Comparison of Artificial Neural Networks and kinetic inverse modeling to predict biomass pyrolysis behavior

Abstract

The prediction of biomass pyrolysis behavior has received immense research attention, which has challenged scientists as well as industrialists. Currently, the main predictive tool is inverse modeling which is based on a specific kinetic scheme. However, real biomass pyrolysis entails a complex reaction mechanism, the appropriate kinetic scheme is being explored. Because pyrolysis prediction is closely related to the accuracy of the proposed kinetic scheme, whether the prediction can be achieved without prior knowledge of the kinetic scheme is intriguing. Therefore, Artificial neural networks (e.g., back propagation neural network and Elman neural network) are applied for prediction based on unknown kinetic schemes. Specifically, Elman neural network, which is an uncommon biomass pyrolysis prediction method, has been applied. Moreover, the neural network structure that was subjected to different data sets was optimized to establish the optimal neuron number in the hidden layer, transfer function, and training function, which were based on thermogravimetric data over a wide heating rate range. Eventually, the predicted results agreed well with the experimental data. Moreover, compared with kinetic inverse modeling coupled with a common heuristic algorithm known as particle swarm optimization, our current ANNs results, which were based on the unknown kinetic scheme exhibited superior prediction ability, especially in the shoulder and peak regions of mass loss rate curves.