Applied machine learning for predicting the properties and carbon and phosphorus fate of pristine and engineered hydrochar

作者：Shiyu Xie, Tao Zhang, Siming You, Santanu Mukherjee, Mingjun Pu, Qing Chen, Yaosheng Wang, Esmat F. Ali, Hamada Abdelrahman, Jörg Rinklebe, Sang Soo Lee, Sabry M. Shaheen · 发表于：Biochar · 年份：2025 · DOI：10.1007/s42773-024-00404-4 · 被引用次数：18 · 研究领域：Catalysis and Hydrodesulfurization Studies、Thermochemical Biomass Conversion Processes、Adsorption and biosorption for pollutant removal

Abstract Application of advanced techniques and machine learning (ML) for designing and predicting the properties of engineered hydrochar/biochar is of great agro-environmental concern. Carbon (C) stability and phosphorus (P) availability in hydrochar (HC) are among the key limitations as they cannot be accurately predicted by traditional one-factor tests and might be overcome by engineering the pristine HC. Therefore, the aims of this study were (1) to determine the optimal production conditions of engineered swine manure HC with high C stability and P availability, and (2) to develop the best ML models to predict the properties of HC derived from different feedstocks. Pristine- (HC) and FeCl 3 impregnated swine manure-derived HC (HC-Fe) were produced by hydrothermal carbonization under different pH (4, 7, and 10), reaction temperature (180, 220, and 260 ℃), and residence time (60, 120, and 180 min) and characterized using thermo-gravimetric, microscopic, and spectroscopic analyses. Also, different ML algorithms were used to model and predict the hydrochar solid yield, properties, and nutrients content. FeCl 3 impregnation increased Fe-phosphate content, while it reduced H/C and O/C ratios and hydroxyapatite P content, and therefore improved C stability and P availability in the HC-Fe as compared to HC, particularly under lower pH (4), temperature of 220 ℃, and at 120 min. The generalized additive ML model outperformed the other models for predicting the HC properties with a...