Chapter 8

Synchronization and Stabilization

of Distributed Power Generation Units

8.1 Outline

The chapter analyzes synchronization and stabilization for distributed power

generation units. First, the problem of stabilization and synchronization is solved
for a model of distributed interconnected synchronous power generators with the use
of flatness-based control. Next, the problem of stabilization and synchronization of
the distributed synchronous generators is solved with the use of flatness-based adap-
tive control. Moreover, the problem of decentralized control and synchronization is
solved for a model of distributed interconnected DC power generation units, such as
photovoltaics.
First, the chapter develops a control method for distributed interconnected power
generation units. The power system comprises Permanent Magnet Synchronous Gen-
erators (PMSGs) which are connected to each other through transformers and tie-
lines. A derivative-free nonlinear Kalman Filtering approach is introduced aiming at
implementing sensorless control of the distributed power generators. In the proposed
derivative-free Kalman Filtering method the generator’s model is first subjected to
a linearization transformation that is based on differential flatness theory and next
state estimation is performed by applying the standard Kalman Filter recursion to the
linearized model. Unlike Lie algebra-based estimator design methods, the proposed
approach provides estimates of the state vector of the permanent magnet synchronous
generator without the need for derivatives and Jacobians calculation. Moreover, by
redesigning the proposed derivative-free nonlinear Kalman Filter as a disturbance
observer it is possible to estimate at the same time the non-measurable elements
of each generator’s state vector, the unknown input power (torque) and the distur-
bance terms induced by interarea oscillations. The efficient real-time estimation of
the aggregate disturbance that affects each local generator makes possible to intro-
duce a counter-disturbance control term thus maintaining the power system on its
nominal operating conditions.
Moreover, the chapter proposes a new nonlinear H-infinity control method for sta-
bilization and synchronization of distributed interconnected synchronous generators.

© Springer International Publishing Switzerland 2016 339

G. Rigatos, Intelligent Renewable Energy Systems,
Green Energy and Technology, DOI 10.1007/978-3-319-39156-4_8
340 8 Synchronization and Stabilization of Distributed Power Generation Units

At first stage local linearization of the distributed generators’ model is performed

round its present operating point. The approximation error that is introduced to the
linearized model is due to truncation of higher-order terms in the performed Taylor
series expansion and is represented as a disturbance. The control problem is now
formulated as a mini-max differential game in which the control input tries to min-
imize the state vector’s tracking error while the disturbances affecting the model
try to maximize it. Using the linearized description of the distributed generators’
dynamics an H-infinity feedback controller is designed through the solution of a
Riccati equation at each step of the control algorithm. The inherent robustness prop-
erties of H-infinity control assure that the disturbance effects will be eliminated and
the state variables of the individual power generators will converge to the desirable
setpoints. The proposed method, stands for a reliable solution to the problem of
nonlinear control and stabilization for interconnected synchronous generators. It is
also a novel approach, comparing to control of synchronous generators based on
global linearization methods. Its efficiency is further confirmed through simulation
experiments.
Additionally, the chapter presents an adaptive fuzzy approach to the problem
of synchronization and control of distributed power generators, which is based on
differential flatness theory and which uses exclusively output feedback. It is shown
that the MIMO model of the distributed power generators is a differentially flat one
and this permits to transform it to the so-called linear canonical form. For the new
description of the system’s dynamics the transformed control inputs contain unknown
terms which depend on the system’s parameters. To identify these terms adaptive
fuzzy approximators are used in the control loop. Thus an adaptive fuzzy control
scheme is implemented in which the unknown or unmodeled system dynamics is
approximated by neurofuzzy networks and next this information is used by a feedback
controller that makes the state vector of the distributed power generators converge
to the desirable setpoints. This adaptive control scheme is exclusively implemented
with the use of output feedback, while the state vector elements which are not directly
measured are estimated with the use of a state observer that operates in the control
loop.
Finally, the chapter develops decentralized control for parallel inverters connected
to the power grid, using differential flatness theory and the Derivative-free nonlinear
Kalman Filter. The problem is of elevated difficulty comparing to the control of stand-
alone inverters because in this case in the dynamics of each inverter one has also to
compensate for interaction terms which are due to the coupling with other inverters.
It is proven that the model of the inverters, is a differentially flat one. By exploiting
differential flatness properties it is shown that the multiple inverters model can be
transformed into a set of local inverter models which are decoupled and linearized.
For each local inverter the design of a state feedback controller becomes possible,
e.g. using pole placement methods. Such a controller processes measurements not
only coming from the individual inverter but also coming from other inverters which
are connected to the grid. Moreover, to estimate the non-measurable state variables of
each local inverter, the Derivative-free nonlinear Kalman Filter is used. As explained
before, this consists of the Kalman Filter recursion applied to the local linearized
8.1 Outline 341

model of the inverter and of an inverse transformation that is based on differential

flatness theory, which enables to compute estimates of the state variables of the initial
nonlinear model of the inverter. Furthermore, by redesigning the aforementioned filter
as a disturbance observer it becomes also possible to estimate and compensate for
disturbance terms that affect each local inverter.

8.2 State Estimation-Based Control of Distributed PMSGs

8.2.1 Outline

In this section, a solution is presented about the problem of distributed control of

interconnected generators. Actually, research results are presented on sensorless con-
trol of interconnected PMSGs (permanent magnet synchronous generators) with the
use of the Derivative-free nonlinear Kalman Filter, in the framework of a multi-area
multi-machine model of distributed power generation units. A large-scale power sys-
tem consists of multiple control areas where local power generation units operate,
while these areas are connected through tie lines and power transformers. Due to
change in active power loads, the frequencies of the areas variate from their nomi-
nal values (synchronous speed) and power exchange through the tie lines will also
deviate from the associated setpoints [107, 108, 116, 165, 167, 233, 302, 359, 465,
477, 479, 495]. Therefore, it is necessary to supply power generator units with local
controllers capable of maintaining the generators’ turn speed at its nominal value. To
stabilize the functioning of the distributed power generation system and eliminate
frequency deviations several control approaches have been implemented, such as
standard PID control, lead-lag compensators where the selection of the controller’s
parameters is performed with the use of Kharitonov’s theorem, H∞ control or adap-
tive control [88, 102, 116, 198, 207, 213, 237, 238, 349]. To control efficiently power
generators in case that the complete state vector of the electric machine cannot be
measured directly, observer-based control schemes have been proposed [121, 267,
290, 291]. Moreover, to control distributed power generators, disturbance rejection
control schemes have been developed [60, 262, 265, 315].
A sensorless control scheme is developed for the local power generators. The
control approach is based on a diffeomorphism (change of coordinates) that fol-
lows differential flatness theory and which enables to transform the initial nonlin-
ear model of the generator into a linear one. The latter model makes possible to
design a state feedback controller for the PMSG. Moreover, to estimate the non-
measurable elements of the PMSG state vector and to use them in the control loop,
the Derivative-free nonlinear Kalman Filter is considered. This filtering approach
exploits the previously mentioned transformation of the generator’s model in the
linear canonical (Brunovsky) form through the application of differential flatness
theory, and performs filtering on the linearized equivalent of the generator using
the standard Kalman Filter recursion. It will be shown that the sensorless control
342 8 Synchronization and Stabilization of Distributed Power Generation Units

scheme for the local power generators can be modified so as to provide the desirable
robustness to the distributed power generation scheme. By redesigning the Kalman
Filter, initially used for estimating the non-measurable elements of the generator’s
state vector, in the form of a disturbances estimator it becomes possible to estimate
in real-time the aggregate disturbance that affects each local generator [98, 99, 169,
292]. Knowing the disturbances’ value it is possible to generate a counter-disturbance
term that serves as auxiliary control input and which finally compensates for the dis-
turbances’ effects. With the previously described control approach each generator
keeps on running at its nominal operating conditions and the distributed generation
units become robust to phenomena such as mechanical input torque variations and
interarea oscillations [15, 439, 474, 483, 504].
The proposed control method is suitable for compensating for both endogenous
and exogenous disturbances that affect the generators’ model. Endogenous distur-
bances are associated with parametric uncertainty, e.g. there can be imprecision in
the values of reactances, stator and rotor circuit coefficients and the rotor’s moment
of inertia. Exogenous disturbances are associated with variations in the mechanical
input torque, as well as with perturbations in the excitation system of each gener-
ator which are due to lack of synchronization with other generators connected to
the grid [50]. The efficient compensation of these disturbances enables to maintain
synchronism between the individual power generators, thus assuring: (i) robustness
improvement of the power grid and maintenance of the synchronization between the
distributed electric power generation units even under adverse conditions in the power
grid, (ii) improvement of the provided power quality (distortions in power quality due
to deviations of the power generation units from their nominal operating conditions
will be avoided), (iii) uninterrupted power flow in the grid, ability to incorporate more
power generation units without affecting stability, and consequently better response
to the increased demand for power supply.
The proposed distributed filtering and control method for the system of the inter-
connected power generators is a highly data-driven one. Actually, to succeed state
estimation and disturbances compensation for a set of n distributed PMSGs it is
necessary that at each sampling period each power generator controller receives n
variables associated with the turn angles of all generators in the power grid. Thus,
the total amount of data that has to be processed consists of n2 variables. This is
a quite intensive computation load which can be managed efficiently thanks to the
distributed control structure that has been implemented (each generator computes
locally its own control signal). Moreover, taking into account that the exchange of
data has to take place in real time one can conclude that communication links of
efficient bit rate have to be established between the generators.
8.2 State Estimation-Based Control of Distributed PMSGs 343

8.2.2 Dynamic Model of the Distributed Power Generation

Units

8.2.2.1 Dynamics of the Local Power Generator Units

It is considered that the third order model of the PMSG is connected to the power
grid thus forming the model of a Single Machine Infinite Bus (SMIB) system. As
explained in Chap. 3 PMSG mechanical dynamics can be represented as follows

δ̇ = ω
ω0 (8.1)
ω̇ = − 2J
D
(ω − ω0 ) + 2J
(Pm − Pe )

where δ is the turn angle of the generator’s rotor, ω is the rotation speed of the rotor
with respect to synchronous reference, ω0 is the synchronous speed of the generator,
J is the moment of inertia of the rotor, Pe is the active power of the generator, Pm is
the mechanical input torque to the generator which is associated with the mechanical
input power, D is the damping constant of the generator and Te is the electrical torque
which is associated to the generated active power. Moreover, the following variables
are defined: Δδ = δ − δ0 and Δω = ω − ω0 with ω0 denoting the synchronous
speed. The generator’s electrical dynamics has already been described in Chap. 1
and is summarized as follows [215, 249, 266]:

Ėq = 1
Td o
(Ef − Eq ) (8.2)

where Eq is the quadrature-axis transient voltage of the generator, Eq is the quadrature
axis voltage of the generator, Tdo is the direct axis open-circuit transient time constant
of the generator and Ef is the equivalent voltage in the excitation coil. The algebraic
equations of the synchronous generator are given by
xd
Eq = xd
Eq − (xd − xd ) xV s cos(Δδ)
 d

Iq = Vs
xd
sin(Δδ)

Eq
Id = xd
− Vs
xd
cos(Δδ)
 
Vs Eq (8.3)
Pe = xd
sin(Δδ)

Vs Eq V2
Qe = cos(Δδ) − xds

 xd
 

Vt = (Eq − Xd Id )2 + (Xd Iq )2

where xd = xd + xT + xL , xd  = xd + xT + xL , xd is the direct-axis synchronous reac-

tance, xT is the reactance of the transformer, xd is the direct axis transient reactance,
xL is the reactance of the transmission line, Id and Iq are direct and quadrature axis
currents of the generator, Vs is the infinite bus voltage, Qe is the generator reactive
power delivered to the infinite bus, and Vt is the terminal voltage of the generator.
344 8 Synchronization and Stabilization of Distributed Power Generation Units

Substituting the electrical equations of the PMSG given in Eq. (8.3) into the equa-
tions of the electrical and mechanical dynamics of the rotor given in Eqs. (8.1) and
(8.2) respectively, the complete model of the Single Machine Infinite Bus model is
obtained
δ̇ = ω − ω0
Vs E 
ω̇ = − 2J
D
(ω − ω0 ) + ω0 P2Jm − ω0 2J1 x q sin(Δδ) (8.4)
d

1 xd −xd
Ėq = − T1 Eq + Tdo xd
Vs cos(Δδ) + 1
Td o
Ef
d 

x
where Td = xdd Tdo is the time constant of the field winding.

The previously analyzed single-machine infinite-bus model of the PMSG is
described by a nonlinear state-space model of the form

ẋ = f (x) + g(x)u (8.5)

 T
where the state vector x is defined as x = Δδ Δω Eq . The vector fields f (x) and
g(x) are defined as
⎛ ⎞
ω − ω0
⎜− D (ω − ω ) + ω Pm − ω 1 Vs Eq
f (x) = ⎝ 2J 0 0 2J 0 2J 
xd
sin(Δδ)⎟
⎠ (8.6)

1 xd −xd
− T1 Eq + 
Tdo xd
Vs cos(Δδ)
d

 T
g(x) = 0 0 1
Tdo
(8.7)

with control input u = Ef being the field voltage (equivalent voltage in the excitation
coil) and measurable output the turn angle of the rotor

y = h(x) = δ − δ0 (8.8)

8.2.2.2 Dynamics of the Interconnected Power Generators

The interconnection between several local power generation units which are described
by the previously analyzed SMIB model results into a multi-area multi-machine dis-
tributed power generation system.
A multi-machine power system (MMPS) with n machines, in which the first
machine is chosen as the reference machine can be described by the following non-
linear dynamic model
8.2 State Estimation-Based Control of Distributed PMSGs 345

δ̇i = ωi − ω0
P
ω̇i = − 2JDi
(ωi − ω0 ) + ω0 2Jmii −
i
− ω0 2J1 i [G ii Eqi 2 + Eqi nj=1,j=i Eqj G ij sin(δi − δj − αij )] (8.9)

1 xdi −xdi
Ėq i = − T1 Eq i + Tdo i xd
Vsi cos(Δδi ) + 1
Td o i
Ef i
di i

where the electric torque Pei which is associated with the active power at the i-th
generator is now given by
n
Pei = G ii Eqi 2 + Eqi 
j=1,j=i Eqj G ij sin(δi − δj − αij ) (8.10)

for i = 1, 2 . . . , n. For a power grid that consists of n generators the aggregate state
vector comprises the state vectors of the local machines, i.e. x = [x 1 , x 2 , . . . , x n ]T ,
where x i = [x1i , x2i , x3i ]T , with x1i = Δδi , x2i = Δωi and x3i = Eqi are the state
variables for the i-th machine and i = 1, 2, . . . , n.

