Machine Learning Accelerates the Directional Construction of the Specific Surface Area of Biochar

Abstract

Biochar exhibits application potential for treating wastes and reducing carbon emissions, thereby improving efficiency in the petrochemical industry. Application prospects are particularly prominent in the remediation of petroleum-contaminated soil and carbon dioxide capture. Specific surface area of biochar serves as a key parameter governing its environmental application performance. However, complexity of biomass precursors and pyrolysis processes poses significant challenges to targeted design and prediction of biochar specific surface area via conventional experimental approaches. In this work four models were constructed and compared. The Random Forest model exhibited the best generalization ability, with a coefficient of determination of 0.79 and a root mean square error of 57.88 on the test set, thus being identified as the optimal prediction model. Pyrolysis process parameters were more dominant than elemental composition of biochar, and pyrolysis temperature was the most critical feature. Recommended pyrolysis parameters include temperatures above 700 °C and time exceeding 3 h, while elemental composition of biochar should favor a chemical composition with high carbon content (>40%) and high nitrogen content (>3%). These findings significantly reduce trial-and-error costs and accelerate the targeted development of biochar-based environmental materials, thereby advancing the practical application of biochar in pollution control and climate change mitigation.