Information Technology Reference
In-Depth Information
8.2.1 Inference from Imprecise Rules
If a dynamic systems is considered, with input variables (
x
1
,...,
x
n
) and output
variable
y
, whose behavior is known by
m
imprecise rules
R
i
:
R
1
:
If
x
1
is
P
11
and
x
2
is
P
12
···
and
x
n
is
P
1
n
,
then
y
is
Q
1
R
2
:
If
x
1
is
P
21
and
x
2
is
P
22
···
and
x
n
is
P
2
n
,
then
y
is
Q
2
.
R
m
:
If
x
1
is
P
m
1
and
x
2
is
P
m
2
···
and
x
n
is
P
mn
,
then
y
is
Q
m
The question is the following: If we observe that the input variables are in the
“states” “
x
1
is
P
1
”, “
x
2
is
P
2
”,
,“
x
n
is
P
n
”, respectively, what can be inferred
for variable
y
? That is, supposing that it lies in the “state” “
y
is
Q
∗
”, what is
Q
∗
?
Without loss of generality let us consider only the case
n
...
=
m
=
2:
R
1
:If
x
1
is
P
11
and
x
2
is
P
12
,then
y
is
Q
1
R
2
:If
x
1
is
P
21
and
x
2
is
P
22
,then
y
is
Q
2
x
1
is
P
1
and
x
2
is
P
2
y
is
Q
,whatis
Q
?
where
x
1
∈
X
1
,
x
2
∈
∈
Y
, and let us represent it in fuzzy logic by means of
the functions
R
1
and
R
2
. This leads to:
X
2
,
y
for convenient continuous t-norms
T
1
,
T
2
and convenient implication functions
J
1
,
J
2
. Convenient, in the sense of adequate to the use of the conditional phrases
R
1
and
R
2
relative to the problem under consideration within the given system
S
.
To find a solution to this problem, the problem of linguistic control, that is a part
of what can be called intelligent systems control, we need to pass throughout several
steps.
First Step: Functions
J
Implication functions
J
are obtained through the interpre-
tation and representation of the rule's use, that is, from the actual meaning of these
conditional phrases.Thus, different types of protoform exist
:[
0
,
1
]×[
0
,
1
]ₒ[
0
,
1
]
a
ₒ
b
⃔
J
(
a
,
b
)
and are used as introduced previously in the topic. For example:
Search WWH ::
Custom Search