8.2.3 Linearization of the Distributed Power Generation

System Using Lie Algebra

As previously analyzed, the dynamic model of the distributed power generation

system is defined by Eq. (8.9). The state vector of the i-th local power generator is
defined as
xi = [x1i , x2i , x3i ]T = [Δδi , Δωi , Eqi ]T (8.11)

The local i-th power generation system is described by

ẋ i = f i (x) + g i (x)ui (8.12)

where ⎛ i⎞ ⎛ ⎞
f1 0
⎝ i⎠ i
f (x) = f2 g (x) =
i ⎝ 0 ⎠ (8.13)
i 1
f3 Td o i

with
f1i (x) = x2i
ω0 2 2
f2i (x) = − 2JDi x2i + 2J Pmi − ω0 2J1 i {G ii x3i + x3i [G ii x3 i +
j
i
j (8.14)
+ x3i nj=1,j=i x3 G ij sin(x1i − x1 − aij )]}
xdi −xd
f3i (x) = − T1 x3i + 1
Td o i xd
i
Vsi cos(x1 i ).
di i
346 8 Synchronization and Stabilization of Distributed Power Generation Units

Fig. 8.1 A multi-machine (3-area) distributed power generation model

Generalizing, the distributed power system is described by

ẋ = f (x) + g(x)u (8.15)

 T
where x = x 1 x 2 . . . x n . Without loss of generality a three-machine distributed
power generation model will be considered next (n = 3), as shown in Fig. 8.1. The
state-space equation of the distributed power generation model is
⎛ 1⎞ ⎛ 1 ⎞ ⎛ ⎞
ẋ1 f1 (x) 0 0 0
⎜ẋ 1 ⎟ ⎜f 1 (x)⎟ ⎜ 0 0 0⎟
⎜ 21 ⎟ ⎜ 21 ⎟ ⎜ ⎟
⎜ẋ ⎟ ⎜f (x)⎟ ⎜g1 0 0⎟
⎜ 32 ⎟ ⎜ 32 ⎟ ⎜ ⎟⎛ ⎞
⎜ẋ ⎟ ⎜f (x)⎟ ⎜ 0 0 0⎟
⎜ 12 ⎟ ⎜ 12 ⎟ ⎜ ⎟ u1
⎜ẋ ⎟ = ⎜f (x)⎟ + ⎜ 0 0 0⎟ ⎝ ⎠
⎜ 22 ⎟ ⎜ 22 ⎟ ⎜ ⎟ u2 (8.16)
⎜ẋ ⎟ ⎜f (x)⎟ ⎜ 0 g2 ⎟
0 ⎟ u3
⎜ 33 ⎟ ⎜ 33 ⎟ ⎜
⎜ẋ ⎟ ⎜f (x)⎟ ⎜ 0 0 0⎟
⎜ 13 ⎟ ⎜ 13 ⎟ ⎜ ⎟
⎝ẋ2 ⎠ ⎝f2 (x)⎠ ⎝ 0 0 0⎠
ẋ33 f33 (x) 0 0 g3

The procedure for linearization with the use of Lie algebra is as follows: The
following state vector is defined x = [x11 , x21 , x31 , x12 , x22 , x32 , x13 , x23 , x33 ]T while the
8.2 State Estimation-Based Control of Distributed PMSGs 347

state vector of the linearized equivalent of the distributed PMSG model is defined as
z∈R9×1 . It holds that z11 = h1 (x) = Δδ 1 , z21 = Lf h1 (x) = x21 and z31 = Lf2 h1 (x) = f21 .
In a similar manner one obtains z12 = h2 (x) = Δδ 2 , z22 = x22 and z32 = f22 . More-
over, in a similar manner one obtains z13 = h3 (x) = Δδ 3 , z22 = Lf h3 (x) = x22 and
z33 = Lf2 h2 (x) = f23 .

Furthermore, for the vectors defined as z3 = [z31 , z32 , z33 ], f˜ = [f21 , f22 , f23 ] and
h̃ = [h1 , h2 , h3 ] it holds that

ż = Lf3 h(x) + Lg Lf2 h(x)u (8.17)

or, equivalently
⎛ 1 ⎞ ⎛ ⎞
⎛ 1⎞ ∂f2 ∂f21 ∂f21 ∂f21 ∂f21 ∂f21 ⎛ ⎞
ż3 ∂x1 1
f + ∂x2 2
f + ··· + f
∂xn n ⎟ ∂x1 1
g ∂x2 2
g ··· g
∂xn 3 ⎟ u1
⎜ ⎜ ∂f
⎝ż3 ⎠ = ⎜ 2 f1 +
∂f ∂f22 ∂f22 ⎟ ⎜ 2 g1 ∂f22 ∂f2
g ⎟ ⎝u2 ⎠
2 2 2
2
⎝ ∂x13 ∂x2 2
f + ··· + f
∂xn n ⎠
+ ⎝ ∂x13 ∂x2 2
g ··· ∂xn 3 ⎠
ż33 ∂f2
f +
∂f23
f + ··· +
∂f23
f
∂f2
g
∂f23
g ···
∂f23
g u3
∂x1 1 ∂x2 2 ∂xn n ∂x1 1 ∂x2 2 ∂xn 3
(8.18)
Thus, it holds
ż3 = fa + Mu = v (8.19)

Therefore, setting ż3 = v, where v = fa + Mu the dynamics of the distributed

power generation system for the i-th power generator i = 1, 2, 3 becomes:
⎛ i⎞⎛ ⎞⎛ i⎞
ż1 010 z1 ⎛ ⎞
⎜ i⎟⎜ ⎟⎜ i⎟ 0
⎜ż ⎟ ⎜0 0 1⎟ ⎜z ⎟ + ⎝0⎠ vi (8.20)
⎝ ⎠⎝
2 ⎠⎝ ⎠2
1
ż3i 000 z3i

In this manner the initial nonlinear power system is transformed into two decou-
pled linear subsystems which are in the canonical Brunovksy form. For each one of
these subsystems i = 1, 2, 3 the appropriate control law is
i
vi = zd(3) − k3 (z̈i − z̈di ) − k2 (żi − żdi ) − k1 (zi − zdi ) − d̃ i (8.21)

where the feedback control gains k1 , k2 , k3 are chosen such that the associated char-
acteristic polynomial of the linearized system is a Hurwitz one.
348 8 Synchronization and Stabilization of Distributed Power Generation Units

8.2.4 Differential Flatness of the Distributed PMSG Model

8.2.4.1 Properties of Differentially Flat Systems

The main principles of the differential flatness theory have been given in the previous
chapters as follows [137, 229, 241, 281, 352, 411, 456]: A finite dimensional system
is considered. This can be written in the general form of an ordinary differential
equation (ODE), i.e. Si (w, ẇ, ẅ, . . . , w(i) ), i = 1, 2, . . . , q. The term w denotes
the system variables (these variables are for instance the elements of the system’s
state vector and the control input) while w(i) , i = 1, 2, . . . , q are the associated
derivatives. Such a system is said to be differentially flat if there is a collection of m
functions y = (y1 , . . . , ym ) of the system variables and of their time-derivatives, i.e.
yi = φ(w, ẇ, ẅ, . . . , w(αi ) ), i = 1, . . . , m satisfying the following two conditions
[137, 229, 241, 281, 352, 411, 456]: (1) There does not exist any differential relation
of the form R(y, ẏ, . . . , y(β) ) = 0 which implies that the derivatives of the flat output
are not coupled in the sense of an ODE, or equivalently it can be said that the flat
output is differentially independent, (2) All system variables (i.e. the elements of the
system’s state vector w and the control input) can be expressed using only the flat
output y and its time derivatives wi = ψi (y, ẏ, . . . , y(γi ) ), i = 1, . . . , s.

8.2.4.2 Differential Flatness of the Distributed Power Generation

Model

Differential flatness for the stand-alone synchronous generator was shown in Chap. 3.
It will be proven that the multi-machine power generation system is also a dif-
ferentially flat one. As flat output of the distributed power generation system,
consisting of n PMSGs, the following vector is defined y = [y11 , y12 , . . . , y1n ] or
y = Δδ 1 , Δδ 2 , . . . , Δδ n . For the n-machines power generation system it holds
x11 = y1 , x12 = y2 , x13 = y3 , . . ., x1n = yn and x21 = Δω1 = ẏ1 , x22 = Δω2 = ẏ2 ,
x23 = Δω3 = ẏ3 , . . ., x2n = Δωn = ẏn . Moreover, it holds

ω0
ẋ2i = − 2J x + 2J
Di i
i 2
Pmi −
ω0 2 n j
i
j (8.22)
− 2Ji
[G ii x3i + x3 j=1,j=i [x3 G ij sin(x1i − x1 − αij )]
i

or, using the flat output variables

ω0
ÿi = − 2JDi i
ẏ + 2J Pmi −
ω0 2 n
i
j
i
(8.23)
− 2Ji
[G ii x3i + x3 j=1,j=i [x3 G ij sin(yi − yj − αij )]
i

The external mechanical torque Pmi is considered to be a piecewise constant variable.

For i = 1, 2, . . . , n one obtains n equations of the form of Eq. (8.23), with unknowns
the state variables x3i , i = 1, 2, . . . , n. By solving this system of equations with respect
to x3i , i = 1, 2, . . . , n one arrives at defining the state variables x3i as functions of the
8.2 State Estimation-Based Control of Distributed PMSGs 349

elements of the flat outputs vector yi , i = 1, 2, . . . , n and of their derivatives. Thus

one has x3i = fx3 (y1 , y2 , . . . , yn ). Additionally, from the relation Ėqi = − T1d Eq i +
i

1 xdi −xdi
Tdo i xd
Vsi cos(Δδi ) + 1
Td o i
Efi and knowing that the state variables x1i , x2i , x3i , i =
i
1, 2, . . . , n can be written as functions of the flat output and its derivatives, one can
solve with respect to the control input ui thus showing that all control inputs ui ,
i = 1, 2, . . . , n can be written as function of the flat output and its derivatives.
By expressing all elements of the multi-machine power generation system and
the associated control inputs as functions of the flat output and its derivatives it is
proven that the distributed power generation system stands for a differentially flat
model.

8.2.4.3 The Case of Unknown Time-Varying Mechanical Input Torque

Functioning of the power generation system under unknown time-varying external

mechanical torque Pmi is the most generic condition, which also comprises the case
of piecewise constant external input torque. It will be shown how with the use of
the differential flatness theory a controller can be designed for the distributed power
system in case of unknown time-varying external mechanical torque. By deriving
the expression about ÿi once more with respect to time one obtains
i
y(3) = ai (x) + b1i (x)g1 u1 + b2i (x)g2 u2 + b3i (x)g3 u3 (8.24)

which means that one arrives at a function of the form ż3i = ai (x) + b1i (x)g1 u1 +
b2i (x)g2 u2 + b3i (x)g3 u3 + d̃ i , where in the case of the distributed power generation
that consists of n = 3 machines, and considering for instance i = 1, j = 2, 3 one has

) x2 + Di ω02 [G ii x3i + x3i nj=1,j=i x3 G ij sin(x1i − x1 − αij )]−
Di 2 i 2 j j
ai = ( 2J i (2Ji )
i + n
ω0 x −x 
− 2J [G x x
j
G sin(x i − x j − 4α )(− 1 x i + ( 1 di di V cos(x i ))]−
i
ii 3 j=1,j =i 3 ij 1 1 ij T  3 oi T i
d xd  si 1
di
ω0 i n j xd −xd
− 2J x j=1,j=i G ij sin(x1i − x1 − αij )(− T1 x3i + ( Td1 xid  i Vsi cos(x1i ))−
i 3 oi i
ω0 i n i − x j − α )x i ω0 x i n
di
j j j j
− 2J x
i 3
x G
j=1,j=i 3 ij cos(x 1 1 ij 2 2Ji 3 j=1,j=i 3 ij cos(x1 − x1 − αij )x2
x G i

(8.25)

while for functions b1i , b2i and b3i it holds

ω0 j j
b1i = − 2Ji
[2G ii x3i + nj=1,j=i x3 G ij sin(x1i − x1 − αij )] Td1
oi
ω0
b2i = − 2J i
G i2 sin(x1i − x12 − αi2 ) Td1 (8.26)
o2
ω0
b3i = − 2J i
G i3 sin(x1i − x13 − αi3 ) Td1
o3

Finally for the disturbance term one has d̃ i = − D2Ji ω20 Pmi + ω0 i
Ṗ
2Ji m
i
350 8 Synchronization and Stabilization of Distributed Power Generation Units

Thus, one has the following description of the dynamics of the i-th power generator

