2020 Virtual AIChE Annual Meeting
(583f) Applied Machine Learning for Prediction of Biochar Properties and Applications
Authors
In the only study of its kind, we performed supervised machine learning (ML) for multi-task prediction of fuel properties of the biochars derived from hydrothermal carbonization and pyrolysis, based on data collected by systematic review of 64 published articles. Results showed that support vector regression (SVR) with optimal hyper-parameters exhibited better generalized performance for prediction of both hydrochar and pyrochar properties with an average R2 of 0.91 and 0.88, respectively, compared to random forest (RF). Further on, the correlations between inputs features and output targets of the SVR model was investigated based on a combined game theory and model-agnostic approximation methodology called the Kernel SHAP (3). It was found that process temperature and C content in the feedstock were the significant features impacting fuel properties of both the chars, and N and H content were other important input feature for hydorchar and pyrochar, respectively. In an ongoing effort, we are currently developing ML models to predict the adsorption of CO2 on biochars based on the physicochemical characteristic of biochars including, BET surface area, pore volume and pore size, ash content and proximate analysis. Given the impetus garnered by data-driven analytics in recent years, the ML models and strategies as presented in this study can help understand the desired properties of chars, in order to better estimate their application, such as heavy metal or CO2 adsorption - to guide experiments based on the model inference.
References:
- Zhu X, Li Y, Wang X. Machine learning prediction of biochar yield and carbon contents in biochar based on biomass characteristics and pyrolysis conditions. Bioresource Technology. 2019 Sep 1;288:121527.
- Zhu X, Wang X, Ok YS. The application of machine learning methods for prediction of metal sorption onto biochars. J Hazard Mater. 2019 15;378:120727.
- Vega García M, Aznarte JL. Shapley additive explanations for NO2 forecasting. Ecological Informatics. 2020 Mar 1;56:101039.