Utilisation of machine learning algorithms for the prediction of syngas composition from biomass bio-oil steam reforming

The aim of this study was to utilise artificial neural network (ANN) and AdaBoost (AB) algorithms to model the synthesis gas composition from the steam reforming of biomass bio-oil. At testing on training data, it was observed that R-2 > 0.999 was achieved for both algorithms for all product selectivity indicating a 99.9% capture of data variability. Also, the RMSE values were <0.007 in most cases. The MAE values were <0.005 in most cases. The ANN predictions were observed to be more accurate than AB predictions for the current application. On the other hand, considering stratified 10-fold cross-validation the proposed models present R-2 > 0.9 using AB considering hydrogen and carbon dioxide, and using ANN considering methane and carbon monoxide.

Automated summary

This study utilized artificial neural network and AdaBoost algorithms to model the synthesis gas composition from the steam reforming of biomass bio-oil. The results indicated that the ANN predictions were more accurate than AB predictions for the current application.