ż1i = z2i
ż2i = z3i (8.27)
ż3i = ai (x) + b1 i g1 u1 + b2 i g2 u2 + b3 i g3 u3 + d̃ i

For the complete system of the 3 generators one has

ż31 = a1 (x) + b1 1 g1 u1 + b2 1 g2 u2 + b3 1 g3 u3 + d̃ 1
ż32 = a2 (x) + b1 2 g1 u1 + b2 2 g2 u2 + b3 2 g3 u3 + d̃ 2 (8.28)
ż33 = a3 (x) + b1 3 g1 u1 + b2 3 g2 u2 + b3 3 g3 u3 + d̃ 1

or in matrix form
ż3 = fa (x) + Mu + d̃ (8.29)

where z3 = [z31 , z32 , z33 ]T , u = [u1 , u2 , u3 ]T and d̃ = [d̃1 , d̃2 , d̃3 ]T while
⎛ ⎞ ⎛ 1 ⎞
a1 (x) b1 g1 b2 1 g2 b3 1 g3
fa (x) = ⎝a2 (x)⎠ , M = ⎝b1 2 g1 b2 2 g2 b3 2 g3 ⎠ (8.30)
a3 (x) b1 3 g1 b2 3 g2 b3 3 g3

Setting v = fa (x) + Mu + d̃, one obtains again the linear canonical form for the
i-th generator given by
⎛ i⎞ ⎛ ⎞⎛ i⎞ ⎛ ⎞
ż1 010 z1 0
⎝ż2i ⎠ = ⎝0 0 1⎠ ⎝z2i ⎠ + ⎝0⎠ (vi + d̃ i ) (8.31)
ż3i 000 z3i 1

In this manner the initial nonlinear power system is transformed into three decoupled
linear subsystems which are in the canonical Brunovksy form. For each one of these
subsystems the appropriate control law is
i
vi = zd(3) − k3 (z̈i − z̈di ) − k2 (żi − żdi ) − k1 (zi − zdi ) − d̃ i (8.32)

Since the disturbance term d̃ i , which is due to the time-varying mechanical input
torque at the i-th generator is unknown, in the computation of the control input of Eq.
(8.110) it will be substituted by its estimate d̃ˆ i , which will be provided by a disturbance
observer. For the implementation of the distributed power generation control scheme,
the controller at the i-th power generator makes use of not only its own state vector
X i = [x1i , x2i , x3i ]T , but also of the state vectors of the rest n − 1 power generators,
8.2 State Estimation-Based Control of Distributed PMSGs 351

j j j
i.e. x j = [x1 , x2 , x3 ], j=i, j = 1, 2, . . . , n. The transformation of the dynamical
model of the local power generators into the linear canonical form enables to obtain
estimates of the transformed state vector of the system Y i = [yi , ẏi , ÿi ]T where
yi = Δδ i , through Kalman Filtering, after processing measurements of only the turn
angle difference Δδ i of the i-th power generator. Therefore one has to compute the
estimates X̂ i = [x̂1i , x̂2i , x̂3i ]T after using the estimate provided by the Kalman Filter
Ŷ i = [ŷi , ẏˆ i , ÿˆ i ]T . As already explained, this is a highly data driven application and
requires the exchange and processing in real-time of a large amount of data. It holds
that
x̂1i = ŷi x̂2i = ẏˆ i (8.33)

while for the computation of x̂3i for i = 1, 2, . . . , n one has to solve with respect to
x̂3i the system of equations

D1 ˆ 1
ÿˆ 1 = − 2J ω0 ω0 2 j
1
ẏ + P
2J1 m1
− 2J1
+ +x̂31 nj=1,j=1 [x̂3 G 1j sin(y1 − yj − α1j )]
[G 11 x31

ÿˆ 2 = − 2J
Di ˆ 2 ω0 ω0 2 j
2
ẏ + 2J2
Pm2 − 2J2
[G 22 x32 + x32 nj=1,j=2 [x̂3 G 2j sin(y2 − yj − α2j )]
···
n
Dn ˆ n
ÿˆ n = − 2J ω0 ω0 j
n
ẏ + P
2Jn mn
− 2Jn
[G nn x3n 2 + x̂3n j=1,j=n [x̂3 G nj sin(y
i
− yj − αij )]
(8.34)
The computation of state estimates for the initial MIMO nonlinear system of the
interconnected power generators, out of the estimates obtained for its linearized
equivalent, can be also formulated as an optimization problem. To obtain state esti-
mates for the initial system one comes against a set of coupled nonlinear equations
described in Eq. (8.34). The latter can be solved with respect to the state vector
elements of the initial system using ordinary computation tools for optimization and
nonlinear programming [251, 307]. Moreover, by generalizing the results about state
estimation in stand-alone synchronous generator to the distributed MIMO case of the
interconnected electric power generators, it is possible to compute the state vector
estimate for the initial system, through a filter of the form described in Eq. (3.15) that
uses the inverse Jacobian matrix of the linearization transformation in its observation
gain.
The proposed control scheme, considers as output of the synchronous generator
the rotor’s difference angle Δδ. However, it is possible to avoid the use of encoder
readings about the rotor’s turn angle, and to indirectly estimate this parameter through
the processing of measurement coming from PMUs. Such measurements are the bus
voltage magnitude and the associated phase angle, the line current magnitude and
the associated angle and the electrical output power at the terminal bus [111, 130,
158, 186].
352 8 Synchronization and Stabilization of Distributed Power Generation Units

8.2.5 Estimation of PMSG Disturbance Input with Kalman

Filtering

8.2.5.1 State Estimation with the Derivative-Free Nonlinear Kalman

Filter

As already shown in Chap. 3, with the use of differential flatness theory the nonlinear
model of each PMSG can be written in the canonical form of Eqs. (3.11) and (3.54).
Thus one has the linear model

ẏf = Af yf + Bf v
(8.35)
zf = Cf yf

where yf = [y1 , y2 , y3 ]T and matrices Af , Bf , Cf are in the canonical form

⎛ ⎞ ⎛ ⎞ ⎛ ⎞
010 0 1
Af = ⎝0 0 1⎠ Bf = ⎝0⎠ CfT = ⎝0⎠ (8.36)
000 1 0

and the measurable variable y1 = δ is associated with the turn angle of the rotor. After
carrying out discretization of matrices Af , Bf and Cf with common discretization
methods one can perform Kalman Filtering on the linear equivalent model. This is
the Derivative-free nonlinear Kalman filtering [352, 353, 371].

8.2.5.2 Kalman Filter-Based Estimation of Disturbances

The procedure for identifying additive disturbance inputs that affect the individual
power generators have been explained in Chap. 3. In the most generic case it can be
assumed that the mechanical input torque Pm varies in time [143, 208]. In that case
the aggregate disturbance input exerted on each generator’s model is
ω0
Tm = −ω0 (2J)
D
2 Pm + Ṗ
2J m (8.37)

It is also assumed that the dynamics of the disturbance term Tm is defined by its
n-th order derivative Tm(n) . Considering now that after expressing the system’s state
variables as functions of the flat outputs and their derivatives the PMSG’s dynamics
is given by
y(3) = fc (y, ẏ, ÿ) + gc (y, ẏ, ÿ)u + 2H
1
Ṗm or
(3) D 2 ω0
y = v − ω0 (2J) Pm + 2J Ṗm or (8.38)
(3)
y = v − Tm
8.2 State Estimation-Based Control of Distributed PMSGs 353

where
fc (y, ẏ, ÿ) = ( 2JD2 )ẏ − ω0 2J
D Pm
2J
+ D
ω V s x3 sin(ẏ)+
(2J)2 0 xd

ω0 Vs 1 −x
x sin(y) − ω2J0 xV s T1do dx d Vs cos(y)sin(y)−
x

2J xd Td 3 (8.39)
d d
ω0 Vs
− 2H x
x3 cos(y)ẏ
d

D ω0 Vs
gc (y, ẏ, ÿ) = − 2M Tdo x 
sin(y) (8.40)
d

⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
ẏ1 010 y1 0  
⎝ẏ2 ⎠ = ⎝0 0 1⎠ ⎝y2 ⎠ + ⎝0⎠ v − Tm (8.41)
ÿ3 000 y3 1

Next, the state vector of the model of Eq. (8.41) is extended to include as additional
state variables the disturbance input Tm [169, 292]. Then, in the new state-space
ω0
description one has z1 = y1 , z2 = y2 , z3 = y3 , z4 = Tm = −ω0 (2J) D
2 Pm + 2J Ṗm ,

z5 = Ṫm , and z6 = T̈m . Without loss of generality, the disturbance input dynamics is
assumed to be described by its third order derivative ż6 = Tm(3) .
Since the dynamics of the disturbance input are taken to be unknown in the design
of the associated disturbances’ estimator one has the following dynamics:

żo = Ão ·z + B̃o ·ṽ + K(Co z − Co ẑ) (8.42)

 T
where ṽ = v Tm(3) , K∈R6×1 is the state estimator’s gain and
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
010000 00 1
⎜0 0 1 0 0 0 ⎟ ⎜0 0⎟ ⎜0 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 1 0 0 ⎟ ⎜ 1 0 ⎟ T ⎜0 ⎟
⎜
Ão = ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎟ B̃o = ⎜0 0⎟ C̃o = ⎜0⎟ (8.43)
⎜0 0 0 0 1 0 ⎟ ⎜ ⎟ ⎜ ⎟
⎝0 0 0 0 0 1⎠ ⎝0 0⎠ ⎝0 ⎠
000000 00 0

and the measurable state variable is z1 . It can be confirmed that the disturbance
observer model of the PMSG defined in Eq. (8.43) is observable. Defining as Ãd ,
B̃d , and C̃d , the discrete-time equivalents of matrices Ão , B̃o and C̃o respectively, a
Derivative-free nonlinear Kalman Filter can be designed for the aforementioned rep-
resentation of the system dynamics. The associated Kalman Filter-based disturbance
estimator is given by [34, 209, 334, 341, 363]
measurement update:

K(k) = P− (k)C̃dT [C̃d ·P− (k)C̃dT + R]−1

ẑ(k) = ẑ− (k) + K(k)[z(k) − C̃d ẑ− (k)] (8.44)
P(k) = P− (k) − K(k)C̃d P− (k)
354 8 Synchronization and Stabilization of Distributed Power Generation Units

time update:

P− (k + 1) = Ãd (k)P(k)ÃTd (k) + Q(k)

(8.45)
ẑ− (k + 1) = Ãd (k)ẑ(k) + B̃d (k)ṽ(k)

To compensate for the effects of the disturbance terms it suffices to use in the control
loop the modified control input, which actually compensates for the effects of the
external disturbance term Tm . This gives v∗ = v − T̂m or v∗ = v − ẑ4 .

8.2.6 Simulation Experiments

To evaluate the performance of the proposed nonlinear control scheme, that uses
Kalman Filtering to estimate the nonmeasurable state vector elements of the distrib-
uted PMSGs and the external disturbances, simulation experiments have been carried
out. Different rotation speed setpoints had been assumed. Moreover, different input
torques (mechanical input power profiles) have been assumed to affect the dynamic
model of each local PMSG. Indicative ratings of the PMSG model are as follows:
direct axis synchronous reactance xd = 2.1 p.u., quadrature axis synchronous reac-
tance xd = 0.4 p.u., infinite bus voltage Vs = 1.0 p.u., direct axis open circuit time
constant Tdo = 5 s, rotor’s moment of inertia J = 50 kgm2 . The nonlinear control
scheme works efficiently even for distributed synchronous generators with uneven
ratings.
The case of distributed PMSG operation under unknown input power (torque)
has been examined. The input power at each local generator was considered to be a
disturbance input to the PMSG model and it was assumed that its change in time was
defined by the third derivative of the associated variable, i.e. Tm(3) where, according
ω0
to Eq. (8.37), Tm = −ω0 (2J) D
2 Pm + 2J Ṗm . The disturbance dynamics was completely

unknown to the controller and its identification was performed in real time by the
disturbance estimator. It is shown that the derivative-free nonlinear Kalman Filter
redesigned as a disturbance observer is capable of estimating simultaneously the
nonmeasurable state variables of the generator (ω and Eq ), as well as the unknown
disturbance input Tm . A first set of results is concerned with the tracking performance
of the control loop in case of piecewise constant disturbance input. The real value of
the state variable of the generator is plotted with blue color, the estimated values is
given with green color, while the associated setpoint is printed in red color. This is
shown in Figs. 8.2a, 8.3a and 8.4a. The estimation of the piecewise constant distur-
bance input for the first generator is shown in Fig. 8.2b, for the second generator is
given in Fig. 8.3b, while for the third generator it is shown in Fig. 8.4b.
Additionally, results about the tracking performance of the control loop in case of
a second piecewise constant disturbance input are shown Figs. 8.3a, 8.5a and 8.7a.
The estimation of the piecewise constant disturbance input for the first generator is
shown in Fig. 8.5b, for the second generator is given in Fig. 8.6b, while for the third
8.2 State Estimation-Based Control of Distributed PMSGs 355

(a) Generator 1
(b) Generator 1
0.15 0.35

Estimation of input torque T m (p.u.)

0.3
0.1

0.25

0.05
Δω (p.u.)

0.2

0.15
0

0.1
−0.05
0.05

−0.1 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.2 Sensorless control of the PMSG No 1 under non-measurable (piecewise constant) mechan-
ical input torque in case of speed reference setpoint 1: a convergence of the real and estimated values
of the angular speed difference Δω b estimation of the external mechanical input torque Tm1

(a) Generator 2
(b) Generator 2
0.12 0.25

0.1
Estimation of input torque Tm (p.u.)

0.2
0.08

0.06
0.15
Δω (p.u.)

