International Journal of Scientific & Engineering Research Volume 3, Issue 3, March -2012 1

ISS N 2229-5518

Neural adaptive control by state space system

UPFC (SSNN) for compensation of active
And reactive power
Bouanane Abdelkrim, Chaker Abdelkader, Addadi Zoubida, Amara Mohamed

Abstract—in our present paper, we present the effectiv eness of the controller's electric al power flow Universal (unif ied power flow
controller UPFC) w ith the choice of a control strategy. To evaluate the performance and robustness of the system, w e proposed a
hybrid control combining the concept of neural networks with conventional regulators vis -à-vis the changes in characteristics of
the transmission line in order to improve the stability of the electrical power network.

Index Terms—UPFC system, adaptive control, neural netw orks, state space (SSNN).

——————————  ——————————

1 Introduction

w ith the rapid development of the modern world, the

demand for electricity is growing and electrical
installations are continually enhanced to meet these
one serial and one paralle l to the transmission line .

requirements. The construction of new plants and new lines

are necessary, but with FACTS devices, we can solve some
problems while using the existing facilities.
Having highlighted the need for rapid control of power
flow in transmission line and the description of the new
concept of "FACTS" that was born to meet the increasing
difficulties in networks including control of the flow on the
axes transport, we are interested in our work to the controller's
electrical power flow universal [1], [2] (Unified Power Flow
Fig. 1 UPFC System Configuration
Controller UPFC).
The UPFC consists of two switching converter, (series
and shunt) (Fig. 1) and even this device is the union of a After all, this composition offers the UPFC's ability to control
the active power and reactive power regardless of where:
parallel compensator and a series compensator. It is capable of
simultaneously and independently control the active power - The series inverter (2) "Inverter2" performs the principal
and reactive power. It can control the three parameters function of the UPFC by injecting an AC voltage in series (AC)
with an amplitude and a phase angle adjustable.
associated with power flow the line voltage, the impedance of
the line and angle of transport. - The parallel UPS (1) "Inverter1 'role is to provide or absorb
the real power demanded by the inverter (2) to the connection
(DC), as it can also produce or absorb reactive power
2 Configuration Variable Universal Charges according to demand, and provide independent shunt
(UPFC): compensation transmission line.
It is assume d that the UPFC [3] shown in Fig. 1 was plugge d The series inverter (2) provides or absorbs the needed reactive
into a simplifie d transmission system, the arrival of the power locally produced active power as a result of injecting a
transmission line (Rece iving End). The two voltage source voltage in series.
inve rters constituting the UPFC are connecte d togethe r through a
common DC circuit. 3 Modelling O f the System UPFC:
Two transformers T1 and T2 are use d to connect the two inve rters,
The simplified circuit of the control system and compensation
of UPFC is shown in (Fig. 2) modeling of this circuit is based
———————————————— on simplifying assumptions in the form of ideal voltage
Boua na ne,, Addadi a nd Amara are teachers in the department of sources then the dynamic equations of the UPFC are divided
electrica l engineering a t the University Dr. MOULAY TAHER,
into three systems of equations: the equations of the series
in SAIDA, Algeria . ([email protected]) branch, the equations of parallel branch and those of the DC
Cha ker is professor in the department of electrical engineering at the
ENSET in ORAN , Algeria. ([email protected]) IJSER © 2012
http://www.ijser.org
 International Journal of Scientific & Engineering Research Volume 3, Issue 3, March -2012 2
ISS N 2229-5518

