Estimation of Several Wood Biomass Calorific Values from Their Proximate Analysis Based on Artificial Neural Networks

Abstract

The accurate estimation of the higher heating value (HHV) of wood biomass is essential to evaluating the latter's energy potential as a renewable energy material. This study proposes an Artificial Neural Network (ANN) model to predict the HHV by using proximate analysis parameters-moisture, volatile matter, ash, and fixed carbon. A dataset of 252 samples (177 for training and 75 for testing), sourced from the Phyllis database, which compiles the physicochemical properties of lignocellulosic biomass and related feedstocks, was used for model development. Various ANN architectures were explored, including one to three hidden layers with 1 to 20 neurons per layer. The best performance was achieved with the 4-11-11-11-1 architecture trained using the backpropagation algorithm, yielding an adjusted R2 of 0.967 with low mean absolute error (MAE) and root mean squared error (RMSE) values. A graphical user interface (GUI) was developed for real-time HHV prediction across diverse wood types. Furthermore, the model's performance was benchmarked against 26 existing empirical and statistical models, and it outperformed them in terms of accuracy and generalization. This ANN-based tool offers a robust and accessible solution for carbon utilization strategies and the development of new energy storage material.