Review and Improvement of Several Optimal Intelligent Pitch Controllers and Estimator of WECS via Artificial Intelligent Approaches - Complex System Modelling and Control Through Intelligent Soft Computations - page 511

These concepts are as follows:

1. Black box, we can handle an ANN as a black box and requiring considering its
building and weight values. Input vectors are mapped by the ANN to quality
vectors and algorithm extracts rules by analyzing the link between input and
output.
2. Dissolution, some algorithms try to collect around corresponding weights to
simplify the extracted rules. these kinds of algorithms take a look at the weights
and search for study patch through the net.

In one of the express activities of RE from neural net, variables have read from a
network. This diversely is not qualify by the network traditional connections
structure (Luo and Unbehaben 1998 ). Also some algorithms have been developed
to extract deterministic

finite-state automata DFA from frequent neural network
(Omlin et al. 1992 ; Giles et al. 1992 ; Giles and Omlin 1993 ).
A little while back some researchers even have aim to generate suppression rules
from neural regressors (Saito and Nakano 2002 ; Setiono 2002 ).

8 Rule Extraction from AI

Neural network rule extract approach attempt to open the NN black box generate
representative rules NN. One of the important catch in some application of NN is
the annoyance with take in system. Therefore extracting comprehension the method
accordingly extracting knowledge from NN in an extensive way has born originated
(Darbari 2000 ; Mitra and Hayashi 2000 ; Santos et al. 2000 ; Zhou et al. 2000 ).
Commonly, it has the form of hypothesis rules many rules extraction approached
that advanced in last few years. A new rule extraction method supported on MLP
NN and GFS optimization has been conferred in this part. It caused called
FRENGA.
The meta-heuristic explore method extracts various rule from MLP network
using GA, for searching optimal solution in enormous space of possible solutions
(Kasiri et al. 2011a , 2012a ).
In this part o NN characterization in classi cation questions by a new genetic

fuzzy algorithm has been presented by proposed method FRENGA completely uses
from investigational data in veri

able article. NN has been trained by this specu-
lative data. That being so this method uses the NN results in de nitional of Fitness

Function (FF). In conclusion GFS has been trained with this FF to extract GFS
fuzzy set rule. Reactions of these processes are given in Table 2 . Table 2 encom-
passes extracted five rules from NN using GFS. These rules set pitch angle in the

best setting to optimally control wind turbine. It can be simply noticed that the set
of if-then rules be in need of cycle, which is composed of

five steps as follows

(Hisao et al. 1999 ).