Kinetic models and distribution of activation energy in complex systems using Hopfield Neural Network

Abstract The knowledge about the thermal behavior of solids is crucial to chemically characterize the products, especially when they are applied in industries as solid fuels. The kinetic studies allow us to accurately predict the thermal processes and are of great value for industries. In complex systems, more than one reaction occurs during the thermal process and the activation energy should be considered as a distribution of the activation energy model (DAEM), or could be determined from the combination of kinetic models, instead of an average parameter for the entire event. The DAEM is equivalent to an ill-conditioned inverse problem and in this work, the Hopfield neural network (HNN) is proposed. Firstly, the HNN algorithm was investigated using simulated data, and after it was tested using experimental data of a Brazilian petroleum coke sample. The HNN results were confronted with the results from the Levenberg-Marquardt algorithm, considering the DAEM and kinetic models determination. The HNN results showed this methodology is more efficient to treat this kind of problem and additionally it was compared with isoconversional methods.