The Estimation of the Higher Heating Value of Biochar by Data-Driven Modeling
Jiefeng Chen,Lisha Ding,Pengyu Wang,Weijin Zhang,Jie Li,Badr A. Mohamed,Jie Chen,Songqi Leng,Tonggui Liu,Lijian Leng,Wenguang Zhou
DOI: https://doi.org/10.32604/jrm.2022.018625
IF: 2.115
2022-01-01
JOURNAL OF RENEWABLE MATERIALS
Abstract:Biomass is a carbon-neutral renewable energy resource. Biochar produced from biomass pyrolysis exhibits preferable characteristics and potential for fossil fuel substitution. For time- and cost-saving, it is vital to establish predictive models to predict biochar properties. However, limited studies focused on the accurate prediction of HHV of biochar by using proximate and ultimate analysis results of various biochar. Therefore, the multi-linear regression (MLR) and the machine learning (ML) models were developed to predict the measured HHV of biochar from the experiment data of this study. In detail, 52 types of biochars were produced by pyrolysis from rice straw, pig manure, soybean straw, wood sawdust, sewage sludge, Chlorella Vulgaris, and their mixtures at the temperature ranging from 300 to 800 degrees C. The results showed that the co-pyrolysis of the mixed biomass provided an alternative method to increase the yield of biochar production. The contents of ash, fixed carbon (FC), and C increased as the incremental pyrolysis temperature for most biochars. The Pearson correlation (r) and relative importance analysis between HHV values and the indicators derived from the proximate and ultimate analysis were carried out, and the measured HHV was used to train and test the MLR and the ML models. Besides, ML algorithms, including gradient boosted regression, random forest, and support vector machine, were also employed to develop more widely applicable models for predicting HHV of biochar from an expanded dataset (total 149 data points, including 97 data collected from the published literature). Results showed HHV had strong correlations (vertical bar r vertical bar > 0.9, p < 0.05) with ash, FC, and C. The MLR correlations based on either proximate or ultimate analysis showed acceptable prediction performance with test R-2 > 0.90. The ML models showed better performance with test R-2 around 0.95 (random forest) and 0.97-0.98 before and after adding extra data for model construction, respectively. Feature importance analysis of the ML models showed that ash and C were the most important inputs to predict biochar HHV.