Modelagem da previsão do rendimento do bio-óleo com base nos parâmetros de processo de pirólise e composição centesimal, elementar e lignocelulósica da biomassa em diferentes reatores usados na indústria química

Abstract

The bio-oil yield, the liquid product of pyrolysis, is affected by biomass composition and process parameters. This research applies different multivariate regression methods, linear and non-linear, to predict bio-oil production, exploring various combinations of predictors in reactors operating in continuous and batch mode in biomass pyrolysis. A database with experimental data from 220 articles in the literature was developed for the modeling. Modeling was performed on eight case studies with different variables. Process parameters analyzed were heating rate, temperature, reaction time, particle size, and vapor residence time. The variables related to the composition of the analyzed biomass were elemental composition, proximate composition, and concentration of lignocellulosic materials. Fourteen multivariate regression models were applied to model the bio-oil yield. The Cubist model provided the best bio-oil prediction results for the batch reactor with RMSE of 0.92%, MAE of 0.73%, and R2 of 0.99. This best result was obtained using, as predictors, all the variables studied and only data with complete observations. The best result for the continuous reactor was obtained with the Extremely Randomized Trees model and a full set of predictors with complete observations. The performance metrics obtained were an RMSE of 2.15%, MAE of 1.75% and R2 of 0.96. Both models showed excellent performance for bio-oil yield prediction for batch and continuous reactors. The modeling was also carried out using the imputation of missing data by k-NN and multiple imputation; the results showed no gain in the prediction efficiency of the models using the data sets containing the imputed values. The particle swarm method was used, in the best models obtained in modeling, to study the optimization of process parameters for rice husk pyrolysis. For the batch reactor, the maximum predicted bio-oil yield was 52.9% at a temperature of 605 ºC, particle size of 7.4 mm, vapor residence time of 1400s and heating rate of 300 ºC/min. For a continuous reactor, the maximum yield predicted by the model was 52.1% at a temperature of 513 ºC, particle size of 0.41 mm and vapor residence time of 0.6 s.