Tar prediction in bubbling fluidized bed gasification through artificial neural networks

Abstract

Tars are one of the main barriers for the implementation of biomass gasification at industrial scale. Among the considerable number of models to predict gas composition, there is a lack of models predicting tar generation in gasification processes, as tar concentration data is far more difficult to collect and analyze. This study makes use of artificial neural networks (ANNs) to predict tar generation in gasification processes carried out in lab-scale bubbling fluidized bed reactors operating with silica sand and woody biomass. An exhaustive review of the existing literature and the different tar collection and analysis methods is conducted to create a consistent database for the ANNs to train on. The model integrates different tar data coming from different sampling protocols and analysis methods (tar protocol and gas chromatography, tar protocol and gravimetric method, and solid phase absorption). The predicted results show good accuracy (R2 > 0.97), concluding this generalized predictive novel model is a useful tool for tar prediction in gasification. The model results are in agreement with the literature, verifying how tar content in the product gas behaves when equivalence ratio (ER) and temperature are varied. The predicted versus experimental values are also compared with previous models for tar prediction. ANN modelling shows a higher accuracy than other models, demonstrating this data-driven modelling can be a good approach for tar content prediction.