Biochar energy prediction from different biomass feedstocks for clean energy generation
Nikhil Pachauri, Chang Wook Ahn, Tae Jong Choi
Abstract
This paper presents a novel approach for predicting various feedstock higher heating values (HHV) using a voting ensemble machine-learning model. The proposed model, referred to as VSGB, combines Support Vector Regression (SR), Gaussian Process Regression (GR), and Boosting (BO) using a weighted sum technique. The Invasive Weed Optimization (IWO) algorithm is employed to estimate hyperparameter values of the VSGB model. Moreover, comparative performance analysis is conducted using several models, such as linear regression (LR), generalized additive model (GAM), bagging (BAG), decision tree (DT), and neural network (NN). The simulation findings demonstrate that the VSGB has a high level of accuracy in predicting the HHV derived from biomass waste. This is evidenced by the lower Root Mean Square Error (RMSE) and Average Absolute Relative Difference (AARD%) values (0.813 and 2.959%, respectively) compared to other Machine Learning (ML) predictive models. Additionally, the present study establishes an empirical correlation between the higher heating value (HHV) and the input characteristics carbon (C), hydrogen (H), oxygen (O), nitrogen (N), and sulphur (S) through the utilization of the IWO algorithm. • A voting ensemble (VSGB) model is proposed for HHV Prediction • SR, GP, and BO are utilized to design a voting ensemble (VSGB) • IWO is used to estimate the hyperparameters of the VSGB. • Detailed performance analysis for VSGB is investigated.