Estimation of fast pyrolysis bio-oil properties from feedstock characteristics using rough-set-based machine learning

In this work, a data-driven rough-set-based machine learning model has been proposed as a pre-processing and predictive modelling tool to predict the pyrolysis bio-oil properties based on pyrolysis temperature and feedstock characteristics. A database consisting of feedstock proximate and ultimate analyses, pyrolysis temperature, bio-oil's pH value, and bio-oil's higher heating value was compiled and used to train the rough-set-based machine learning model. The resulting rule-based rough-set-based machine learning model demonstrated promising strength, certainty, and coverage factor. Furthermore, the emergent patterns and mechanistic plausibility of the rough-set-based machine learning models were analysed. The generated rules illustrated reasonable predictive capability in estimating the higher heating value and pH value of bio-oil based on the feedstock characterisation and pyrolysis temperature. Rough-set-based machine learning model is thus demonstrated to be a simple and straightforward approach for feedstock composition and pyrolysis temperature selection in pyrolysis/co-pyrolysis bio-oil production.

Automated summary

A data-driven rough-set-based machine learning model is proposed in this work to predict the properties of pyrolysis bio-oil. The model is trained based on a database consisting of feedstock proximate and ultimate analyses, pyrolysis temperature, bio-oil's pH value, and bio-oil's higher heating value. The results demonstrate that the model has good predictive capability and can be used for feedstock composition and pyrolysis temperature selection in pyrolysis bio-oil production.