circuit.
. v pb - v cb - v rb 
di pb rp 1 (4)
By applying Kirchhoff’s laws we will have the following - . i pb 
dt Lp Lp
equations.
di
1 
.  v - v - v 
pc r
UPFC  - p .i 
M odel of the line Is Ir dt Lp pc Lp  pc cc rc 
With a transformation marke r d, q we have the syste m of
r L Vc ~ equations (5):
Lp d di
~ Vs rp ~ Vr pd
 ω . ipq -
rp
.i 
1 
 v - v - v 
dt Lp pd Lp  pd cd rd 
d Vp d
di pq rp 1   (5)
~ dt
  . i pd -
Lp
. i pq  v - v - v 
L p  pq cq rq 
(Sending) ip (Receiving)
d
End End The matrix form is give n as follows:
d i pd   rp l p    i d  1  v sd  v cd  v rd 
i       r l .i   . v  v  v 
p
Fig. 2 Equivalent circuit of UPFC
3.1 The modelling of the UPFC seri es branch: dt  pq   p p  q l  sq cq p rq 

3.3 The modelling of the UPFC continues branch:

The mathematical model is given by the following equations: By passing on the principle of balance of powe r and ne glecting
di
dt L sa L sa ca

sa  - r . i  1 v - v - v
ra
 the losses of the converters. The DC voltage V dc by the following
equation:
disb
dt
r 1
 - . i  . v - v - vrb
L sb L sb cb
  (1)
dv
dt
c 1

C vc
p e  p ep 
(6)

di
dt L sc L sc cc rc

sc  - r . i  1 . v - v - v  He nce pe = vca isa + vcb isb + vccisc

The transformation of Park aims to model this three-phase pep = vpa ipa + vpb i pb + vpc ipc
system (a, b, c) two-phase (d, q) as follows: (2) With: pe: active powe r consumption of the AC syste m

x d   cos (  t ) - sin (  t ) 1 2
T
x a  Pe p: active powe r injecte d by the shunt inve rte r AC syste m
  2  
 x q   3 cos (  t - 120 )  sin (  t - 120 ) 1 2 x b 
   
Applying the Park transformation on equa tion (6) we obtain:
x o   cos (  t  120 )  sin ( t  120 ) 1 2 x c 
Where x; can be a voltage or current. dv
 v i (7)
 v pq i pq - v i - v cq i q 
dc  3
dt 2Cv  pd pd cd d 
dc
In our case, the component x0 is not seen as the power
system is assumed to be symmetric. After the transformation The UPFC se ries and shunt UPFC's are ide ntical in
of Park, equation (1) is expressed in the d q reference by the e very respe ct. The commands use d to se t the inve rte r are the same
for the shunt inve rter.
equations:
4 Configuration Control Circuit:
di
sd   . i - r . i  1
dt sq L sd L
 v
sd
-v
cd
-v
rd

The ore tically the UPFC should be treate d as a
(3) multivariable system because both se ries and shunt converte rs are
. isq  vsq - vcq - vrq 
disq r 1 connecte d from one side to the transmission line and the other
  . isd -
dt L L side in continuous DC circuit and thus each has two inputs and
The matrix form of the dq axis can be re writte n as follows : two outputs. This for to facilitate the synthesis of se ttings,
treatme nt of the two converte rs will be done se parate ly. The
possibility of this se paration is justifie d by two main factors. First,
d i sd   r l    i sd  1  v sd  v cd  v rd 
  .   .  the coupling be tween the two conve rters of the transmission line
dt i sq      r l i sq  l  v sq  v cq  v rq  is quite small. Second, the dynamic variation of the voltage of DC
side of the continuum is dominate d by the paralle l conve rter.
3.2 The modelling of the shunt branch: Control of the converte r in paralle l UPFC is very similar to that of
The mathematical mode l of the UPFC shunt is give n similarly the compe nsator SVC. So to control the flow of active powe r in
the transmission line , the controlle r of the UPFC series must
by the following e quations :
adjust the angle of the phase of the compe nsation voltage V c while
di
1 
.  v - v - v 
pa r to adjust the flow of reactive powe r, the amplitude of the input
 - p .i 
dt Lp pa Lp  pa ca ra  voltage range must be controlle d. To e nsure syste m stability, a
chain of control is imple mente d with PI control.

