A Study On Modeling and Operational Optimization of Biomass Gasification Processes Using Neural Networks

Monday, November 8, 2010: 2:23 PM
Grand Ballroom H (Marriott Downtown)
Maurício B. De Souza Jr1, Amaro G. Barreto Jr1, Leonardo C. Nemer1, Patrícia O. Soares2 and Cristina P. B. Quitete2, (1)Chemical Engineering, Federal University of Rio de Janeiro, Rio de Janeiro, Brazil, (2)Cenpes, PETROBRAS, Rio de Janeiro, Brazil

Gasification is a promising technology for energy production based on complex thermochemical processes, at high temperatures. Carbonaceous solid material is converted in fuel gases, volatile compounds, carbon and ashes through gasification. Particularly, biomass is transformed in permanent gases as hydrogen, carbon monoxide, carbon dioxide and methane, together with organic vapors that condensate under ambient conditions and are collectively known as tar, and a solid residual consisting of char and ashes. Modeling for simulation and prediction of performance of biomass gasification processes is still an incipient activity, due to the inherent complexity of these processes. The development of a phenomenological model in this case requires that many idealizations and assumptions are made, resulting in a very simplified model, with little predictive capability. In this context, neural network models provide an interesting alternative as they are universal approximators with the ability to learn directly from input-output data. Additionally, a large body of literature on biomass gasification report experimental data consisting of characterization of biomass and gaseous products and operational conditions of different gasifiers. The target of the present work was to develop neural network based models that correlated the yields of the gases produced with both the characteristics of the biomass and the operational conditions of the gasifier. With that purpose, multilayer perceptrons were trained and validated using data from the literature for different biomasses and for circulating and bubbling fluidized bed gasifiers. High correlations values (ranging from 0.94 to 0.99) between predicted and observed values were obtained and sensitivity techniques were used in order to evaluate and discard input variables. The neural network models were further used together with the particle swarm optimization technique (PSO) to calculate the conditions that allowed the maximization in the yield of a desired gas. The results obtained indicate that the developed approach provides a valuable tool to help in the efficient design, operation and control of fluidized bed gasifiers.

Extended Abstract: File Uploaded
See more of this Session: Biomass Gasification
See more of this Group/Topical: Fuels and Petrochemicals Division