Artificial Intelligence: Summary and Hybrid Schemes - Digital Signal Processing in Power System Protection and Control - page 305

• Combine at least two intelligent techniques,
• Are highly capable of reasoning and learning in an uncertain and imprecise
environment,
• Allow for gaining greater tractability, robustness and lower the cost of solutions.
Applying in one system such techniques as FL, ANN, GA and ES is meant as
complementary rather than competitive. The virtues of particular techniques:
• Dealing with imprecision and uncertainty (FL),
• Possibility of learning and curve fitting (ANN),
• Search and optimization (GA),
• Decision-making in face of many often different in nature input signals (ES) are
merged together to provide an optimal solution for given problem.
Any combination of AI techniques is conceivable; the examples may
include fuzzy-expert systems, neural-expert hybrids, fuzzy-neural networks, fuzzy
controlled genetic algorithms, etc. The literature study confirms that neuro-fuzzy
hybridization is the most visible integration realized so far.
A hybrid intelligent system can be good or bad—it depends on components
which constitute the hybrid. Metaphorically one can say that a good hybrid would
be ''British Police, German Mechanics, French Cuisine, Swiss Banking, Italian
Love and Polish Hospitality'', but ''British Cuisine, German Love, French
Mechanics, Italian Banking, Polish Police and Swiss Hospitality'' would be a bad
one. Therefore careful analyses are needed to select the right components for
building a good hybrid system for the application at hand.
Generally, the hybrid schemes proposed are either fused (melted) schemes or
cooperative hybrids. The two classes of schemes are outlined below with some
more detail.
15.2.1 Fused Hybrid Schemes
The first category of hybrid schemes include those where one of the schemes is
melted into another one to realize certain partial function or where given AI
technique is represented by another one in order to exploit some virtues of this
technique (e.g. training of a fuzzy scheme).
In the paper [ 19 ] a fuzzy version of the neural network model called Fuzzy
Self-Organizing Map was introduced. The neurons of the original ANN model
were replaced by fuzzy rules, which are composed of fuzzy sets. The fuzzy sets
defined an area in the input space, where each fuzzy rule fires. The output of each
rule was a singleton. The outputs were aggregated by a weighted average, with
the firing strengths of the fuzzy rules acting as the weights. Thus the Fuzzy
Self-Organizing Map performed a mapping from a n-dimensional input space to
one-dimensional output space. The learning capability of the Fuzzy Self-
Organizing Map enabled it to model a continuous valued function to an arbitrary