0.04

0.1
0.02

0
0.05
−0.02

−0.04 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.3 Sensorless control of the PMSG No 2 under non-measurable (piecewise constant) mechan-
ical input torque in case of speed reference setpoint 1: a convergence of the real and estimated values
of the angular speed difference Δω b estimation of the external mechanical input torque Tm2

generator it is shown in Fig. 8.7b. The units of the PMSG state variables have been
expressed again in p.u. (per unit system).
The simulation experiments have confirmed that the proposed state estimation-
based control scheme not only enables implementation of distributed PMSG control
through the measurement of a small number of variables (e.g. of only the rotor’s turn
356 8 Synchronization and Stabilization of Distributed Power Generation Units

(a) Generator 3
(b) Generator 3
0.25 0.7

0.2
0.6

Estimation of input torque Tm (p.u.)

0.15
0.5
0.1
Δω (p.u.)

0.05 0.4

0 0.3

−0.05
0.2
−0.1
0.1
−0.15

−0.2 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.4 Sensorless control of the PMSG No 3 under non-measurable (piecewise constant) mechan-
ical input torque in case of speed reference setpoint 1: a convergence of the real and estimated values
of the angular speed difference Δω b estimation of the external mechanical input torque Tm3

(a) Generator 1 (b) Generator 1

0.25 0.5
Estimation of input torque Tm (p.u.)

0.2
0.4
0.15

0.3
Δω (p.u.)

0.1

0.05
0.2

0
0.1
−0.05

−0.1 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.5 Sensorless control of the PMSG No 1 under non-measurable (piecewise constant) mechan-
ical input torque in case of speed reference setpoint 2: a convergence of the real and estimated values
of the angular speed difference Δω b estimation of the external mechanical input torque Tm1

angle) but also improves the robustness of the PMSG control loop in case of varying
speed setpoints and varying mechanical input torque.
Comparison to the Extended Kalman Filter, which is the most commonly used
nonlinear estimation technique, has shown that the Derivative-free nonlinear Kalman
Filter succeeds estimation of smaller variance for both the nonmeasurable state vector
elements of the PMSG and the term Tm expressing external torques and disturbances
8.2 State Estimation-Based Control of Distributed PMSGs 357

(a) Generator 2
(b) Generator 2
0.2 0.35

Estimation of input torque Tm (p.u.)

0.3
0.15
0.25

0.1
Δω (p.u.)

0.2

0.15
0.05

0.1

0
0.05

−0.05 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.6 Sensorless control of the PMSG no 2 under non-measurable (piecewise constant) mechan-
ical input torque in case of speed reference setpoint 2: a convergence of the real and estimated values
of the angular speed difference Δω b estimation of the external mechanical input torque Tm2

(a) Generator 3
(b) Generator 3
0.4 0.8
Estimation of input torque Tm (p.u.)

0.3 0.7

0.6
0.2

0.5
Δω (p.u.)

0.1
0.4
0
0.3

−0.1
0.2

−0.2 0.1

−0.3 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.7 Sensorless control of the PMSG No 3 under non-measurable (piecewise constant) mechan-
ical input torque in case of speed reference setpoint 2: a convergence of the real and estimated values
of the angular speed difference Δω b estimation of the external mechanical input torque Tm3

inputs. Estimation with the Extended Kalman Filter was based on local linearization
of the PMSG model and on the computation of the Jacobian of Eq. (8.6). The obtained
results are summarized in Table 8.1, and correspond to experiments (a) and (b)
associated with the two turn speed setpoints given in Figs. 8.2, 8.3, 8.4, 8.5, 8.6 and
8.7 under sinusoidal variations of Tm , and considering different noise levels.
358 8 Synchronization and Stabilization of Distributed Power Generation Units

Table 8.1 RMSE for nonlinear filters

DKFa EKFa DKFb EKFb
Δδ 0.0006 0.0007 0.0011 0.0012
Δω 0.0061 0.0093 0.0069 0.0119
Eq 0.0167 0.1654 0.0376 0.2253
Tm 0.0172 0.0843 0.0184 0.0810
Ṫm 0.0137 0.0374 0.0146 0.0339
T̈m 0.0054 0.0083 0.0057 0.0080

8.3 Nonlinear H-Infinity Control of Distributed

Synchronous Generators

8.3.1 Overview

In this section a new nonlinear H-infinity optimal control method for distributed
synchronous generators is developed and specific results are obtained for the problem
of synchronization and stabilization of a three-area multi-machine power system. The
nonlinear dynamical model of the interconnected synchronous generators undergoes
linearization round temporary operating points which are computed at each iteration
of the control algorithm. These operating points consist of the present value of the
system’s state vector and of the last value of the control inputs vector that was exerted
on it. This linearization procedure is based on the concept of Taylor series expansion
of the system’s states-space model and on the computation of the associated Jacobian
matrices [34, 341, 344]. The modelling error due to this approximate linearization is
considered to be a perturbation that is compensated by the robustness of the control
scheme [352, 376, 380].
The feedback control input is based on the solution of an H-infinity control prob-
lem [352, 376, 380]. This stands for a mini-max differential game between the control
signal and the disturbance terms that affect the model of the distributed generators.
The feedback control gain is found from the solution of an algebraic Riccati equation
at each iteration of the control algorithm [352, 376, 380]. The stability of the con-
trol scheme is demonstrated through Lyapunov analysis. Actually, it is shown that
the control loop satisfies an H-infinity tracking performance criterion which signi-
fies elevated robustness against model uncertainty and perturbation inputs. Besides,
under moderate conditions the asymptotic stability of the control scheme is proven.
Through simulation experiments it is further confirmed that the proposed control
method assures fast and accurate tracking of reference setpoints.
8.3 Nonlinear H-Infinity Control of Distributed Synchronous Generators 359

8.3.2 Dynamic Model of the Multi-machine Power System

8.3.2.1 State-Space Description of the Distributed SGs Model

The state-space model of the multi-machine power system was analyzed in the previ-
ous sections. A multi-machine power system (MMPS) with n machines, in which the
first machine is chosen as the reference machine can be described by the following
nonlinear dynamic model [352, 375]

δ̇i = ωi − ω0
P
ω̇i = − 2Ji (ωi − ω0 ) + ω0 2Jmii −
Di
(8.46)
−ω0 2J1 i [G ii Eqi 2 + Eqi nj=1,j=i Eqj G ij sin(δi − δj − αij )]
xd −x 
Ėq i = − T1 Eq i + Td1 xi  di Vsi cos(Δδi ) + Td1 Ef i
di oi d i oi

where the electric torque Pei which is associated with the active power at the i-th
generator is now given by
n
Pei = G ii Eqi 2 + Eqi 
j=1,j=i Eqj G ij sin(δi − δj − αij ) (8.47)

for i = 1, 2 . . . , n. For a power grid that consists of n generators the aggregate state
vector comprises the state vectors of the local machines, i.e. x = [x 1 , x 2 , . . . , x n ]T ,
where x i = [x1i , x2i , x3i ]T , with x1i = Δδi , x2i = Δωi and x3i = Eqi are the state
variables for the i-th machine and i = 1, 2, . . . , n.
Without loss of generality the three-machine power system of Fig. 8.8 is consid-
ered. The associated dynamic model is given by:

ẋ1 = x2 − ω0 (8.48)

P
ẋ2 = − 2JD1
(x2 − ω0 ) + ω0 2Jm11 −
ω0
1 (8.49)
− 2J1
{G 11 x32 + x3 [x6 G 12 sin(x1 − x4 − a12 ) + x9 G 13 sin(x1 − x7 − a13 )]


1 xd1 −xd1
ẋ3 = − T1 x3 + Tdo1 xd
Vs cos(x1 ) + 1
Td o 1
u1 (8.50)
d1 1

ẋ4 = x5 − ω0 (8.51)

P
ẋ5 = − 2JD2
(x5 − ω0 ) + ω0 2Jm22 −
ω0
2 (8.52)
− 2J2
{G 22 x62 + x6 [x3 G 21 sin(x4 − x1 − a21 ) + x9 G 23 sin(x4 − x7 − a23 )]
360 8 Synchronization and Stabilization of Distributed Power Generation Units

Fig. 8.8 A 3-area distributed power generation model consisting of 3 synchronous machines


1 xd2 −xd2
ẋ6 = − T1 x6 + Tdo2 xd
Vs cos(x4 ) + 1
Td o 2
u2 (8.53)
d2 2

ẋ7 = x8 − ω0 (8.54)

P
ẋ8 = − 2JD2
(x8 − ω0 ) + ω0 2Jm33 −
ω0
3 (8.55)
− 2J3
{G 32 x92 + x9 [x3 G 31 sin(x7 − x1 − a31 ) + x6 G 32 sin(x7 − x4 − a32 )]


1 xd3 −xd3
ẋ9 = − T1 x9 + Tdo3 xd
Vs cos(x7 ) + 1
Td o 3
u3 (8.56)
d3 3

The system is also written in the matrix form:

8.3 Nonlinear H-Infinity Control of Distributed Synchronous Generators 361
⎛ ⎞
ẋ1 ⎛ ⎞ ⎛ ⎞
⎜ẋ2 ⎟ f1 (x) 0 0 0
⎜ ⎟ ⎜f2 (x)⎟ ⎜ 0 0 0 ⎟
⎜ẋ3 ⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜f3 (x)⎟ ⎜g1 (x) 0 0 ⎟
⎜ẋ4 ⎟ ⎜ ⎟ ⎜ ⎟⎛ ⎞
⎜ ⎟ ⎜f4 (x)⎟ ⎜ 0 0 0 ⎟
⎜ẋ5 ⎟ ⎜ ⎟ ⎜ ⎟ u1
⎜ ⎟ = ⎜f5 (x)⎟ + ⎜ 0 0 0 ⎟ ⎝u2 ⎠ (8.57)
⎜ẋ6 ⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜f6 (x)⎟ ⎜ 0 g2 (x) 0 ⎟ u3
⎜ẋ6 ⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎜f7 (x)⎟ ⎜ 0 0 0 ⎟
⎜ẋ7 ⎟ ⎜ ⎟ ⎜ ⎟
⎜ ⎟ ⎝f8 (x)⎠ ⎝ 0 0 0 ⎠
⎝ẋ8 ⎠
f9 (x) 0 0 g3 (x)
ẋ9

where
f1 (x) = x2 − ω0 (8.58)

P
f2 (x) = − 2JD1
(x2 − ω0 ) + ω0 2Jm11 −
ω0
1 (8.59)
− 2J1
{G 11 x32 + x3 [x6 G 12 sin(x1 − x4 − a12 ) + x9 G 13 sin(x1 − x7 − a13 )]
xd1 −xd
f3 (x) = − T1 x3 + 1
Td o 1 xd
1
Vs cos(x1 ) (8.60)
d1 1

f4 (x) = x5 − ω0 (8.61)

P
f5 (x) = − 2JD2
(x5 − ω0 ) + ω0 2Jm22 −
ω0
2 (8.62)
− 2J2 {G 22 x6 + x6 [x3 G 21 sin(x4 − x1 − a21 ) + x9 G 23 sin(x4 − x7 − a23 )]
2


1 xd2 −xd2
f6 (x) = − T1 x6 + Tdo2 xd
Vs cos(x4 ) (8.63)
d2 2

f7 (x) = x8 − ω0 (8.64)

P
f8 (x) = − 2JD2
(x8 − ω0 ) + ω0 2Jm33 −
ω0
3 (8.65)
− 2J3 {G 32 x9 + x9 [x3 G 31 sin(x7 − x1 − a31 ) + x6 G 32 sin(x7 − x4 − a32 )]
2


1 xd3 −xd3
f9 (x) = − T1 x9 + Tdo3 xd
Vs cos(x7 ) (8.66)
d3 3

while one has that

g1 = 1
Td o 1
g2 = 1
Td o 2
g3 = 1
Td o 3
u3 (8.67)
362 8 Synchronization and Stabilization of Distributed Power Generation Units

8.3.3 Linearization of the Model of the Distributed

Synchronous Generators

Local linearization is performed for the state-space model of the distributed power
generators, though Taylor series expansion, round the operating point (x ∗ , u∗ ) where
x ∗ is the present value of the system’s state vector and u∗ is the last value of the control
input that was exerted on the machine. Thus, one obtains the linearized description

ẋ = Ax + Bu + d̃ (8.68)

For the previous description of the distributed power generators’ model (3-area
3-machine model) by the state-space equation of Eq. (8.57) it holds that A =
∇x ∇x [f (x) + g(x)u] with
⎛ ∂f1 ∂f1 ⎞
∂f1 ∂f1
∂x1
· · · ∂x
∂x29 ∂x3
⎜ ∂f2 ∂f2 ⎟
∂f2
· · · ∂x ∂f2
⎜ ∂x1 ∂x29⎟ ∂x3
⎜ ∂f3 ∂f3 ⎟
∂f3 ∂f3
⎜ ∂x1 · · · ∂x
∂x2 ⎟∂x3
⎜ ∂f4 ∂f4 ⎟
9
A = ∇x [f (x) + g(x)u] = ⎜ ∂f4
· · · ∂x ⎟∂f4 (8.69)
⎜ ∂x1 ∂x29⎟ ∂x3
⎜· · · · · · · · · · · · · · ·⎟
⎜ ⎟
⎝· · · · · · · · · · · · · · ·⎠
∂f5 ∂f5 ∂f5 ∂f5
∂x1 ∂x2 ∂x3
· · · ∂x 9

∂f1
For the first row of the Jacobian matrix A = ∇x [f (x) + g(x)u] one has: ∂x1
= 0,
∂f1 ∂f1 ∂f1 ∂f1 ∂f1 ∂f1 ∂f1 ∂f1
∂x2
= 1, ∂x3
= 0, ∂x4
= 0, ∂x5
= 0, ∂x6
= 0, ∂x7
= 0, ∂x8
= 0, ∂x9
= 0.

