Aplicação de redes neurais artificiais na análise de dados cinéticos de processos de gaseificação

Abstract

The world is becoming more and more aware of the need to use clean sources of energy, especially in the current conjuncture, where the geopolitical crisis as the supply of oil has become more limited in the world. One of the alternative sources of energy are the processes that make use of residual organic matter. Since biomass is a renewable raw material and can be applied in energy generation processes, such as biomass gasification, which has emerged as a promising source of energy generation. The big problem with the process is that many biomasses may not be suitable for it and end up with low efficiency. To avoid this, previous studies of organic raw materials are needed in order to provide a prediction of their effectiveness for the generation of energy in gasification. These studies are often delayed and, to try to circumvent this objective problem, we used this work to predict the kinetic parameters of activation energy and pre-exponential factor for CO2 gasification of different biomasses using artificial neural networks (ANNs). These variations can help in the analysis of the feasibility of using biomass in gasification processes, without the need for costly studies. An artificial neural network was coded and tests were performed in the same way, to try to obtain an ANN that could predict this kinetic data being fed with more easily obtained data. The structures of Feed-Forward, Cascade-Forward and Elman network RNAs were used, fed with data of percentages of two main chemical elements present in the biomass, percentages of zinc, volatile matter and fixed carbon and other data that are considered relevant for a performing two tests. When the initial results were not satisfactory, there were several variations in the methodology to correct the problem. However, these variations did not completely correct the problem, so I noticed that the tests performed showed a tendency towards overfitting, so suggestions were given to improve the network and improve it. The results of the work based on the obtained data may not be useful in further work for the development of a viable model for the analysis of gasification with ANNs.