Biochar production from valorization of agricultural Wastes: Data-Driven modelling using Machine learning algorithms

Biochar is an important nanomaterial that can be used in wastewater treatment. The production of biochar is often done by heating biomass in the absence of oxygen, a process known as pyrolysis. The valorization process involves a complex chemical reaction which is often not easily demystified. The data obtained from the valorization of the biomass can be employed to model the process for the purpose of understanding the relationship between the input and targeted parameters thereby optimizing the process. This study employs a data-driven approach to model biochar production from agricultural wastes. The parametric analysis shows that the biochar yields obtained from the different agriculture wastes were significantly influenced by temperature, heating rate, residence time, and Nitrogen flow rate. The support vector machine (SVM) models with different kernel functions displayed predictive potentials of the biochar production with R2 in the range of 0.5-0.8. The Gaussian process regression models with different kernel functions offer better prediction potentials compared to the SVM models as indicated by a higher R2 > 0.7. The artificial neural network-based algorithms outperformed the SVM and GPR as indicated by the R2 > 0.9 and low predictive errors. The relative importance analysis shows that the residence time during the valorization reaction has the most significant influence on the predicted biochar produced from agricultural wastes.

Automated summary

Biochar, produced through pyrolysis, is a crucial nanomaterial for wastewater treatment. This study uses a data-driven approach to model biochar production from agricultural wastes and finds that temperature, heating rate, residence time, and nitrogen flow rate significantly impact biochar yields. Support vector machine (SVM) and Gaussian process regression (GPR) models show potential for predicting biochar production, with GPR outperforming SVM. Artificial neural network-based algorithms provide even better prediction accuracy. Residence time during valorization reaction is identified as the most influential factor.