Optimization of hydrothermal gasification process through machine learning approach: Experimental conditions, product yield and pollution

This study involves the development of a machine learning algorithm based Tunable Decision Support System (TDSS) and Tunable Recommendation System (TRS) for optimizing the process conditions and product yield values of hydrothermal gasification (HTG) of biomass. More than 500 datasets were collected from various studies and laboratory experiments performed. Initially, these data were correlated and the coefficients were determined using the Pearson matrix. Based on the statistical consolidation, weighted rank aggregates were formed and trends among these aggregates were derived and stored in the cloud repository. Test datasets were analysed using the predicted trends and iterations were continued until the required response values were derived. The precision of the predicted results was compared with actual results and %Accuracy values showed more than 94% accuracy in predicting HTG process conditions for a given set of biomass input characteristics. The correlation studies revealed that %Hydrogen, %Carbon, %Volatile matter, %Moisture & %Oxygen present in biomass are highly responsible for the yield and quality of product gas by influencing HTG conditions like Temperature (T in degrees C), Pressure (P in MPa), Catalyst Loading (Cat. Load in wt. %), Solvent to Biomass ratio (S/B) and Time (t in min). As a part of the optimization study, the production of polluting gases like CO2 and CO (collectively accounted for pollution index) were also modeled. The TRS function available in this framework suggests suitable alternatives when the input conditions do not suit to yield an expected response outcome. Multiple data and parameter-driven processes like HTG can be optimized efficiently using this system.