For the second row of the Jacobian matrix A = ∇x [f (x) + g(x)u] one has:
∂f2 ω0 ∂f2
∂x1
= − 2J1
x3 [x6 G 12 cos(x1 − x4 − a13 )] + x9 G 13 cos(x1 − x7 − a13 ), ∂x2
= − 2J
D1
1
,
∂f2 −ω0
∂x3
= 2J1
{G 11 2x3 + [x6 ]G 12 sin(x1 − x4 − a12 ) + x9 G 13 sin(x1 − x7 − a13 )]},
∂f2 ω0 ∂f2 ∂f2 ω0
∂x4
= x x G cos(x1
2J1 3 6 12
− x4 − a12 ), ∂x5
= 0, ∂x6
= − 2J1
x3 G 12 sin(x1 − x4 − a12 ),
∂f2 ω0 ∂f2 ∂f2 ω0
∂x7
= x x G cos(x1
2J1 3 9 13
− x7 − a13 ), ∂x8
= 0, ∂x9
= − 2J1
x3 G 13 sin(x1 − x7 − a13 ).
∂f3
For the third row of the Jacobian matrix A = ∇x [f (x) + g(x)u] one has: ∂x1
=

1 xd1 −xd1 ∂f3 ∂f3 ∂f3 ∂f3 ∂f3 ∂f3
− 
Tdo1 xd
Vs sin(x1 ), ∂x2
= 0, ∂x3
= − T1 , ∂x4
= 0, ∂x5
= 0, ∂x6
= 0, ∂x7
= 0,
1 d1
∂f3 ∂f3
∂x8
= 0, ∂x9
= 0.
∂f4
For the fourth row of the Jacobian matrix A = ∇x [f (x) + g(x)u] one has: ∂x1
= 0,
∂f4 ∂f4 ∂f4 ∂f4 ∂f4 ∂f1 ∂f1 ∂f1
∂x2
= 0, ∂x3
= 0, ∂x4
= 0, ∂x5
= 1, ∂x6
= 0, ∂x7
= 0, ∂x8
= 0, ∂x9
= 0.
∂f5
For the fifth row of the Jacobian matrix A = ∇x [f (x) + g(x)u] one has: ∂x1
=
ω0 ∂f5 ∂f5
x [x G cos(x4 − x1 − a21 )
2J2 6 3 21
+ x3 G 23 cos(x4 − x7 − a23 )], ∂x2
= 0, ∂x3
=
ω0 ∂f5 ω0
− 2J 2
x6 G 21 sin(x4 −x1 −a21 ), ∂x4
= − 2J2
x6 [x3 G 21 cos(x4 −x1 −a21 )+x9 G 23 cos(x4 −
8.3 Nonlinear H-Infinity Control of Distributed Synchronous Generators 363

∂f5 D2 ∂f5 ω0
x7 −a23 )], ∂x5
= − 2J ,
2 ∂x6
= − 2J2
{G 22 2x6 +[x3 G 21 sin(x4 −x1 −a21 )+x9 G 23 sin(x4 −
∂f5 ω0 ∂f5 ∂f5 ω0
x7 − a23 )]}, ∂x7
= 2J2
x6 x9 G 23 cos(x4 − x7 − a23 ), ∂x8
= 0, ∂x9
= − 2J2
x6 G 23 sin(x4 −
x7 − a23 ).
∂f6
For the sixth row of the Jacobian matrix A = ∇x [f (x) + g(x)u] one has: ∂x 1
= 0,
∂f6 ∂f6 ∂f6 xd2 −xd ∂f6 ∂f6 ∂f6 ∂f6
∂x2
= 0, ∂x3
=, ∂x4
= − Td1 xd
2
Vs sin(x4 ), ∂x5
= 0, ∂x6
= − T1 , ∂x7
= 0, ∂x8
= 0,
o2 2 d2
∂f6
∂x9
= 0.
∂f7
For the seventh row of the Jacobian matrix A = ∇x [f (x)+g(x)u] one has: ∂x1
= 0,
∂f7 ∂f7 ∂f7 ∂f7 ∂f7 ∂f7 ∂f7 ∂f7
∂x2
= 0, ∂x3
= 0, ∂x4
= 0, ∂x5
= 0, ∂x6
= 0, ∂x7
= 0, ∂x8
= 1, ∂x9
= 0.
∂f8
For the eight row of the Jacobian matrix A = ∇x [f (x) + g(x)u] one has: ∂x1
=
ω0 ∂f8 ∂f8
x [x G cos(x7 − x1 − a31 ) + x6 G 32 cos(x7 − x4 − a32 )], ∂x
2J3 9 3 31 2
= 0, = ∂x3
ω0 ∂f8 ω0 ∂f8 ∂f8
− 2J3 x9 G 31 sin(x7 − x1 − a31 ), ∂x4 = 2J3 x9 x6 G 32 cos(x7 − x4 − a32 ), ∂x5 = 0, = ∂x6
ω0 ∂f8 ω0
− 2J3 x9 G 23 sin(x7 −x4 −a32 ), ∂x7 = − 2J3 x9 [x3 G 31 cos(x7 −x1 −a31 )+x6 G 32 cos(x7 −
∂f8 D3 ∂f8 ω0
x4 −a32 )], ∂x 8
= − 2J3 , ∂x9 = − 2J 3
{G 33 2x9 +[x3 G 31 sin(x7 −x1 −a31 )+x6 G 32 sin(x7 −
x4 − a32 )]}.
∂f9
For the ninth row of the Jacobian matrix A = ∇x [f (x) + g(x)u] one has: ∂x 1
= 0,

∂f9 ∂f9 ∂f9 ∂f9 ∂f9 x −x ∂f9
= 0, ∂x = 0, ∂x = 0, ∂x = 0, ∂x = − Td1 x3  d3 Vs sin(x7 ), ∂x = 0, and
d
∂x2 3 4 6 7 o3 d 8
3
∂f9
finally ∂x9
= − T1 .
d3
Moreover, it holds that B = ∇u ∇x [f (x) + g(x)u] with
⎛ ⎞
0 0 0
⎜ 0 0 0 ⎟
⎜ ⎟
⎜g1 (x) 0 0 ⎟
⎜ ⎟
⎜ 0 0 0 ⎟
⎜ ⎟
B = ∇u [f (x) + g(x)u] = ⎜
⎜ 0 0 0 ⎟ ⎟ (8.70)
⎜ 0 g2 (x) 0 ⎟
⎜ ⎟
⎜ 0 0 0 ⎟
⎜ ⎟
⎝ 0 0 0 ⎠
0 0 g3 (x)

8.3.4 The Nonlinear H-Infinity Control

8.3.4.1 Mini-Max Control and Disturbance Rejection

The design of an H-infinity controller was clearly analyzed in the case of stand-alone
synchronous generators. The initial nonlinear model of the distributed synchronous
generators is in the form
ẋ = f (x, u) x∈Rn , u∈Rm (8.71)
364 8 Synchronization and Stabilization of Distributed Power Generation Units

Linearization of the system (distributed synchronous generators) is performed at

each iteration of the control algorithm round its present operating point (x ∗ , u∗ ) =
(x(t), u(t − Ts )). The linearized equivalent of the system is described by

ẋ = Ax + Bu + L d̃ x∈Rn , u∈Rm , d̃∈Rq (8.72)

where matrices A and B are obtained from the computation of the Jacobians
⎛ ∂f ⎞ ⎛ ∂f ∂f ⎞
1 ∂f1 ∂f1 ∂f1
∂x ∂x
· · · ∂x 1
∂u1 ∂u2
1
· · · ∂u
⎜ ∂f21 ∂f22 n
∂f2 ⎟ ⎜ ∂f ∂f ∂f2 ⎟
m

⎜ ∂x1 ∂x2 · · · ∂x ⎟ ⎜ 2 2 · · · ∂u ⎟
A= ⎜ n ⎟ | ∗ ∗ B = ⎜ ∂u1 ∂u2
(x ,u ) m⎟| ∗ ∗
(x ,u ) (8.73)
⎝· · · · · · · · · · · ·⎠ ⎝· · · · · · · · · · · ·⎠
∂fn ∂fn ∂fn ∂fn ∂fn ∂fn
∂x1 ∂x2
· · · ∂xn ∂u1 ∂u2
· · · ∂um

and vector d̃ denotes disturbance terms due to linearization errors. The problem of
disturbance rejection for the linearized model that is described by

ẋ = Ax + Bu + L d̃
(8.74)
y = Cx

where x∈Rn , u∈Rm , d̃∈Rq and y∈Rp , cannot be handled efficiently if the classical
LQR control scheme is applied. This because of the existence of the perturbation
term d̃. The disturbance term d̃ apart from modeling (parametric) uncertainty and
external perturbation terms can also represent noise terms of any distribution.
In the H∞ control approach, a feedback control scheme is designed for trajectory
tracking by the system’s state vector and simultaneous disturbance rejection, con-
sidering that the disturbance affects the system in the worst possible manner. The
disturbances’ effect are incorporated in the following quadratic cost function:

1 T T
J(t) = 2 0
[y (t)y(t) + ruT (t)u(t) − ρ 2 d̃ T (t)d̃(t)]dt, r, ρ > 0 (8.75)

As in the case of nonlinear H-infinity control for stand-alone generators, in the case
of distributed synchronous generators the significance of the negative sign in the
cost function’s term that is associated with the perturbation variable d̃(t) is that the
disturbance tries to maximize the cost function J(t) while the control signal u(t) tries
to minimize it. The physical meaning of the relation given above is that the control
signal and the disturbances compete to each other within a mini-max differential
game. This problem of mini-max optimization can be written as

minu maxd̃ J(u, d̃) (8.76)

Again the objective of the optimization procedure is to compute a control signal u(t)
which can compensate for the worst possible disturbance, that is externally imposed
to the system. However, the solution to the mini-max optimization problem is directly
8.3 Nonlinear H-Infinity Control of Distributed Synchronous Generators 365

related to the value of the parameter ρ. This means that there is an upper bound in
the disturbances magnitude that can be annihilated by the control signal.

8.3.4.2 H-Infinity Feedback Control

For the linearized system of the distributed generators given by Eq. (8.74) the cost
function of Eq. (8.75) is defined, where the coefficient r determines the penaliza-
tion of the control input and the weight coefficient ρ determines the reward of the
disturbances’ effects.
It is assumed that (i)The energy that is transferred from the disturbances signal
∞
d̃(t) is bounded, that is 0 d̃ T (t)d̃(t)dt < ∞, (ii) the matrices [A, B] and [A, L] are
stabilizable, (iii) the matrix [A, C] is detectable. Then, the optimal feedback control
law is given by
u(t) = −Kx(t) (8.77)

with
K = 1r BT P (8.78)

where P is a positive semi-definite symmetric matrix which is obtained from the

solution of the Riccati equation

AT P + PA + Q − P( 1r BBT − 1
2ρ 2
LL T )P =0 (8.79)

where Q is also a positive definite symmetric matrix. The worst case disturbance is
given by
d̃(t) = ρ12 L T Px(t) (8.80)

The diagram of the considered control loop is depicted in Fig. 8.9.

8.3.5 Lyapunov Stability Analysis

The Lyapunov stability analysis for the H-infinity control of the distributed generators
follows the procedure that was explained in the case of the stand-alone generators.
Through Lyapunov stability analysis it will be shown that the proposed nonlinear
control scheme assures H∞ tracking performance for the distributed synchronous
generators, and that in case of bounded disturbance terms asymptotic convergence
to the reference setpoints is achieved.
The tracking error dynamics for the distributed synchronous generators is written
in the form
ė = Ae + Bu + L d̃ (8.81)
366 8 Synchronization and Stabilization of Distributed Power Generation Units

Fig. 8.9 Diagram of the control scheme for the distributed synchronous generators

where in the distributed synchronous generators’ case L = I∈R9×9 with I being the
identity matrix. Variable d̃ denotes model uncertainties and external disturbances of
the distributed generators’ model. The following Lyapunov equation is considered

V = 21 eT Pe (8.82)

where e = x − xd is the tracking error. By differentiating with respect to time one

obtains
V̇ = 21 ėT Pe + 21 ePė⇒
(8.83)
V̇ = 2 [Ae + Bu + L d̃]T P + 21 eT P[Ae + Bu + L d̃]⇒
1

V̇ = 21 [eT AT + uT BT + d̃ T L T ]Pe+
(8.84)
+ 21 eT P[Ae + Bu + L d̃]⇒

V̇ = 21 eT AT Pe + 21 uT BT Pe + 21 d̃ T L T Pe+
(8.85)
e PAe + 21 eT PBu + 21 eT PL d̃
1 T
2
8.3 Nonlinear H-Infinity Control of Distributed Synchronous Generators 367

The previous equation is rewritten as

V̇ = 21 eT (AT P + PA)e + ( 21 uT BT Pe + 21 eT PBu)+

(8.86)
+ ( 21 d̃ T L T Pe + 21 eT PL d̃)

Assumption: For given positive definite matrix Q and coefficients r and ρ there exists
a positive definite matrix P, which is the solution of the following matrix equation

AT P + PA = −Q + P( 2r BBT − 1
ρ2
LL T )P (8.87)

Moreover, the following feedback control law is applied to the system

u = − 1r BT Pe (8.88)

