Information Technology Reference
In-Depth Information
set by discretization of the data. Granulation liquid-binder spray
rate and spray pressure have been recognized as the key factors
infl uencing granule size. Other important parameters used as
secondary splitting criteria are product, and inlet and outlet air
temperature. Built decision trees have, apart from forecasting
medium and median particle sizes for new batches, also enabled
identifi cation of key processing parameters for in-line control of the
fl uidized-bed granulation process.
5.4 Evolutionary computing and
genetic algorithms
5.4.1 Introduction
The theory of evolution based on survival of the fi ttest was fi rst introduced
by Charles Darwin. The possibility of application of evolutionary
principles in machine learning and artifi cial intelligence concepts has
been researched for many years (Fogel et al., 1966; Goldberg, 1989;
Holland, 1992). Evolutionary computing has broadened into several
strategies, even though there are still discrepancies in nomenclature and
explanations. Some of these are GAs, evolution strategies, evolutionary
programming, genetic programming (GP), etc. Their similarities and
differences are discussed in more detail elsewhere (Eiben and Schoenauer,
2002).
5.4.2 Theory
GAs attempt to mimic the evolutionary process, by which biological
systems self-organize and adapt (Rowe and Roberts, 1998). GAs are
an optimization technique based on the concept of biological evolution
(Rowe and Roberts, 1998). They are usually combined with other
computational methods, or can serve as an optimizer module. Often
GAs are used to optimize ANN topology and are an example of a
stochastic algorithm capable of fi nding global minima/maxima.
Traditional optimization methods, such as gradient ascent/descent,
search in the direction of the local gradient vector, and thus can easily
Search WWH ::
Custom Search