Page 262 - Biomass Gasification, Pyrolysis And Torrefaction Practical Design and Theory

238 Biomass Gasification, Pyrolysis and Torrefaction

7.5.2.6 Reactor Hydrodynamics
The kinetic model considers the physical mixing process and therefore requires
knowledge of reactor hydrodynamics. The hydrodynamics may be defined in
terms of the following types with increasing sophistication and accuracy:
Zero dimensional (stirred tank reactor)
One dimensional (plug flow)
Two dimensional
Three dimensional
Unlike other models, the kinetic model is sensitive to the gas solid con-
tacting process involved in the gasifier. Based on this process, the model
may be divided into three groups: (i) moving or fixed bed, (ii) fluidized bed,
and (iii) entrained flow. Short descriptions of these are given in Section 7.6.

7.5.2.7 Neural Network Models
An alternative to the sophisticated modeling of a complex process, especially
for one not well understood, is an ANN. An ANN model mimics the working
of the human brain and provides some human characteristics in solving mod-
els (Abdulsalam, 2005). It cannot produce an analytical solution, but it can
give numerical results. This technique has been used with reasonable success
to predict gas yield and composition from gasification of bagasse, cotton
stem, pine sawdust, and poplar in fluidized beds (Guo et al., 1997); in MSW;
and also in a fluidized bed (Xiao et al., 2009).
The ANN model can deal with complex gasification problems. It uses a
high-speed architecture of three hidden layers of neurons (Kalogirou, 2001):
one to receive the input(s), one to process them, and one to deliver output(s).
Figure 7.9 shows the arrangement of neuron layers and the connection pat-
terns between them. Kalogirou (2001) suggested the following empirical for-
mula to estimate the number of hidden neurons:
1
Number of hidden neurons 5 ðinputs 1 outputsÞ
2
(7.78)
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
1 number of training patterns

Input layer Hidden layer Output layer

FIGURE 7.9 Schematic diagram of a multilayer feed-forward neural network (Source: Kalogirou,
2001).