Machine learning prediction of fuel properties of hydrochar from co-hydrothermal carbonization of sewage sludge and lignocellulosic biomass

Abstract

Machine learning approaches are emerging as a promising method for assisting in the control of thermochemical processes. eXtreme Gradient Boosting (XGB) and Random Forest (RF) were applied, for the first time, for prediction of fuel properties of hydrochar from co-hydrothermal carbonization of sewage sludge (SS) and biomass. XGB outperformed RF in the prediction of carbon content, O/C, higher heating value, and mass and energy yields, while RF surpassed XGB in the prediction of H/C, N/C, and fuel ratio. The R2 between the predicted and experimental values for the best models was in [0.94–1] and [0.83–0.95], respectively for training and test. The feature importance and partial dependence analyses were used to interpret models and provide comprehensive understanding of the input features’ impact. Based on the best models, a graphical user interface was created to make prediction easier for other researchers. By only knowing the properties of SS and lignocellulosic biomass, the authors could prior to experiments explore various co-HTC conditions and SS ratios to find the most appropriate conditions to obtain some given properties of hydrochar. This will save time and resources that are usually spent on several trial experiments that may sometimes not yield positive results.