Torrefied biomass quality prediction and optimization using machine learning algorithms

Abstract

Torrefied biomass is a vital green energy source with applications in circular economies, addressing agricultural residue and rising energy demands. In this study, ML models were used to predict durability (%) and mass loss (%). Firstly, data was collected and preprocessed, and its distribution and correlation were analyzed. Gaussian Process Regression (GPR) and Ensemble Learning Trees (ELT) were then trained and tested on 80 % and 20 % of the data, respectively. Both machine learning models underwent optimization through Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) for feature selection and hyperparameter tuning. GPR-PSO demonstrates excellent accuracy in predicting durability (%), achieving a training R2 score of 0.9469 and an RMSE value of 0.0785. GPR-GA exhibits exceptional performance in predicting mass loss (%), achieving a training R2 value of 1 and an RMSE value of 9.7373e-05. The temperature and duration during torrefaction are crucial variables that are in line with the conclusions drawn from previous studies. GPR and ELT models effectively predict and optimize torrefied biomass quality, leading to enhanced energy density, mechanical properties, grindability, and storage stability. Additionally, they contribute to sustainable agriculture by reducing carbon emissions, improving cost-effectiveness, and aiding in the design and development of pelletizers. This optimization not only increases energy density and grindability but also enhances nutrient delivery efficiency, water retention, and reduces the carbon footprint. Consequently, these outcomes support biodiversity and promote sustainable agricultural, ecosystem, and environmental practices.