By substituting Eqs. (8.87) and (8.88) one obtains

V̇ = 21 eT [−Q + P( 2r BBT − 2ρ1 2 LL T )P]e+

(8.89)
+ eT PB(− 1r BT Pe) + eT PL d̃⇒

V̇ = − 21 eT Qe + ( 2r PBBT Pe − 2ρ1 2 eT PLL T )Pe

(8.90)
− 1r eT PBBT Pe) + eT PL d̃

which after intermediate operations gives

V̇ = − 21 eT Qe − 1 T
2ρ 2
e PLL T Pe + eT PL d̃ (8.91)

or, equivalently
V̇ = − 21 eT Qe − 2ρ1 2 eT PLL T Pe+
(8.92)
+ 21 eT PL d̃ + 21 d̃ T L T Pe

Lemma: The following inequality holds

1 T
2
e L d̃ + 21 d̃L T Pe − 1 T
2ρ 2
e PLL T Pe≤ 21 ρ 2 d̃ T d̃ (8.93)

Proof : The binomial (ρα − ρ1 b)2 is considered. Expanding the left part of the above
inequality one gets

ρ 2 a2 + 1 2
ρ2
b − 2ab ≥ 0 ⇒ 21 ρ 2 a2 + 1 2
2ρ 2
b − ab ≥ 0 ⇒
(8.94)
ab − 1 2
2ρ 2
b ≤ 21 ρ 2 a2 ⇒ 21 ab + 21 ab − 2ρ1 2 b2 ≤ 21 ρ 2 a2

The following substitutions are carried out: a = d̃ and b = eT PL and the previous
relation becomes
1 T T
2
d̃ L Pe + 21 eT PL d̃ − 1 T
2ρ 2
e PLL T Pe≤ 21 ρ 2 d̃ T d̃ (8.95)
368 8 Synchronization and Stabilization of Distributed Power Generation Units

Equation (8.95) is substituted in Eq. (8.92) and the inequality is enforced, thus giving

V̇ ≤ − 21 eT Qe + 21 ρ 2 d̃ T d̃ (8.96)

Equation (8.96) shows that the H∞ tracking performance criterion is satisfied. The
integration of V̇ from 0 to T gives
T T T
V̇ (t)dt≤ − 21 0 ||e||2Q dt + 21 ρ 2 0 ||d̃||2 dt⇒
0 T T (8.97)
2V (T ) + 0 ||e||2Q dt≤2V (0) + ρ 2 0 ||d̃||2 dt

Moreover, if there exists a positive constant Md > 0 such that

∞
0 ||d̃||2 dt ≤ Md (8.98)

then one gets

∞
0 ||e||2Q dt ≤ 2V (0) + ρ 2 Md (8.99)
∞
Thus, the integral 0 ||e||2Q dt is bounded. Additionally, V (T ) is bounded and from
the definition of the Lyapunov function V in Eq. (8.82) it becomes clear that e(t)
will be also bounded since e(t) ∈ e = {e|eT Pe≤2V (0) + ρ 2 Md }. According to
the above and with the use of Barbalat’s Lemma one obtains limt→∞ e(t) = 0.

8.3.6 Robust State Estimation with the Use of the H∞

Kalman Filter

Robust state estimation with the use of the H-infinity Kalman Filter was already
analyzed in the case of multi-phase machines. The control loop has to be implemented
with the use of information provided by a small number of sensors and by processing
only a small number of state variables. To reconstruct the missing information about
the state vector of the distributed synchronous generators it is proposed to use a
filtering scheme and based on it to apply state estimation-based control [375]. The
recursion of the H∞ Kalman Filter, for the model of the distributed SGs, can be
formulated in terms of a measurement update and a time update part
Measurement update:

D(k) = [I − θ W (k)P− (k) + C T (k)R(k)−1 C(k)P− (k)]−1

K(k) = P− (k)D(k)C T (k)R(k)−1 (8.100)
x̂(k) = x̂ − (k) + K(k)[y(k) − C x̂ − (k)]

Time update:
x̂ − (k + 1) = A(k)x(k) + B(k)u(k)
(8.101)
P− (k + 1) = A(k)P− (k)D(k)AT (k) + Q(k)
8.3 Nonlinear H-Infinity Control of Distributed Synchronous Generators 369

where it is assumed that parameter θ is sufficiently small to assure that the covariance
matrix P− (k) − θ W (k) + C T (k)R(k)−1 C(k) will be positive definite. When θ = 0
the H∞ Kalman Filter becomes equivalent to the standard Kalman Filter. One can
measure only a part of the state vector of the distributed SGs, such as state variables
xi = θi , i = 1, 2, 3 and can estimate through filtering the rest of the state vector
elements.

8.3.7 Simulation Tests

The problem of synchronization and stabilization of the distributed synchronous

generators has been solved with the application of a nonlinear H-infinity (optimal)
control method. The performance of the control method is evaluated through sim-
ulation experiments. It can be observed that the control scheme assures fast and
accurate convergence of the rotation speeds of the individual generators to the refer-
ence setpoints, as shown in Figs. 8.10, 8.11 and 8.12. As it happens with all optimal
control methods the transients of the state variables are affected by values of the
controller’s gains, that is r, ρ and by values of the elements of matrix Q appearing
in Eq. (8.79). The values of these parameters affect also the variation ranges of the
control inputs. It can be noted that the H-infinity algorithm exhibited remarkable
robustness to uncertainty in the model of the distributed power generators which was
due to approximate linearization. It also remained robust to measurement noise and
to external perturbations affecting the model of the distributed generators (e.g. vari-
ations in the mechanical torque that excites the rotational motion of the machines).
The computational aspects of the proposed control method have been also
assessed. For the three-area multi-machine power system the dimension of the state
vector was x∈R9 , while the dimension of the feedback gain matrix P that was obtained
from the solution of the Riccati equation given in Eq. (8.79) was P∈R9×9 . Using an
i7 processor at 2.0Ghz the solution of the Riccati equation was performed in msec
and definitely in time that was significantly smaller than the sampling period of the
control algorithm. For a distributed power system, that comprises more generators
and which is of higher dimensionality, the real-time implementation of the control
algorithm is not going to be hindered. It can be concluded that the proposed H-infinity
control method is computationally very efficient.

8.4 Flatness-Based Adaptive Control of Distributed PMSGs

As previously explained, distributed power generation is becoming progressively an

issue of major importance for the functioning of the smart grid [84, 106, 167, 270, 359,
375, 510]. To cover the increasing demand for electric power it has become necessary
to interconnect a large number of power generation units, located at different areas.
For efficient functioning of such a decentralized power generation scheme stability
370 8 Synchronization and Stabilization of Distributed Power Generation Units

(a) (b)
1 1
ω1 (p.u,)

ω 1 (p.u,)
0.5 0.5

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

1 1
ω2 (p.u,)

ω 2 (p.u,)
0.5 0.5

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

1 1
ω3 (p.u,)

ω 3 (p.u,)
0.5 0.5

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

Fig. 8.10 Tracking of reference setpoints by the rotational speeds ωi i = 1, 2, 3 of the distributed
synchronous generators a setpoint 1 (p.u) b setpoint 2 (p.u)

(a) (b)
1 1
ω 1 (p.u,)
ω 1 (p.u,)

0.5 0.5

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

1 1
ω 2 (p.u,)
ω (p.u,)

0.5 0.5
2

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

1 1
ω 3 (p.u,)
ω 3 (p.u,)

0.5 0.5

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

Fig. 8.11 Tracking of reference setpoints by the rotational speeds ωi i = 1, 2, 3 of the distributed
synchronous generators a setpoint 3 (p.u) b setpoint 4 (p.u)

properties should hold (despite modeling uncertainties that may affect the power
generators) and synchronization between the individual power generators should
be succeeded (despite their exposure to external perturbations) [88, 102, 116, 198,
213, 237, 238, 495]. This section proposes adaptive fuzzy control based on differen-
tial flatness theory and using exclusively output feedback, for solving the problem
of synchronization between distributed synchronous generators and for succeeding
stabilization of a distributed power generation system. The model of the individual
8.4 Flatness-Based Adaptive Control of Distributed PMSGs 371

(a) (b)
1 1

ω 1 (p.u,)
ω (p.u,)

0.5 0.5
1

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

1 1

ω (p.u,)
ω (p.u,)

0.5 0.5

2
2

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

1 1

ω 3 (p.u,)
ω (p.u,)

0.5 0.5
3

0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time (sec) time (sec)

Fig. 8.12 Tracking of reference setpoints by the rotational speeds ωi i = 1, 2, 3 of the distributed
synchronous generators a setpoint 5 (p.u) b setpoint 6 (p.u)

power generator units (synchronous power generators) is considered to be completely

unknown while the state vector of these generators is not completely measurable [60,
121, 262, 265, 315]. Thus, finally one has to do with a joint estimation and nonlinear
control problem in which the control signal makes use of the estimated state vector
of the generators [15, 439, 474].
First, the chapter shows that the dynamic model of the distributed power generation
system is a differentially flat one. This means that all its state variables and the
control input can be written as functions of one single algebraic variable, which is
the flat output, and also as functions of the flat outputs derivatives [20, 137, 229,
241, 281, 375, 411, 456]. This change of variables (differential flatness theory-based
diffeomorphism) enables to transform the nonlinear model of the distributed power
generators into the linear canonical (Brunovsky) form. In the latter description of the
power generators system the transformed control inputs contain elements which are
associated with the unknown nonlinear dynamics of the system. These are identified
on-line with the use of neurofuzzy approximators and the estimated system dynamics
is finally used for the computation of the control signal that will make the generators’
state vector track the desirable setpoints. Thus an adaptive fuzzy control scheme is
implemented. The learning rate of the neurofuzzy approximators is determined by
the requirement that the first derivative of the Lyapunov function of the control loop
will always be a negative one [34, 143, 344, 375].
Next, another problem that has to be dealt with is that only output feedback can
be used for the implementation of the distributed power generators’ control scheme
[50, 483]. The nonmeasurable state variables of the power generators have to be
reconstructed with the use of a state estimator (observer), which functions again in
the control loop [111, 112, 197, 249, 251, 266, 307, 320]. Thus, finally, the Lyapunov
372 8 Synchronization and Stabilization of Distributed Power Generation Units

function for the proposed control scheme comprises three quadratic terms: (i) a term
that describes the tracking error of the generators’ state variables, (ii) a term that
describes the error in the estimation of the non-measurable state vector elements
with respect to the reference setpoints, and (iii) a sum of quadratic terms associated
with the distance of the weights of the neurofuzzy approximators from the values that
give the best approximation of the unknown dynamics of the generators. It is proven
that an adaptive (learning) control law can be found assuring that the first derivative
of the Lyapunov function will remain negative, thus assuring that the stability of
the control loop will be preserved and that accurate tracking of the setpoints by the
systems state variables will be succeeded (H-infinity tracking performance) [117,
215, 226, 263].

8.4.1 An Adaptive Fuzzy Control for the System

of the Distributed Synchronous Generators

In the previous sections it was shown that, for the case of the N distributed synchro-
nous generators, the dynamic model of the i-th input-output linearized model of the
synchronous machine is given by
i
y(3) = ai (x) + b1i (x)g1 u1 + b2i (x)g2 u2 + b3i (x)g3 u3 (8.102)

which means that one arrives at a function of the form ż3i = ai (x) + b1i (x)g1 u1 +
b2i (x)g2 u2 + b3i (x)g3 u3 + d̃ i , where in the case of the distributed power generation
that consists of n = 2 machines, and considering for instance i = 1, j = 2 (Fig. 8.13)
one has

) x2 + Di ω02 [G ii x3i + x3i nj=1,j=i x3 G ij sin(x1i − x1 − αij )]−
Di 2 i 2 j j
ai = ( 2J i (2Ji )
i + n
ω0 x −x 
− 2J [G x x
j
G sin(x i − x j − α )(− 1 x i + ( 1 di di V cos(x i ))]−
i
ii 3 j=1,j =i 3 ij 1 1 ij T  3oi T i
d xd  si 1
di
(8.103)
ω0 i n j xd −xd
− 2J x j=1,j=i G ij sin(x1i − x1 − αij )(− T1 x3i + ( Td1 xid  i Vsi cos(x1i ))−
i 3 oi i
ω0 i n ω0 i n
di
j j j j j
− 2J x j=1,j=i x3 G ij cos(x1i − x1 − αij )x2i 2J
i 3
x j=1,j=i x3 G ij cos(x1i − x1 − αij )x2
i 3

while for functions b1i , b2i it holds

ω0 j j
b1i = − 2J [2G ii x3i + nj=1,j=i x3 G ij sin(x1i − x1 − αij )] Td1
i
ω0
oi (8.104)
b2i = − 2J i
G i2 sin(x1i − x12 − αi2 ) Td1
o2

Finally for the disturbance term one has d̃ i = − D2Ji ω20 Pmi + 2J
ω0 i
i
Ṗm . Thus, one has the
i
following description of the dynamics of the i-th power generator
8.4 Flatness-Based Adaptive Control of Distributed PMSGs 373

Fig. 8.13 A multi-machine (2-area) distributed power generation model

ż1i = z2i
ż2i = z3i (8.105)
ż3i = ai (x) + b1 i g1 u1 + b2 i g2 u2 + d̃ i

For the complete system of the 2 generators one has

ż31 = a1 (x) + b1 1 g1 u1 + b2 1 g2 u2 + d̃ 1
(8.106)
ż32 = a2 (x) + b1 2 g1 u1 + b2 2 g2 u2 + d̃ 2

or in matrix form
ż3 = fa (x) + Mu + d̃ (8.107)

where z3 = [z31 , z32 ]T , u = [u1 , u2 ]T and d̃ = [d̃1 , d̃2 ]T while

   1 
a1 (x) b1 g1 b2 1 g2
fa (x) = 2 , M= (8.108)
a (x) b1 2 g1 b2 2 g2

Setting v = fa (x) + Mu + d̃, one obtains again the linear canonical form for the
i-th generator given by
⎛ i⎞ ⎛ ⎞⎛ i⎞ ⎛ ⎞
ż1 010 z1 0
⎝ż2i ⎠ = ⎝0 0 1⎠ ⎝z2i ⎠ + ⎝0⎠ (vi + d̃ i ) (8.109)
ż3i 000 z3i 1
374 8 Synchronization and Stabilization of Distributed Power Generation Units

In this manner the initial nonlinear power system is transformed into two decoupled
linear subsystems which are in the canonical Brunovksy form. For each one of these
subsystems the appropriate control law is
i
vi = zd(3) − k3 (z̈i − z̈di ) − k2 (żi − żdi ) − k1 (zi − zdi ) − d̃ i (8.110)

The proposed control scheme, considers as output of the synchronous generator

the rotor’s difference angle Δδ. However, it is possible to avoid the use of encoder
readings about the rotor’s turn angle, and to indirectly estimate this parameter through
the processing of measurement coming from PMUs. Such measurements are the bus
voltage magnitude and the associated phase angle, the line current magnitude and
the associated angle and the electrical output power at the terminal bus [111].

