Академический Документы
Профессиональный Документы
Культура Документы
..
,
e
<
,
0
0 '-'
.. , ..
TImals)
,
3.5
., ,
""""""'"
__
"
\.
"
..
Tlme{$j
Figure 6. Output and control signals for crisp PD control system
45
r-
c
.9
Input
,- "
(Crisp)
."
0
.c
-
~
'-
Inference
mechanism
f _
Rule-base
r-
0
:;
,
Output
(Crisp)
"
"
u
Q
'-
Figure 7. Fuzzy Logic controller
Implementation of an FLC requires the choice of four key factors (Mamdani, 1977):
number of fuzzy sets that constitute linguistic variables, mapping of the measurements
onto the support sets, control protocol (hat determines the controller behaviour and
shape of membership functions. Thus, FLCs can be tuned not just by adj usting
controller parameters but also by changing control rules, membership functions etc.
Rule base, inference mechanism and defuzzification methods are the sources of
nonlinearities in FLCs. But it's possible to construct a rule base with linear input-output
characteristics. For an FLC to become a linear controller with a control signal
U = E + CE where E is "error" and CE is "change of error", some conditions must be
satisfied (Jantzen, 2007):
I . Support sets of input linguistic variables must be large enough so that input
values stay in limits.
2. Linguistic values must consist of syrnmeTric triangular fuzzy sets that intercept
with neighbouring sets at a membership value of J.l = 0.5 so that for any time
instant, membership values add to t .
3. Rule base must consist of A-combinations of all fuzzy sets.
4. Outpullingui stic variables must consist ofsinglelon fuzzy sets (Si.l ) positioned
at the sum of the peak positions of input fuzzy sets.
5. A should be multiplication and defuzzification method mus[ be "centre of
gravity" (COGS).
4.1 FPD controiler design
Figure 8 shows an FPD controller that acts on the same signals with a PD controller but
the control strategy is constructed as fuzzy rules (Jantzen, 2007) .
6
.e ~ ~ E
u U
~ - -
GU FLC
de/dt ~ CE
~ G C E > __I
Figure 8. FPO controller
Control sjgnaJ U(n) is a nonlinear function of "error" and "change of error". Thus,
U(n) = f(GE x e(n). GeE x e(n)) x GU (20)
where f represents the control algorithm. A linear approximation should be obtained
with a suitable choice:
I(GE x e(n). GeE x e(n)) '" GE x e(n) + GeE x e(n) (21)
Then
U(n) = (GE x e(n) + GeE x e(n))x GU (22)
GeE )
U(n) = GE x GU x e(n) + GE x e(n)
(23)
(
When we compare thi s equation with the control signal of a crisp PD controller. the
relationship between ga ins of a PO conrrolJer and of an fPD controller is:
GE x GU ::; Kp (24)
GeE
(25)
GE = TD
Consequeml y. parameter values of a linear FPO controller may be detemlined from a
tuned PD controller.
Figure 9 shows the contro l system wi th an FPD controller.
47
. ( 1.,-'(l)
K, I
I
FPD
"(ll
d
K-
, dl
yet)
-
PLANT
Figure 9. Control system wi th an FPD controller
4.2 Defuzzification methods
Defuzzification interface uses the implied fuzzy sets or the overall impli ed fuzzy set to
obtain a crisp output va lue. There are many defuzziiic3lion methods but the most
common methods are as fo llows:
1) Center of gravity (COG)
2) Bisector of area (BOA)
3) Smallest of maximum (SOM)
4) Mean of maximum (MOM)
5) Largest of maximum (LOM)
For discrete sets COG is called center of gravity for singletons (COGS) where the crisp
control value UCOGS is the abscissa of the center of gravity of the fuzzy set. UCOGS is
calculated as follows:
L f JlcCXaX/
(26)
UcOGS = Ltllc(Xi)
where Xi is a point in the universe of the conclusion (i = 1,2, ... ) and .u,exD is the
membership value of the resulting conclusion set. For continuous sets summations are
replaced by integrals.
The bisector of area (BOA) defuzziflcation method calcul ates the abscissa of the
vertical line that divides the area of the resulting membership functi on into two equal
areas. For discrete sets. U 60A is the abscissa Xj that minimizes
j tmllX
I - I L i <j < imax
(27)
i=l I:j+l
Here i mcu: is the index of the largest abscissa x i " BOA is a computationaJly complex
max
method.
48
Another approach to obtain the crisp value is to choose the point with the highest
membershi p. There may be several points in the overall implied fuzzy set which have
maximum membership value. Therefore it ' s a common practice to calculate the mean
value of these paints. This method is called mean of maximum (MOM) and the crisp
value is calcul ated as follows:
LiEf xi
UMOM = I = (i I = I'm=) (28)
III
Here J is the (crisp) set of indices i where J.Lc(xtl reaches its maximum and PI is
ils cardinali ty (the number of members).
One can also choose the leftmost poim among the points which have maximum
membership to the overall impli ed fuzzy set. This metbod is called smallest of
maximum (SOM) or the leftmosl maximum (lJvl) defuzzification method. Crisp value is
calculated as follows:
USO M == xmln (f)
(29)
Another possibility is to choose the rightmost point among the points which have
maximum membership to the overall implied fuzzy set. This method is called largest of
maximum (LOM) or the rightmost maximum (RM) defuzzification method where crisp
value is calculated as:
U WM = xmax (I)
(30)
4.3 Simuli nk implemen tation
Inputs of FPD are "error" and "change of error" where the output is "control". Input and
output variables of FPO consisr of seven fuzzy sets namely NB (negative big), NM
(negative medium), NS (negative smaJl ), Z (zero), PS (positive small), PM (positive
medium) and PB (positive big) as shown in Figure lO(a) and (b). Table 4 shows fuzzy
rules.
,
,VB .\').1 .VS z PM P8
"
(8) Fuzzy input variables "error"" and "change of error"
n
(b) Fuzzy output variable "output"
Figure 10. Fuzzy input-output variables
49
Table 4. Fuzzy rules
PB
NM I NS "'H Z i PS
, IJ\
NB NB I NB NO I NM NS Z NB
NM NS Z PS NM I NB NB I NB
N$ , Z PS PM NS NB NB NM
Z ! PS PM PB Z NO NM NS
PS , PM NS Z PB PB PS NM
ni l NS PS PM I PB PB PB z
PB ' PB PS PM PB PB PB Z
Figure 11 shows the fuzzy PD control system designed in Simulink.
"'......
I.....
OpIinl2:lti::ln
r I
'"
Kd
l-JoIVa 1116's
r:J
11 VI) I
na; ,.,
l oc moJor
Loo'
""""
Figure 11. Fuzzy PO control system
Different defuzzification methods were used to obtain the control signal. Table 5 shows
the tuned values of the controller parameters for different defuzzification methods.
Table 5. Controller parameters for different defuzzificacion methods
Kp Method KD
2. 2484 0.01 Bisector
4.1236 0.01 SOM
0.1901 MOM 4.5538
4.7623 0 .1649 LOM
Figure 12(a)-(d) shows the system responses and conrml signals for the fuzzy control
systems with different defuzzification methods.
Table 6 shows the values of the perfounance criteria for different defuzzification
methods with the tuned controller parameters.
50