Predicción del comportamiento termogravimétrico de la energía de activación de los residuos cáscara de papa (Solanum tuberosum)

Abstract

The objective of this study was to develop an artificial neural network capable of predicting the thermogravimetric behaviour of the activation energy in the residues of potato peel (Solanum tuberosum) from the Guano canton. The experimental data of the thermogravimetric analysis and calculations using the kinetic models proposed by Flynn-Wall-Ozawa, Kissinger-Akahira-Sunose and Friedman for the activation energy have been used when establishing a base of 100 data in the prediction model. The development of the network was carried out in the Matlab software with three input variables, the different number of neurons in the hidden layer, an output variable and the Levenberg Marquardt, Bayesian Regularization and Scaled Conjugate Gradient training algorithms. The time, temperature and weight of the sample from the thermogravimetric analysis have been selected as input variables, while the activation energy calculated by the kinetic method of Flynn Wall Ozawa as output variables. A Pearson correlation coefficient of 1 and a mean square error of 2.327E-09 show the good performance of the network during its training with 375 neurons in the hidden layer and the Bayesian regularization algorithm. A P-value greater than 0.05 allows the prediction model to be validated with 95% confidence, statistically confirming that there is no significant difference between the real and predicted activation energy by the artificial neural network. The results indicate that the artificial neural network proves to be efficient in predicting the activation energy of potato peel residues analysed by thermogravimetry using heating rates of 5 and 15 ° C / min in an inert atmosphere with nitrogen injection at 20 mL/min. It is recommended to use the prediction model in those projects aimed at pyrolysis or gasification of biomass made up of potato peel residues.