Estimation of Biomass Fuels' HHVs Based on Ultimate and Proximate Analysis and Their Combination Data Using MLP-ANN Models

The most important thing to know when investigating the feasibility of energy generation from biomass materials is the higher heating values (HHVs). 12 biochars were obtained from zeyrek pulp by hydrothermal carbonization method. Fuel properties (proximate, ultimate and calorific value) and structural properties (by IR spectroscopy) of the obtained biochars were determined. To predict HHVs of biomass, the multi-layer perceptron artificial neural network (MLP-ANN) technique is used. For this purpose, 66 real data points were extracted from both our data and reliable references for the model's training and validation. Based on input data from the proximate analysis, ultimate analysis and combined proximate-ultimate analysis, three different MLP-ANN models were developed. The prediction accuracies of these models were compared statistically to the experimental data. MLP-ANN models have been shown to predict the HHV of biomass with high accuracy. The performance of the MLP-ANN models was also evaluated and it was discovered that the combined proximate-ultimate analysis based MLP-ANN model providing the best results such as coefficient of correlation (R-2), root mean square error (RMSE) and mean absolute percentage error (MAPE).