A deep learning approach for prediction of syngas lower heating value from CFB gasifier in Aspen plus®

Abstract

Aspen Plus® is one of the practicable software for investigation of the biomass gasification characteristics. Also, artificial neural networks (ANN) as a deep learning approach are often used in the prediction of parameters such as syngas composition, lower heating value (LHV), exergy, etc. However, to our best knowledge, a universal deep learning model based on the thermodynamic equilibrium approach to predict LHV of syngas in circulating fluidized bed (CFB) gasifier is not available in literature yet. In this paper, a unique CFB gasifier model was developed in Aspen Plus® as a tool to create a total of over 1 million datasets for the training of a deep learning model that predicts the LHV of the syngas. The CFB gasifier model was found to be in agreement with the results when compared with the experimental data in the literature. 56 biomass with various elemental and proximate properties were gasified in a newly developed CFB gasifier model under different operating conditions by using sensitivity analysis in Aspen Plus®. A novel artificial neural network model, which is regularized with Levenberg-Marquardt algorithm was used as a deep learning model with a 6-12-1 tangent sigmoid architecture to predict LHV of syngas in circulating fluidized bed (CFB) gasifier, requiring minimal specificity as compared to commercial simulators require significant modelling effort and test runs. Results showed that the estimated LHV of the syngas in an agreement with the calculated values. The coefficient of determination score was calculated as R2 > 0.99 for all datasets.