Bayesian optimized multilayer perceptron neural network modelling of biochar and syngas production from pyrolysis of biomass-derived wastes

Abstract

Biochar and syngas are important products of pyrolysis that can be employed for a wide range of applications such as catalysts for biodiesel production, wastewater treatment, and the production of oxygenated fuel. This study employs Bayesian optimized multilayer perceptron neural network for modelling the prediction of biochar and syngas from pyrolysis of biomass-derived wastes. Sixty neural networks were configured by considering the effect of hyperparameters such as the connecting layers of the network, the size of the network, and the type of neural network algorithms. The feature analysis using F-test algorithms revealed that temperature, biomass composition, N2 flow rates, residence time, and bed height influence the biochar and syngas yield obtained from the pyrolysis process. There is a significant interaction effect between the features as shown by the parametric analysis. The performance of the neural networks was significantly influenced by the number of connecting layers and the size of the hidden neurons. The five-layer neural network with an architecture of 3–2-10–10-1 displayed the best performance in predicting the biochar yield obtained from the pyrolysis process as indicated by R2 of 0.984, and RMSE of 0.34. While the five-layer neural network with an architecture of 3–7-10–3-1 displayed the best performance in predicting the syngas yield from the pyrolysis process as indicated by R2 of 0.999.