Fitting performance of artificial neural networks and empirical correlations to estimate higher heating values of biomass

Biomass residue is one of the raw materials of most interest as a source of renewable energy. There are three major types of biomasses to obtain energy: lipids, sugars/starches and cellulose/lignocellulose. The estimate of the gross calorific value (PCS), whose determination methods require long periods and are relatively expensive, is crucial in the analysis and development of bioenergy systems. There are empirical correlations in the literature for higher heating value (HHV) determination based on both elemental analysis data (more demanding in terms of instrumentation) and proximate analysis data (simpler and easier to achieve experimentally). This study evaluates the feasibility of using artificial neural networks (ANN) and empirical correlations (Matlab 2013) to fit and estimate the gross calorific value of biomass from proximate analysis databases available in the literature. Matlab eases the development of ANNs, as provides easy programming and many functions that can be directly used. Starting from a database of 100 records and raising then the database to 225 and thereafter to 350, it was possible to analyze the differences between basic fitting characteristics of ANNs and correlation models. Twelve HHV values of biomass available in the literature were used to verify the validity of the fittings.