CFD-based reduced-order modeling of fluidized-bed biomass fast pyrolysis using artificial neural network

Abstract

In order to reduce the computational effort of design and optimization for biomass fast pyrolysis reactor, the reduced-order modeling technology was applied to develop reduced-order models (ROMs) based on the CFD data from multi-fluid model (MFM) simulation of biomass fast pyrolysis in a bubbling fluidized bed reactor. The CFD simulations at nine different pyrolysis temperatures were performed, and the product yields and the influence of temperature on product yields were in a good agreement with experiments, which fully validated the CFD approach. The back-propagation (BP) artificial neural network (ANN) was used to map the species mass fraction data of CFD simulation to pyrolysis temperature and coordinates of each computational node in the reactor. The number of neurons and active function in the ANN was optimized. The ability of the developed ROMs to predict the species distributions at both training and testing temperature was investigated. The influence of sample method and number of outputs was also studied.