IJSER © 2012
http://www.ijser.org
 International Journal of Scientific & Engineering Research Volume 3, Issue 3, March -2012 3
ISS N 2229-5518

Control of the se ries branch Zie gle r-Nichols, use d in the pre sent thesis is base d on a trial
Control of paralle l branch and the game conducte d in close d loop with a simple analog proportional
continues. controlle r. The gain Kp of the regulator is gradually increase d
until the stability limit, which is characterize d by a steady
oscillation. Base d on the results obtaine d, the paramete rs of the PI
controlle r give n by the analog transfer function.
4.1 Description of the Control System of UPFC:  1 
K ( s )  Kp 
 1  
 Ti s 


The active and reactive powe rs P and Q are give n by

We can say in the conditions of the converter, the e xcess
2

the equations: P  3 V .i  V .i
sd sd sq sq
3
 
Q  Vsd .isq  Vsq .isd
2
(8)
 curre nts should be minimal. There fore , the introduction of a
simple condition K i  (r L).k p . We obtain the transfer function
Where ird = isd + i pd
F(s)  k p k p  s of the form is first class with a time constant
irq = isq + i pq
T  1 k p .He nce F(s) 
1 (10)
The re fere nce power and reactive P * and Q * of the desire d real
1  s.T
powe rs P and Q are use d as input to the control syste m of UPFC.
From equation (8) the re fe rence currents isd * and isq * can be Thus, de termining the time constant de pe nds on the
calculate d as follows: maximum allowable change of control variable s and V cd V cq for
(9) controlle r se ries and the same for the shunt converte r. So
 * *   * * 
2  P .Vsq  Q .Vsd  i*  2 
P .Vsq  Q .Vsd
 according to the method of Zie gle r-Nichols, the critical gain Kpc
i*
q  3 q
  
 3  

    and the pe riod Tc of the oscillations is measure d by the choice of
the table as follows:
With   V 2  V 2 Kp =0,45 kp c e t Ti =0,83 Tc avec Td=0
sd sq
UPFC (Série)
The re fere nce powe r and reactive P * and Q * of the desire d real
powe rs P and Q are use d as input to the control syste m of UPFC. * - v cd
isd + + 1
From equation (8) the re fe rence currents isd*and isq * can be PI isd
+ - S+r/l
calculate d as follows. The refe re nce curre nts I rdref I rqref and are

+ 
calculate d according to equations (8). These re fe rence values and - +
I rdref I rqref are the n compare d with actual line curre nts from the
re ce iver. + +
 
The outputs of the PI e ditors provide the curre nt value s of control isq
*
+ - 1
voltages V sd and V cq. The goal is to have active and reactive power PI + + S+r/l
at the finish line (Rece iving End) ide ntical to those instructions - v cq isq
(P *, Q *) by forcing the line curre nts (I sd , I sq) to monitor properly
+
the ir refe re nces. The re fere nce curre nts calculate d (8) compare d Fig. 3 Block Diagram of the UPFC Control (series)
-
to the actual line curre nts and afte r a correction to the curre nt one
e nds control voltages (UPFC se ries) and V cd Vcq re prese nting the 4.2 Simulation Results with PI-D UPFC:
re fe re nce voltage control circuit (PWM) of the inve rter in Fig. 3.

4.2 Decoupled PI Controller (a)

According to the system of e quations (3) or (5), one can have
the system contains a coupling be twee n the reactive and active
curre nt Id Iq. The inte raction betwee n curre nt loops cause d by the
coupling term (ω) (Fig. 3). This e xplains the deviation of reactive
powe r with respect to the re fere nce . To re duce the inte raction
be twee n the active and reactive power, a decoupling of the two
curre nt loops is nee de d.The function of decoupling is to remove
the product ωL and I q controlle r along the axis d and adding the
product te rm ωL I d and the controller along the axis q. The design
of the control system must be gin with the se lection of variables to
adjust and the n that of the control variables and the ir association
(b)
with variables se t. There is various adjustment te chniques we ll
suite d to the PI controlle r. The re are two we ll-known e mpirical
approaches propose d by Zie gle r and Tit for de termining the
optimal paramete rs of the PI controller. The method
IJSER © 2012
http://www.ijser.org
 International Journal of Scientific & Engineering Research Volume 3, Issue 3, March -2012 4
ISS N 2229-5518

