Predicción de potencialidades energéticas a partir de los residuos sólidos generados en el mercado municipal del cantón Caluma mediante el uso de redes neuronales artificiales

Abstract

The objective of this study was to predict the energy potentialities of organic solid waste using artificial neural networks. The experimental data were obtained from the thermogravimetric analysis of the organic residues, data that were evaluated using kinetic models Flynn-Wall-Ozawa, Kissinger-Akahira-Sunose (KAS) and Friedman. Subsequently, a statistical analysis was performed in which the kinetic model with the lowest data dispersion was selected for the ramp of 5 ° C / min and 15 ° C / min. The neural network was designed with Matlab in which three input variables were defined time, temperature and weight from thermogravimetric analysis and activation energy as output variable. For the heating rate of 5 ° C / min and 15 ° C / min, the network architecture has 25 neurons in the hidden layer and the Bayesian Regularization algorithm because it obtained the lowest root mean square error (MSE) with 5.25464E-1 and the R closest to 1 equivalent to 9.99997E-1. The results predicted by the artificial neural network were validated by a statistical analysis performed in Statgraphics, resulting in a P-value greater than 0.05 with a confidence level of 95% concerning the real and predicted activation energy. In conclusion, the use of neural networks for the prediction of energy potentialities prevents us from carrying out extensive calculations or long hours in laboratories. These results may be used to carry out projects that aim to generate clean energy.