State of the Art and Development Trend - Computational Intelligence in Time Series Forecasting - page 325

engineers. Here, it should be recalled that the learning rule of Hebb, himself a
neurophysiologist, was formulated after his study of the learning principle of
neurons.
A modest step in this direction is the special issue of Control Systems
Magazine , August 1994, devoted to biological networks and cell regulation.
In the area of fuzzy logic technology , the intensive research trend towards
knowledge extraction from data or data understanding using rule-based systems
(Duch et al. , 2004) is remarkable. The beginning of this research has roots in the
achievements in image interpretation using the methods of artificial intelligence.
The aim behind this was to explain the meaning of the images from the collected
data, mainly using
x perceptual knowledge , which supports interpretation in terms of lines,
patterns, areas, etc .
x semantic knowledge , which enables the use of some abstract concepts like
the object shapes, the relationships between the objects, etc .
x functional knowledge , i.e. the problem-oriented knowledge that finds out
the best image interpretation by conducting intelligently the inference
process.
Thus far, the rule-based data understanding approach has chiefly been used for
data-based medical diagnostics, like for diagnosis of cancer and diabetes diseases
(Setiono, 2000; Mertz and Murphy, 1993). However, it is highly possible that the
approach can also find application in the production industry for material analysis,
product quality inspection, and for production performance identification. What is
interesting from our point of view is that rule-based data understanding can help to
elucidate more inherent knowledge from the time series to be analyzed in this way
than using the approaches described in previous chapters.
In the area of genetic algorithms , the trend toward development of new, more
advanced search algorithms is noteworthy. The most prominent example represents
the development trend in particle swarm optimization , invented by Kennedy and
Eberhart (1995), i.e. by a social psychologist and an electrical engineer. It is a
population-based search approach placed somewhere between genetic algorithms
and evolutionary programming because it works in the following way:
x Each particle (representing a potential problem solution) keeps track of its
coordinates in the problem space related with the best solution, called
pbest , based on the fitness obtained thus far. By evaluation of pbest values
across the particle's population the global best temporary solution, called
gbest , is found and the parameter's adjustments are performed. This, in
principle, corresponds to the crossover operation of GAs.
x Like evolutionary programming, the particle swarm concept also relies on
the stochastic processes within the population.
The advantage of the particle swarm algorithm, compared with both of its
precursors, i.e. with genetic algorithms and evolutionary programming, lies in the
programming simplicity, which is due to the simplicity of its underlying concept.