Comprehensive review of hydrothermal liquefaction data for use in machine-learning models

Hydrothermal liquefaction is a new, sustainable pathway to generate biogenic liquids from organic resources. The technology is compatible with a wide variety of resources such as lignocellulosic resources, organic waste, algae, and sewage sludge. The chemistry is complex and predictions of yields are notoriously difficult. Understanding and modeling of hydrothermal liquefaction is currently mostly based on a simplified biochemical analysis and product yield data. This paper presents a large dataset of 2439 experiments in batch reactors that were extracted from 171 publications in the scientific literature. The data include biochemical composition data such as fiber content and composition, proteins, lipids, carbohydrates, and ash. The experimental conditions are recorded for each experiment as well as the reported yields. The objective of this paper is to make a large database available to the scientific community. This database is analyzed with machine-learning tools. The results show that there is no consensus on the analysis techniques, experimental procedures, and reported data. There are many inconsistencies across the literature that should be improved by the scientific community. Machine-learning tools with a large dataset allow the generation of reliable yield production tools with a large application field. Given the accuracy of the data, the overall precision of prediction in an extrapolation to new results can be expected to be around 10%.