Prediction of combustion reactivity for lignocellulosic fuels by means of machine learning

Abstract

Energy demand and environmental concerns made biomass a sustainable energy source that can be used as a substitute for coal in many applications. Therefore, the combustion efficiency of biomass is a major concern for policy makers and engineers. Thermogravimetric analysis (TGA) is a robust method for determining the combustion characteristics of biomass using combustion index. TGA instruments, on the other hand, are quite expensive, and performing the experiments themselves requires a lot of time and a trained operator. Developing a method that is both faster and more reliable to obtain combustion characteristics without the use of TGA is therefore very important. In this study, a machine learning approach based on artificial neural network (ANN) was developed to predict the instantaneous combustion index defined for biomass combustion process with the help of biomass properties and combustion conditions, without using instruments and complex equations. Thus, a total of 6721 data sets were generated by using the 24 thermogravimetric experiments which conducted in this work. The Bayesian regularization optimization algorithm was used to train the developed ANN model, which is based on a multilayer perceptron architecture. The results showed that there was good agreement between the predicted and measured values of the combustion index for the training, testing, and external validation data sets. The mean absolute percentage error(MAPE), regression coefficient (R2), root-mean-square error(RMSE) for each biomass sample under the different experimental conditions were investigated, and the results were found to be satisfactory withR2 > 0.99