A combined genetic algorithm and least squares fitting procedure for the estimation of the kinetic parameters of the pyrolysis of agricultural residues

The main objective of this study is to implement a two-step fitting procedure to estimate the kinetic parameters of two distinct pyrolysis models to study the pyrolysis of pine bark, wheat straw and rice husk. Thermogravimetric curves were obtained for the three biomass fuels for heating rates of 5, 10 and 15 K/min in an inert atmosphere of Argon to investigate the impact of the type of biomass in the pyrolysis behavior under different heating conditions. Distinctive thermogravimetric and differential thermogravimetric curves were obtained owing to the different composition of the biomass fuels. For the conditions examined, the impact of the heating rate on the profile curves was marginal. In order to better understand the impact of the biomass composition in the pyrolysis, their main components were estimated. Additionally, a two-step algorithm was used to calibrate the global kinetic parameters of a single reaction model and of a three parallel reaction model, based on the fitting of predicted curves to the experimental ones. The first step was a genetic algorithm procedure. An evaluation function that minimizes the deviation between the experimental and predicted pyrolysis yields, while preserving the characteristics of the mass decomposition during pyrolysis, is presented in this work. The second step was a least squares minimization that was used for further refining the solution obtained in the first step. The method showed excellent repeatability. For each biomass fuel, all heating rates were globally fitted, with errors of the order of ∼5% for the single reaction model and of less than 1.6% for the three parallel reaction model. The activation energies obtained by fitting each model to the experimental data are generally within the values reported in the literature. Finally, a sensitivity analysis showed that the variation of the composition of each biomass used in this study does not affect significantly the predictions of the three parallel model.