Batch and continuous pilot-scale hydrothermal liquefaction of food waste: kinetic modelling and response surface methodology optimization

Hydrothermal liquefaction (HTL) presents a viable method for converting high moisture biomass to valuable products. Although extensively studied, its complex reaction mechanisms and accurate prediction of product yields remain limited. In this study, design of experiments and Response Surface Methodology (RSM) were applied to optimize HTL of food waste. An optimal range of 325–350 °C and 10–20 min of residence time were identified, resulting in maximum mass and carbon biocrude yields of 28 % and 43 %, respectively. Additionally, a predictive equation for biocrude yield was derived. To advance the mechanistic understanding a lumped kinetic model was developed by applying recommended best practices, including a modified Arrhenius equation, solution of the full system coupled with the actual temperature profile, and inclusion of short residence time data. Our findings highlight that model optimization alone does not guarantee either the validity of the reaction network or the accuracy of the estimated kinetic parameters. However, if such a tool is coupled with reaction pathway analysis, rigorous assessment of the parameters variance, and a critical analysis of the model bias and results, it can achieve physically consistent solutions with predictive capabilities. Both models were validated against independent data from a continuous pilot-scale HTL unit, showing robust predictive performances: biocrude yields predicted by the RSM and the kinetic models were within 7 % and 9 % of experimental values, respectively. Overall, this study advances the mechanistic understanding of HTL and highlights strategies for enhancing the predictive capabilities of kinetic modelling together with its inherent limitations . • RSM and kinetic models compared for their HTL product yields prediction ability. • Multiple kinetic parameter sets can predict HTL product yields equally well. • Certain reaction pathways can be excluded from consideration. • Models predict biocrude yields from continuous HTL processing within 7 % and 9 % error. • Suggestions for refining HTL kinetic model are provided.