8.4.2 Flatness-Based Adaptive Fuzzy Control for MIMO

Nonlinear Systems

8.4.2.1 Transformation of MIMO PMSG System into the Brunovsky

Form

It is assumed now that after defining the flat outputs of the initial MIMO nonlinear
system of the distributed synchronous generators, and after expressing the system
state variables and control inputs as functions of the flat output and of the associated
derivatives, the system can be transformed in the Brunovsky canonical form [375]:

ẋ1 = x2
ẋ2 = x3
···
ẋr1 −1 = xr1 y1 = x1
p
ẋr1 = f1 (x) + j=1 g1j (x)uj + d1 y2 = xr1 −1
(8.111)
ẋr1 +1 = xr1 +2 ···
ẋr1 +2 = xr1 +3 yp = xn−rp +1
···
ẋp−1 = xp
p
ẋp = fp (x) + j=1 gpj (x)uj + dp

where x = [x1 , . . . , xn ]T is the state vector of the transformed system (according to

the differential flatness formulation), u = [u1 , . . . , up ]T is the set of control inputs,
y = [y1 , . . . , yp ]T is the output vector, fi are the drift functions and gi,j , i, j =
1, 2, . . . , p are smooth functions corresponding to the control input gains, while dj
is a variable associated to external disturbances. In holds that r1 + r2 + · · · + rp = n.
Having written the initial nonlinear system into the canonical (Brunovsky) form it
holds
8.4 Flatness-Based Adaptive Control of Distributed PMSGs 375

p
yi(ri ) = fi (x) + j=1 gij (x)uj + dj (8.112)

Equivalently, in vector form, one has the following description for the system
dynamics
y(r) = f (x) + g(x)u + d (8.113)

where the following vectors and matrices are be defined

(r )
y(r) = [y1(r1 ) , . . . , yp p ]
f (x) = [f1 (x), . . . , fp (x)]T
g(x) = [g1 (x), . . . , gp (x)]
(8.114)
with gi (x) = [g1i (x), . . . , gpi (x)]T
A = diag[A1 , . . . , Ap ], B = diag[B1 , . . . , Bp ]
C T = diag[C1 , . . . , Cp ], d = [d1 , . . . , dp ]T

where matrix A has the MIMO canonical form, i.e. with elements
⎛ ⎞
0 1 ··· 0
⎜0 0 · · · 0 ⎟
⎜ ⎟
⎜ ⎟
Ai = ⎜ ... ... · · · ... ⎟
⎜ ⎟
⎝0 0 · · · 1 ⎠ (8.115)
0 0 · · · 0 ri ×ri
 
BiT = 0 0 · · · 0 1 1×ri
 
Ci = 1 0 · · · 0 0 1×ri

Thus, Eq. (8.112) can be written in state-space form

ẋ = Ax + B[f (x) + g(x)u + d̃]

(8.116)
y = CT x

which can be also written in the equivalent form:

ẋ = Ax + Bv + Bd̃
(8.117)
y = CT x

where v = f (x) + g(x)u. The reference setpoints for the system’s outputs y1 , . . . , yp
are denoted as y1m , . . . , ypm , thus for the associated tracking errors it holds

e1 = y1 − y1m
e2 = y2 − y2m
(8.118)
···
ep = yp − ypm
376 8 Synchronization and Stabilization of Distributed Power Generation Units

The error vector of the outputs of the transformed MIMO system is denoted as

E1 = [e1 , . . . , ep ]T
ym = [y1m , . . . , ypm ]T
··· (8.119)
(r)
ym(r) = [y1m (r) T
, . . . , ypm ]

(r)
where yim denotes the r-th order derivative of the i-th reference output of the MIMO
dynamical system. Thus, one can also define the following vectors: (i) a vector
containing the state variables of the system and the associated derivatives, (ii) a
vector containing the reference outputs of the system and the associated derivatives
r −1
x = [x1 , . . . , x1r1 −1 , . . . , xp , . . . , xpp ]T (8.120)

r1 −1 r −1
Ym = [y1m , . . . , y1m , . . . , ypm , . . . , ypm
p
]T (8.121)

while in a similar manner one can define a vector containing the tracking error of the
system’s outputs and the associated derivatives
r −1
e = Ym − x = [e1 , . . . , e1r1 −1 , . . . , ep , . . . , epp ]T (8.122)

It is assumed that matrix g(x) is a nonsingular one, i.e. g −1 (x) exists and is bounded
for all x∈Ux , where Ux ⊂Rn is a compact set. In any case, the problem of singularities
in matrix g(x) can be handled by appropriately modifying the state feedback-based
control input.
The objective of the adaptive fuzzy controller, denoted as u = u(x, e|θ ) is: all the
signals involved in the controller’s design are bounded and it holds that limt→∞ e = 0,
(ii) the H∞ tracking performance criterion is succeeded for a prescribed attenuation
level.
In the presence of non-gaussian disturbances wd , successful tracking of the refer-
ence signal is denoted by the H∞ criterion [352, 375]:
T T
0 eT Qedt ≤ ρ 2 0 wd T wd dt (8.123)

where ρ is the attenuation level and corresponds to the maximum singular value of
the transfer function G(s) of the linearized model associated to Eqs. (8.116) and
(8.117).

8.4.2.2 Control Law

The control signal of the MIMO nonlinear system of the distributed synchronous
generators which has been transformed into the Brunovsky form as described by Eq.
(8.117) contains the unknown nonlinear functions f (x) and g(x). In case that the
8.4 Flatness-Based Adaptive Control of Distributed PMSGs 377

complete state vector x is measurable these unknown functions can be approximated

by
fˆ (x|θf ) = Φf (x)θf
(8.124)
ĝ(x|θg ) = Φg (x)θg

where
Φf (x) = (ξf1 (x), ξf2 (x), . . . ξfn (x))T (8.125)

with ξfi (x), ı = 1, . . . , n being the vector of kernel functions (e.g. normalized fuzzy
Gaussian membership functions), where

ξfi (x) = (φfi,1 (x), φfi,2 (x), . . . , φfi,N (x)) (8.126)

thus giving ⎛ ⎞
φf1,1 (x) φf1,2 (x) · · · φf1,N (x)
⎜ 2,1 2,2 2,N ⎟
⎜φ (x) φf (x) · · · φf (x)⎟
Φf (x) = ⎜ f ⎟ (8.127)
⎝ ··· ··· ··· ··· ⎠
n,1 n,2 n,N
φf (x) φf (x) · · · φf (x)

while the weights vector is defined as

 
θf T = θf1 , θf2 , . . . θfN (8.128)

j = 1, . . . , N is the number of basis functions that is used to approximate the compo-

nents of function f which are denoted as i = 1, . . . , n. Thus, one obtains the relation
of Eq. (8.124), i.e. fˆ (x|θf ) = Φf (x)θf .
In a similar manner, for the approximation of function g one has
 T
Φg (x) = ξg1 (x), ξg2 (x), . . . ξgN (x) (8.129)

with ξgi (x), ı = 1, . . . , N being the vector of kernel functions (e.g. normalized fuzzy
Gaussian membership functions), where
 
ξgi (x) = φgi,1 (x), φgi,2 (x), . . . , φgi,N (x) (8.130)

thus giving ⎛ ⎞
φg1,1 (x) φg1,2 (x) · · · φg1,N (x)
⎜φg2,1 (x) φg2,2 (x) · · · φg2,N (x)⎟
Φg (x) = ⎜
⎝ ···
⎟ (8.131)
··· ··· ··· ⎠
φgn,1 (x) φgn,2 (x) · · · φgn,N (x)

while the weights vector is defined as

 p
θg = θg1 , θg2 , . . . , θg (8.132)
378 8 Synchronization and Stabilization of Distributed Power Generation Units

where the components of matrix θg are defined as

 T
j
θg = θgj 1 , θgj 2 , . . . θgj N (8.133)

j = 1, . . . , p is the number of basis functions that is used to approximate the com-

ponents of function g which are denoted as i = 1, . . . , n. Thus one obtains about
matrix θg ∈RN×p
⎛ 1 2 p ⎞
θg1 θg1 · · · θg1
⎜ θg1 θg2 · · · θgp2 ⎟
θg = ⎜ 2 2 ⎟
⎝· · · · · · · · · · · ·⎠ (8.134)
p
θgN θgN · · · θgN
1 2

It holds that ⎛ ⎞ ⎛ 1 p⎞
g1 g1 g12 · · · g1
⎜ g2 ⎟ ⎜ g 1 g22 · · · g2 ⎟
p
⎜ ⎟
g=⎝ ⎠=⎝ 2 ⎜ ⎟ (8.135)
··· ··· ··· ··· · · ·⎠
p
gn gn1 gn2 · · · gn

Using the above, one finally has the relation of Eq. (8.124), i.e. ĝ(x|θg ) = Φg (x)θg .
If the state variables of the system are available for measurement then a state-feedback
control law can be formulated as

u = ĝ −1 (x|θg )[−fˆ (x|θf ) + ym(r) − K T e + uc ] (8.136)

where fˆ (x|θf ) and ĝ(x|θg ) are neurofuzzy models to approximate f (x) and g(x),
respectively. uc is a supervisory control term, e.g. H∞ control term that is used
to compensate for the effects of modelling inaccuracies and external disturbances.
Moreover, K T is the feedback gain matrix that assures that the characteristic poly-
nomial of matrix A − BK T will be a Hurwitz one.

8.4.2.3 Estimation of the State Vector

The control of the system described by Eq. (8.113) becomes more complicated when
the state vector x is not directly measurable and has to be reconstructed through a
state observer. The following definitions are used
• error of the state vector e = x − xm
• error of the estimated state vector ê = x̂ − xm
• observation error ẽ = e − ê = (x − xm ) − (x̂ − xm )
When an observer is used to reconstruct the state vector, the control law of Eq.
(8.136) is written as

u = ĝ −1 (x̂|θg )[−fˆ (x̂|θf ) + ym(r) − K T ê + uc ] (8.137)

8.4 Flatness-Based Adaptive Control of Distributed PMSGs 379

Applying Eq. (8.137) to the nonlinear system described by Eq. (8.113), results
into

y(r) = f (x) + g(x)ĝ −1 (x̂)[−fˆ (x̂) + ym(r) − K T ê + uc ] + d⇒

y = f (x) + [g(x) − ĝ(x̂) + ĝ(x̂)]ĝ −1 (x̂)[−fˆ (x̂) + ym(r) − K T ê + uc ] + d⇒
(r)

y(r) = [f (x) − fˆ (x̂)] + [g(x) − ĝ(x̂)]u + ym(r) − K T ê + uc + d

(8.138)
It holds e = x − xm ⇒ y(r) = e(r) + ym(r) . Substituting y(r) in the above equation gives

e(r) + ym(r) = ym(r) − K T ê + uc + [f (x) − fˆ (x̂)]+

(8.139)
+ [g(x) − ĝ(x̂)]u + d

and equivalently

ė = Ae − BK T ê + Buc + B{[f (x) − fˆ (x̂)]+

(8.140)
+ [g(x) − ĝ(x̂)]u + d̃}

e1 = C T e (8.141)

where e = [e1 , e2 , . . . , ep ]T with ei = [ei , ėi , ëi , . . . , eiri −1 ]T , i = 1, 2, . . . , p and

equivalently ê = [ê1 , ê2 , . . . , êp ]T with êi = [êi , ėˆ i , ëˆ i , . . . , êiri −1 ]T , i = 1, 2, . . . , p.
Matrices A,B and C have been defined in Eq. (8.115).
A state observer is designed according to Eqs. (8.140) and (8.141) and is given
by [375]:
ê˙ = Aê − BK T ê + Ko [e1 − C T ê] (8.142)

ê1 = C T ê (8.143)

The feedback gain matrix is denoted as K∈Rn×p . The observation gain matrix
is denoted as Ko ∈Rn×p and its elements are selected so as to assure the asymptotic
elimination of the observation error.

