TY - JOUR
TI - The possibility of modeling agricultural biomass ash by neural networks concerning proximate analysis inputs
AU - Matin Ana
AU - Špelić Karlo
AU - Jurišić Vanja
AU - Matin Božidar
AU - Grubor Mateja
AU - Tomić Ivana
AU - Majdak Tugomir
AU - Brandić Ivan
JN - Thermal Science
PY - 2024
VL - 28
IS - 6
SP - 4771
EP - 4780
PT - Article
AB - Agricultural biomass is an important RES with significant environmental and eco¬nomic benefits. However, high ash content in biomass can lead to problems such as slagging, fouling, and corrosion and can reduce the efficiency of energy systems. This study analyzes the proximate composition of different biomass samples, focusing on ash content, and uses machine learning to model ash content based on other components. Six biomass types, including rapeseed, barley, wheat, corn, soybean and sunflower, were examined to analyze the content of coke, fixed carbon, volatile matter, and ash. The results showed considerable variability, with ash content ranging from 8.25% for rapeseed to 12.3% for soybean. Artificial neural networks were used to model ash content with a high accuracy of R² = 0.92. The model effectively estimated the ash content based on the input parameters and demonstrated the potential of machine learning to optimize biomass selection for energy production.