Predictive modeling and optimization of hydrochar properties from food waste hydrothermal carbonization using machine learning techniques

Hydrothermal carbonization (HTC) efficiently converts food waste into hydrochar for energy, soil improvement, and environmental remediation. However, optimizing HTC is challenging due to the complex interactions of temperature, residence time, moisture content, and catalyst dosage. This study explores the application of machine learning (ML) techniques for predicting and optimizing hydrochar yield and properties from the HTC of food waste. Key process parameters-temperature (120-360 °C), catalyst dosage (0-2 g), residence time (30-270 min), and moisture content were analyzed using three ML models: XGBoost (XGB), Support Vector Regression (SVR), and Linear Regression (LR). Among them, XGBoost demonstrated the highest predictive accuracy (R² = 0.87, MAE = 0.81, RMSE = 1.02), outperforming SVR (R² = 0.9386, MAE = 3.08) and LR (R² = 0.85, MAE = 0.86). Temperature was identified as the most significant factor (importance score = 0.91), followed by catalyst dosage (0.61). Optimized conditions yielded 48.5 g of hydrochar per 100 g of dry food waste, with an energy recovery efficiency of 68 %. The inclusion of TiO₂ nanoparticles as a catalyst significantly enhanced hydrochar properties, increasing carbon retention from 56.8 % to 77.6 %, bulk density from 0.37 g/cm³ to 0.89 g/cm³, and high heating value from 18.4 MJ/kg to 22.7 MJ/kg. These findings demonstrate the potential of ML-driven optimization for real-time control of HTC, reducing reliance on trial-and-error experimentation and advancing sustainable waste-to-energy conversion technologies.