Modelling Biomass Elemental Composition: A Neurofuzzy Approach

Due to growing concerns about energy security and environmental impacts, biomass has been considered as a viable renewable energy source. However, the development of biomass combustion systems that can process biomass from different sources requires an accurate understanding and prediction of the major properties of biomass. Although adaptive neuro-fuzzy inference systems (ANFIS) have been deployed in the prediction of elemental composition, the impact of clustering technique has not been reported despite its significance in optimal model development. In this study, the effect of clustering technique on the performance of standalone ANFIS in the prediction of elemental composition of biomass was evaluated. The minimum proximate parameters which are fixed carbon, ash and volatile matter were considered as the input. Two clustering methods called grid partitioning (GP) and fuzzy c-means (FCM) were compared to generate an optimal fuzzy inference system (FIS) structure. The results show that FCM based ANFIS model performed better in biomass elemental composition prediction. Its performance metrics based on Root Mean Squared Error (RMSE), Mean Absolute Deviation (MAD), Mean Absolute Percentage Error (MAPE), Coefficient of Correlation (CC), Log of accuracy (LAR) and Variable accounted for (VAF) are 0.6856, 0.419, 9.5603, 0.764, 0.001, 0.583 at an optimal computation time (CT) of 5.48s for Hydrogen content; 3.894, 2.442, 5.001, 0.898,0.001, 0.805 at an optimal computation time (CT) 5.55 secs for carbon; 4.381, 2.787, 10.020, 0.895, 0.002, 0.801 at an optimal computation time (CT) 5.95 secs for Oxygen respectively. This technique could be utilized as a quick tool for a reliable and accurate approximation of elemental composition of biomass feedstock.