Fig. 5 Answers Powers to Change the Reactance

of 30%
Fig. 4 (a), (b). Answers of Pow ers with PI-D.

Figures (4-a, b) illustrates the behaviour of active and reactive

powe r, whe re we see that the control system has a fast dynamic
re sponse to the forces reach the ir steady states after a change in
the re fe rence values.
We also note the presence of the interaction be twee n the
two compone nts (d and q). These influe nces are cause d by the
PWM inve rter is unable to produce continuous signals nee de d by
the decoupling, thus increasing the e rror in the PI -D controlle rs.
To test robustness, we teste d for a variation of the reactance
XL to 30% and I had variations on the output powe r following
Following the se changes, the active and reactive powe r
(Fig. 5) unde rgo large deviations more or less with an ove rflow at
times of great change instructions (Pre f, Qref), which means the
pe rformance de gradation of the PI controlle r, interpre te d by the
loss of system stability
We simulate d this time by introducing perturbation (Fig. 6)
duration of 25 ms and amplitude 1.5 to test again its robustness
and stability of the system.
The response of active and reactive powe r at 30% of XI could
be de tecte d; the e rror message give n by "MATLAB command"
indicate d the saturation of the orde r at infinity in the spike s of the Fig.6. UPFC System Perturbed to Test Stability
re fe re nce signals. 5 Adaptive Neural Controls:
We can check the response of the se rvo not only continue d but
also regulation by adding a disturbance The inte rest in adaptive control [4] appears mainly at the
le ve l of parame tric pe rturbations, ie act on the characte ristics of
the process to be controlle d, disturbance acting on the variable s to
control or order. In this pape r we present the metho d of
adjustme nt propose d for the UPFC, e mphasizing the classical
approach base d on ne ural ne tworks.

5.1Choi x a Neural Network:

In this paper the Elman ne twork [5] said hidde n layer
ne twork is a recurre nt ne twork, and there fore be tter suite d for
mode lling dynamic systems. His choice in the ne ural control by
state space is justifie d by the fact that, in particular, the ne twork
can be inte rpre te d as a state space mode l nonlinear. Learning by
back propagation algorithm standard is the law use d for the
ide ntification of the UPFC.

5.2State Space Adaptive Neural Control« SSNN»:

The inte gration of these two approaches (ne ural adaptive
control) [6] in a single hybrid structure, that each be nefits from
the othe r, but to change the dynamic be haviour of the UPFC
system was adde d against a reaction calculate d from the state
vector (state space ) (Fig. 7).

E (t) + U (t) Process’s Y (t)

IJSER © 2012
http://www.ijser.org
UPFC
 International Journal of Scientific & Engineering Research Volume 3, Issue 3, March -2012 5
ISS N 2229-5518

