ABSTRACT
The present study attempts a new approach for optimizing higher heating value (HHV) of agricultural biomass from only proximate analysis data, thereby requiring nominal specificity compared to elemental composition-based models. This objective was achieved by classifying the largest extracted 1140 data into five categories and by adopting five machine learning (ML) models namely: Multi Layer Perceptron, Random Forest, Support Vector Machine, Linear Regression and Artificial Neural Network. Further, relative correlation and contribution of different input parameters on target variable was performed to forecast the impact level and interaction. It showed higher relevancy factor for FC and VM as 0.82 and 0.91 respectively. Compared to other models, Artificial neural network and Random Forest model provided better accuracy with higher correlation coefficient values (0.991 and 0.989 respectively) and lower root mean square error 0.215 and 0.283 respectively. The validity and applicability of the database and ML modeling were verified using Box plots, in which more than 96% of the data for each category of biomass were located in the valid region. The values of statistical indicators proved better prediction accuracy of our proposed models in comparison to other reported literature models for all predefined categories of biomass.
Abbreviations/Nomenclatures
Abbreviation | = | Full form |
HHV | = | Higher Heating Value (MJ/kg) |
FC | = | Fixed carbon (wt%) |
VM | = | Volatile matter (wt%) |
ASH | = | Ash content (wt%) |
ML | = | Machine Learning |
ANN | = | Artificial Neural Network |
RF | = | Random Forest |
MLP | = | Multi Layer Perceptron |
SVM | = | Support Vector Machine |
LR | = | Linear Regression |
WEKA | = | Waikato Environment for Knowledge Analysis |
AW | = | Agri waste |
FBP | = | Fruits by product |
GL | = | Grass leaves and tree species |
OWM | = | Other waste materials |
WC | = | Wood chips |
wt% | = | Mass fraction or percentage by weight |
PCC | = | Pearson Correlation Coefficient |
CC or R2 | = | Coefficient of correlation |
RMSE | = | Root mean square error |
Disclosure statement
No potential conflict of interest was reported by the author(s).
Data availability
The datasets of 1140 data generated in this study are available from the corresponding author on reasonable request.