Modeling of Higher Heating Value Using Physico-chemical Properties of Selected Biomass Available in Bangladesh

The global shift toward renewable energy underscores the importance of biomass as a sustainable and carbon-neutral energy source. This study develops precise empirical and machine learning (ML)-based models to predict the higher heating value (HHV) of agricultural biomass in Bangladesh using their physico-chemical properties. A dataset comprising 150 biomass samples, including proximate, ultimate, and combined compositional analyses with measured HHVs, was compiled from the literature. Traditional regression methods and advanced ML algorithms were applied, and their performances were assessed using R², root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). The best-fitting regression models achieved R² values above 90%, MAPE below 7%, and RMSE within 1.2. Non-linear ML models, particularly Random Forest (RF) and Support Vector Regression (SVR), demonstrated superior accuracy, with RF yielding the best overall results across proximate (R² = 89.7%), ultimate (R² = 91.3%), and combined (R² = 92.0%) datasets. To validate the developed models, laboratory analyses were conducted on seven widely available Bangladeshi biomass types, including rice husk, maize, wheat straw, and sugarcane bagasse. Among these, coconut husk exhibited the highest HHV (20 MJ/kg), while maize recorded the lowest (12.2 MJ/kg). The proposed models consistently outperformed 50 published HHV correlations, with prediction errors below 10% for most cases. The study exhibits that ML models can accurately predict HHV from compositional data, reducing the need for extensive experimental measurements. This approach offers a cost-effective and scalable solution for biomass energy estimation, paving the way for enhanced utilization of renewable energy sources in Bangladesh and beyond.