hidde n laye r and output laye r. The laye rs of input and output
inte rfere with the e xternal e nvironme nt, which is not the case for
the interme diate laye r calle d hidde n laye r ie the network input is
K the command U (t) and its output is Y (t).
x (t) The state vector X (t) from the hidde n laye r is inje cte d into the
Fig. 7 Block Diagram of Feedback Control State of the UPFC. input laye r.
The state fee dback control is to conside r the process mode l in We de duce the following e quations:
the form of an e quation of state : X (t) = W r X (t-1) + W h U (t-1) ( 21)
X‫( ٭‬t) = A x (t) + B u (t) (11) Y(t) = W o X (t) (22)
And observation e quation: Where , Wh ; Wr e t Wo are the we ight matrices. Equations are
Y (t) = C x (t) +D u (t) (12) standard de scriptions of the state space of dynamical systems.
Where u(t) is the control vector, x (t) the state vector, and y (t) the The orde r of the syste m de pe nds on the number of states e quals
output vector of dime nsion for a discre te syste m to the sampling the numbe r of hidde n laye rs. Whe n an input-output data is
process parame ters Te at times of Te sample k are formalize d as
prese nte d to the network at ite ration k square d error at the output
follows :
of the ne twork is de fine d as
x(t+1) = Ad x(t) + B ud(t) (14)
E t  y d ( t )  y( t ) 
1 2 (23)
Y(t) = C d x(t) + Dd u(t) (15)
2
The transfer function G (s) = Y (s) / U (s) of our process UPFC can For all data u (t), y d (t) de t = 1,2,….. N, the sum of square d e rrors
is: N
(24)
be writte n as: E   Et
t 1

1 (16) The we ights are modifie d at each ite ration, for W 0 we have :
G s   E t y( t )
 y d ( t )  y( t ) 
sr L (25)
W0 W0
We de duce the e quations of state re presentation of the UPFC:  y d (t )  y(t ).x T (t )
 * r
 x   L .x  u (17)
   For Wh e t Wr, we re : E t E t y( t ) x ( t )
 . .
 Wh y( t ) x ( t ) Wh
y  x

  y d (t )  y(t ).W0T .u(t ) (26)
With: U (t)=e (t) –k x (t)
Le t: X (t +1) = [Ad – K B d] x (t) + B d e (t) (18) E t E t y( t ) x i ( t )
 . .
Wri y( t ) x i ( t ) Wri
Y (t) = C d x (t) (19)
x i ( t )
 y d ( t )  y( t ) .W0i
(27)
The dynamics of the process correcte d by state space is prese nte d Wri
base d on the characte ristic e quation of the matrix [Ad - B d K], x i x ( t  1)
whe re K is the matrix state space controlle d process.  X T ( t  1)  Wri .
The latter we obtain: W r
i
Wri (28)
Our syste m is describe d in matrix form in the state space :
The variation of the we ight matrix base d on the learning gain is
x *  A.x  B.u (20)

 writte n as: W  . E t (29)
 y  C.x  D.u
 W
Where : Elman network
 r  1  W 0 , W h et W r
 L  L 0 1 0 0 0 
A   B  C   
 D   

   
   r  0  1 0 1 0 0
 L   L 

 v cd  i sd 
x  i sd i sq 
    e (t)= i *sd
u y 
T V(t)= id
 + G(s)
 v cq  i sq 
    u(t)=Vc UP FC in system the state
d space

5.3 Identification Based Network Elman:

K
Process ide ntifie d [7] [8] will be characte rize d by the Estimated state vector
X
mode l structure (Fig. 8), of his orde r and paramete r values. It is
Fig. 8 Netw ork ELMAN and State Space.
there fore , a corollary of the simulation process for which using a
Note that the performance of the ide ntification is bette r whe n the
mode l and a set of coe fficie nts to pre dict the response of the
input signal is sufficie ntly high in freque ncy to e xcite the differe nt
system. The Elman network consists of three laye rs: an input layer,
IJSER © 2012
http://www.ijser.org
 International Journal of Scientific & Engineering Research Volume 3, Issue 3, March -2012 6
ISS N 2229-5518

