A study to predict pyrolytic behaviors of refuse-derived fuel (RDF): Artificial neural network application

The present study demonstrates the thermal behaviors of refuse-derived fuel (RDF), a highly heterogeneous fuel, at high temperature region by bringing experimental and modelling studies together. In the first part, RDF was pyrolyzed in thermal analyzer from room temperature to 900 degrees C at varying heating rates as well as the evolved gas analysis was monitored by using TG-FTIR-MS. Afterwards, obtained data was used to develop an artificial neural network (ANN) model that can predict thermal behaviors of RDF at a new heating rate without performing any experiments. The temperature and heating rate were selected as input parameters while temperature dependent weight loss was selected as output parameter. The effects of parameters such as neuron number, training number, and the transfer function type on the network performance were investigated in detail to optimize network topology. Optimization studies showed that the best performance was achieved with ANN that had 7-6 neurons trained 25 times with tansig-logsig non-linear function combination. Prediction performance of the optimized ANN was tested by introducing a new experimental dataset. The good agreement between experimental and predicted values revealed that ANN can be a promising tool in pyrolytic behaviors estimation of even heterogeneous fuels such as RDF. (C) 2016 Elsevier B.V. All rights reserved.