Predicting bio-oil yield obtained from lignocellulosic biomass pyrolysis using artificial neural networks

The prediction of bio-oil yield from pyrolysis of lignocellulosic biomass is important for processes optimization, modeling, and installation's design. This work models the bio-oil of lignocellulosic biomass yield by Artificial Neural Networks (ANN), being the inputs: cellulose, hemicellulose, lignin, pyrolysis temperature, heating rate, N-2 flow rate, and particle size. A database was created with 34 biomass types, modeling with 329 samples, training with 80%, and validating with 20%. A previous stage of screening was carried out with 100% of data for choosing the algorithm of the second phase and the number of neurons in the hidden layer; the selection criteria were the mean absolute error (MAE) and the correlation coefficient. The best performance was for backpropagation/Levenberg-Marquardt with 7:13:1 as ANN architecture. All ANN with less than 1% of MAE were tested for validating and the weight's matrix of the best one is shown. The selected network with a correlation coefficient of 0.9739 and MAE of 1.7159% for validation, only had four outlier values between 5 and 6%, the remaining 62 samples with all 263 used in training, had less than 5% of difference compared to the experimental values, thus representing a very accurate model for predicting bio-oil yield.

Automated summary

This study used Artificial Neural Networks to model the bio-oil yield from pyrolysis of lignocellulosic biomass and obtained an accurate prediction model through validation.