modes of process. The three we ights Wo, Wr and Wh which are

re spective ly the matrices of the equation of state of the process
system (UPFC) [CA and B] became stable afte r a rough time
t = 0.3s and se veral ite rations (fig. 9). Fig. 11 Neural Adaptive Control (SSNN) Pow ers P, Q
Note: For the Elman network ne uron type is assume d through the And at (± 25% XL)
vector is zero (D = 0).
In online learning Elman ne twork, the task of ide ntifying 6 Conclusion
and correcting same synthesis are one afte r the other. Or
The ide ntification process will be characterize d by the
correction of the numerical values of the parame ters is done
re peate dly so the estimation error (Fig. 10) take s about almost a mode l structure , its orde r and paramete r values. It is the re fore a
corollary of the simulation process which uses a mode l and a set
second (t = 1s) to conve rge to zero ie re gulation in pursuit.
of coe fficie nts to pre dict the response of the syste m. The use of
more advance d low-ide ntification ne ural ne twork can be
de pe ndable in the calculation algorithm of the orde r.

In this pape r we use d for the ide ntification of system

paramete rs a ne uron ne twork said Elman ne twork with three
laye rs. As we have already see n. Note that the performance of the
Fig. 9 Changing Weight ide ntification is be tte r whe n the input signal is sufficie ntly high in
fre que ncy to e xcite the diffe re nt modes of process.

Ne ural adaptive control by state fee dbac k (SSNN: state

space ne ural ne twork)) is a hybrid control base d on the
re prese ntation of system status UPFC was teste d. The
pe rformance of the latter are slightly de grade d, this is may be due
to the de lay cause d by the algorithm, it can't be re duce d at wil l, or
it may due to the choice of the gain K of the close d loop.
Fig. 10 Estimation Error
5.4 Robustness Test:
Finally, the ide ntification process base d on learning of
To check the robustness to the controller, two tests were
Elman ne ural ne twork, provide s the dynamic be havior of the
pe rforme d. For each test we varie d the parame te rs of the
process and to estimate the syste m output as its state vector base d
transmission line but the controller remains unchange d. It can be
on the information that are control signal and the measure d
see n that the variation of the reactance (± 25%) has almost no
output.
influe nce on the output characte ristics of the UPFC syste m (Fig.
11).To compare the responses of active and reactive powe r of
UPFC syste m, we gave the three cases (Fig. 11) Where (a) and (c)
are the responses to changes (± 25%) and (b) is the response of the
7 References
system without any change in reactance .

(a) [1] I. Papic, P. Zunko, D. Povh, M. Weinhold, ‚Basic control of unified

power flow controller‛. IEEE Trans. Power Syst., vol. 12, no. 4, pp.
1734–1739, 1997.
[2] Y. H. song and A .T. Johns,’ flexible AC transmission systems
(FACTS).’ IEE power and Energy series 30.1999.
[3] Gyugyi L.: unified power flow control concept for flexible AC
transmission systems. IEE Proc. 139 (1992) 323-331
[4] O. pages ‘étude de comparaison de différentes struc tures de
commande multi- contrôleurs application à un axe robotise.’ thèse
2001. (LAMII/CESALP).université de S avoie.
(b) [5] Xiang Li, Guanrong C., Zengqian C., and Z. Yuan ‘Chaotifying
Linear Elman Networks’ IEEE transaction neural networks. vo1, 3. 5
September 2002.
[6] M. A. Denaï, T. Allaoui, ‚Adaptive fuzzy Decoupling of UPFC-
Power Flow Compensation‛, 37th UPEC2002, 9-11 September 2002,
S traffordshires University UK.
[7] S. zebirate, A.chaker,O.traiaia , ‘commande addaptative
(c) decouple neuronale d’un compensateur de flux de puissance
UPFC’reference480.laboratoire LAAS.CCECE 03-CCGEI
2003.Montreal.IEEE May 2003.
IJSER © 2012
http://www.ijser.org
 International Journal of Scientific & Engineering Research Volume 3, Issue 3, March -2012 7
ISS N 2229-5518

[8] S. zebirate, A.c haker, ‘commande addaptative decouple neuronale

d’un UPFC (unified power flow control)’, JCGE Saint Nazaire ,Juin
2003.

IJSER © 2012
http://www.ijser.org