors influencing model accuracy included feedstock composition, reaction temperature, duration, solid–liquid ratio, and Ca and Fe concentrations. We found that the impact of time on TPS and IPL was minimal when the reaction time was less than 200 min, while pH showed a positive correlation with TPS and IPL. NMR and XRD analyses indicated that as the reaction severity increased, the organic P content in the hydrochar became more uniform. These findings highlight the potential of AI-based methodologies for optimizing HT processes, contributing to more sustainable and effective solutions for safe recycling, management, and development of bioresources. Graphical Abstract