8.4.3 Application of Flatness-Based Adaptive Fuzzy Control

to the Distributed Power Generators’ Model

8.4.3.1 Differential Flatness of the Synchronous Power Generator

For the model of the 2-area distributed power generation system it holds that
(3)
x1,1 = f1 (x) + g1 (x)u
(3) (8.144)
x1,2 = f2 (x) + g2 (x)u
380 8 Synchronization and Stabilization of Distributed Power Generation Units

Moreover, it holds that

ẋ1,1 = x2,1
ẋ2,1 = x3,1
ẋ3,1 = f1 (x) + g1 (x)u
(8.145)
ẋ1,2 = x2,2
ẋ2,2 = x2,3
ẋ3,2 = f2 (x) + g2 (x)u

Moreover, from Eq. (8.145) it holds

     
(3)
x1,1 f1 (x) g1 (x)
(3) = + u i.e.
x1,2 f2 (x) g2 (x)
 −1     (8.146)
g (x) x (3) f1 (x)
u= 1 { 1,1(3) − }
g2 (x) x1,2 f2 (x)

Therefore, the considered power generator system is a differentially flat one. Next,
taking into account also the effects of additive disturbances the dynamic model
becomes
(3)
x1,1 = f1 (x, t) + g1 (x, t)u + d1
(3) (8.147)
x1,2 = f2 (x, t) + g2 (x, t)u + d2
       
(3)
x1,1 f1 (x, t) g (x, t) d
(3) = + 1 u+ 1 (8.148)
x1,2 f2 (x, t) g2 (x, t) d2

The following control input is defined

 −1  (3)       
ĝ (x, t) xd1,1 fˆ1 (x, t) K1T uc1
u= 1 { − − e + } (8.149)
ĝ2 (x, t) (3)
xd1,2 fˆ2 (x, t) K2T uc2

where [uc1 uc2 ]T is a robust control term that is used for the compensation of the
model’s uncertainties and of external disturbances and KiT = [k1i , k2i , . . . , kn−1
i
, kni ].
Substituting Eq. (8.149) into Eq. (8.148) the closed-loop tracking error dynamics is
obtained       −1
(3)
x1,1 f1 (x, t) g1 (x, t) ĝ1 (x, t)
(3) = + ·
x1,2 f2 (x, t) g2 (x, t) ĝ2 (x, t)
 (3)          (8.150)
xd1 fˆ1 (x, t) K1T uc1 d1
·{ − ˆ − e+ }+
xd2(3) f2 (x, t) K2T uc2 d2

Equation (8.150) can now be written as

8.4 Flatness-Based Adaptive Control of Distributed PMSGs 381
     
(3)
x1,1 f1 (x, t) g (x, t) − ĝ1 (x, t)
(3) = +{ 1 +
x1,2 f2 (x, t) g2 (x, t) − ĝ2 (x, t)
   −1
ĝ1 (x, t) ĝ1 (x, t)
+ } · (8.151)
ĝ2 (x, t) ĝ (x, t)
 (3)     2    
xd1 fˆ1 (x, t) K1T uc1 d
·{ (3) − ˆ − T e+ }+ 1
xd2 f2 (x, t) K 2 u c2 d2

and using Eq. (8.149) this results into

 (3)     
e1 f1 (x, t) − fˆ1 (x, t) g1 (x, t) − ĝ1 (x, t)
= + u−
e(3)
2 f2 (x, t) − fˆ2 (x, t) g2 (x, t) − ĝ2 (x, t)
 T     (8.152)
K1 uc1 d
− e + + 1
K2T uc2 d2

The following description for the approximation error is defined

   
f1 (x, t) − fˆ1 (x, t) g1 (x, t) − ĝ1 (x, t)
w= + u (8.153)
f2 (x, t) − fˆ2 (x, t) g2 (x, t) − ĝ2 (x, t)

Moreover, the following matrices are defined

⎛ ⎞ ⎛ ⎞
0 10000 00
⎜0 0 1 0 0 0⎟ ⎜0 0 ⎟
⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 0 0⎟ ⎜ ⎟
A=⎜ ⎟ , B = ⎜1 0 ⎟
⎜0 ⎟
0 0 0 1 0⎟ ⎜0 0 ⎟
⎜ ⎜ ⎟ (8.154)
⎝0 00001 ⎠ ⎝0 0 ⎠
0 00000 01
 1 1 1 1 1 1
K K K K K K
KT = 1 2 3 4 5 6
K12 K22 K32 K42 K52 K62

Using matrices A, B, K T , Eq. (8.152) is written in the following form

 
f1 (x, t) − fˆ1 (x, t)
ė = (A − BK )e + Buc + B{
T
+
f2 (x, t) − fˆ2 (x, t)
  (8.155)
g (x, t) − ĝ1 (x, t)
+ 1 u + d̃}
g2 (x, t) − ĝ2 (x, t)

When the estimated state vector x̂ is used in the feedback control loop, equivalently
to Eq. (8.140) one has
382 8 Synchronization and Stabilization of Distributed Power Generation Units
 
f1 (x, t) − fˆ1 (x̂, t)
ė = Ae − BK ê + Buc + B{
T
+
f2 (x, t) − fˆ2 (x̂, t)
  (8.156)
g (x, t) − ĝ1 (x̂, t)
+ 1 u + d̃}
g2 (x, t) − ĝ2 (x̂, t)

and considering that the approximation error w is now denoted as

   
f1 (x, t) − fˆ1 (x̂, t) g1 (x, t) − ĝ1 (x̂, t)
w= + u (8.157)
f2 (x, t) − fˆ2 (x̂, t) g2 (x, t) − ĝ2 (x̂, t)

Equation (8.156) can be also written as

ė = Ae − BK T ê + Buc + Bw + Bd̃ (8.158)

The associated state observer will be described again by Eqs. (8.142) and (8.143).

8.4.3.2 Dynamics of the Observation Error

The observation error is defined as ẽ = e − ê = x − x̂. Subtracting Eq. (8.142) from

Eq. (8.140) as well as Eq. (8.143) from Eq. (8.141) one gets

ė − ê˙ = A(e − ê) + Buc + B{[f (x, t) − fˆ (x̂, t)]+

+ [g(x, t) − ĝ(x̂, t)]u + d̃} − Ko C T (e − ê)
e1 − ê1 = C T (e − ê)

or equivalently

ẽ˙ = Aẽ + Buc + B{[f (x, t) − fˆ (x̂, t)]+

+ [g(x, t) − ĝ(x̂, t)]u + d̃} − Ko C T ẽ
ẽ1 = C T ẽ

which can be written as

ẽ˙ = (A − Ko C T )ẽ + Buc + B{[f (x, t) − fˆ (x̂, t)]+

(8.159)
+[g(x, t) − ĝ(x̂, t)]u + d̃}

ẽ1 = C T ẽ (8.160)

or equivalently, it can be written as

ẽ˙ = (A − Ko C T )ẽ + Buc + Bw + d̃} (8.161)

ẽ1 = C T ẽ (8.162)
8.4 Flatness-Based Adaptive Control of Distributed PMSGs 383

Fig. 8.14 Neurofuzzy approximator used for estimating the unknown dynamics of the distributed
generators

8.4.3.3 Approximation of Functions f (x, t) and g(x, t)

Next, the following approximators of the unknown dynamics of the distributed power
generators are defined
 
fˆ (x̂|θf ) x̂∈R6×1 fˆ1 (x̂|θf ) ∈ R1×1
fˆ (x̂) = ˆ1 (8.163)
f2 (x̂|θf ) x̂∈R6×1 fˆ2 (x̂|θf ) ∈ R1×1

with kernel functions n

j=1 μAj (x̂j )
i
i,j
φf (x̂) = N n (8.164)
i=1 j=1 μA (x̂j )
i
j

where N is the number of fuzzy rules, n is the number of fuzzy sets in the antecedent
part of each rule, x̂ is the estimate of the state vector and μAij (x̂) is the i-th membership
function of the antecedent (IF) part of the i-th fuzzy rule. Similarly, the following
approximators of the unknown system dynamics are defined (Fig. 8.14)
 
ĝ (x̂|θg ) x̂∈R6×1 ĝ1 (x̂|θg ) ∈ R1×2
ĝ(x̂) = 1 (8.165)
ĝ2 (x̂|θg ) x̂∈R6×1 ĝ2 (x̂|θg ) ∈ R1×2

The values of the weights that result in optimal approximation are

θf∗ = arg minθf ∈Mθf [supx̂∈Ux̂ (f (x) − fˆ (x̂|θf ))]

(8.166)
θg∗ = arg minθg ∈Mθg [supx̂∈Ux̂ (g(x) − ĝ(x̂|θg ))]

where the variation ranges for the weights are defined as

384 8 Synchronization and Stabilization of Distributed Power Generation Units

Mθf = {θf ∈Rh : ||θf ||≤mθf }

(8.167)
Mθg = {θg ∈Rh : ||θg ||≤mθg }

The value of the approximation error defined in Eq. (8.153) that corresponds to the
optimal values of the weights vectors θf∗ and θg∗ one has
   
w = f (x, t) − fˆ (x̂|θf∗ ) + g(x, t) − ĝ(x̂|θg∗ ) u (8.168)

which is next written as

 
w = f (x, t) − fˆ (x̂|θf ) + fˆ (x̂|θf ) − fˆ (x̂|θf∗ ) +
  (8.169)
+ g(x, t) − ĝ(x̂|θg ) + ĝ(x̂|θg ) − ĝ(x̂|θg∗ ) u

which can be also written in the following form

 
w = wa + wb (8.170)

where
wa = {[f (x, t) − fˆ (x̂|θf )] + [g(x, t) − ĝ(x̂|θg )]}u (8.171)

wb = {[fˆ (x̂|θf ) − fˆ (x̂|θf∗ )] + [ĝ(x̂, θg ) − ĝ(x̂|θg∗ )]}u (8.172)

Moreover, the following weights error vectors are defined

θ̃f = θf − θf∗
(8.173)
θ̃g = θg − θg∗

8.4.4 Lyapunov Stability Analysis

8.4.4.1 Design of the Lyapunov Function

In case of the distributed synchronous generators, the adaptation law of the neuro-
fuzzy approximators weights θf and θg as well as the equation of the supervisory
control term uc are derived from the requirement for negative definiteness of the
Lyapunov function

V = 21 êT P1 ê + 21 ẽT P2 ẽ + θ̃ θ̃
1 T
2γ1 f f
+ 1
2γ2
tr[θ̃gT θ̃g ] (8.174)

The selection of the Lyapunov function is based on the following principle of indirect
adaptive control ê : limt→∞ x̂(t) = xd (t) and ẽ : limt→∞ x̂(t) = x(t). This yields
limt→∞ x(t) = xd (t). Substituting Eqs. (8.142), (8.143) and (8.159), (8.160) into
Eq. (8.174) and differentiating results into
8.4 Flatness-Based Adaptive Control of Distributed PMSGs 385

V̇ = 21 ê˙ T P1 ê + 21 êT P1 ê˙ + 21 ẽ˙ T P2 ẽ + 21 ẽT P2 ẽ+

˙
(8.175)
+ γ11 θ̃˙fT θ̃f + tr[θ̃˙ g θ̃g ]
T
1
γ2
⇒

V̇ = 21 {(A − BK T )ê + Ko C T ẽ}T P1 ê+

+ 21 êT P1 {(A − BK T )ê + Ko C T ẽ} + 21 {(A − Ko C T )ẽ + Buc +
(8.176)
+ Bd̃ + Bw}T P2 ẽ + 21 ẽT P2 {(A − Ko C T )ẽ + Buc + Bd̃ + Bw}+

+ γ11 θ̃˙fT θ̃f + tr[θ̃˙ g θ̃g ]

T
1
γ2

V̇ = 21 {êT (A − BK T )T + ẽT CKoT }P1 ê + 21 êT P1 {(A − BK T )ê + Ko C T ẽ}+

+ 21 {ẽT (A − Ko C T )T + ucT BT + wT BT +
+ d̃ T BT }P2 ẽ + 21 ẽT P2 {(A − Ko C T )ẽ + Buc + Bw + Bd̃}+

+ γ11 θ̃˙fT θ̃f + tr[θ̃˙ g θ̃g ]

T
1
γ2
⇒
(8.177)
V̇ = 21 êT (A − BK T )T P1 ê + 21 ẽT CKoT P1 ê + 21 êT P1 (A − BK T )ê + 21 êT P1 Ko C T ẽ+
+ 21 ẽT (A − Ko C T )T P2 ẽ + 21 (ucT + wT + d̃ T )BT P2 ẽ+
+ 21 ẽT P2 (A − Ko C T )ẽ + 21 ẽT P2 B(uc + w + d̃)+

+ γ11 θ̃˙fT θ̃f + γ12 tr[θ̃˙ g θ̃g ]

T

(8.178)

Assumption 1: For given positive definite matrices Q1 and Q2 there exist positive
definite matrices P1 and P2 , which are the solution of the following Riccati equations
[375]
(A − BK T )T P1 + P1 (A − BK T ) + Q1 = 0 (8.179)

T
(A − Ko C T ) P2 + P2 (A − Ko C T )−
(8.180)
−P2 B( 2r − ρ12 )BT P2 + Q2 = 0

The conditions given in Eqs. (8.179) to (8.180) are related to the requirement that
the systems described by Eqs. (8.142), (8.143) and (8.159), (8.160) have stable poles.
Substituting Eqs. (8.179) to (8.180) into V̇ yields
V̇ = 21 êT {(A − BK T )T P1 + P1 (A − BK T )}ê + ẽT CKoT P1 ê + 21 ẽT {(A − Ko C T )T P2 +
P (A − K C T )}ẽ + ẽT P B(u + w + d̃) + 1 θ̃˙ T θ̃ + 1 tr[θ̃˙ θ̃ ]
T
2 o 2 c γ1 f f γ2 g g
i.e.
V̇ = − 21 êT Q1 ê + ẽT CKoT P1 ê−
− 21 ẽT {Q2 − P2 B( 2r − ρ12 )BT P2 }ẽ + ẽT P2 B(uc + w + d̃)+ (8.181)
+ 1 θ̃˙ T θ̃ + 1 tr[θ̃˙ θ̃ ]
T
γ1 f f γ2 g g
386 8 Synchronization and Stabilization of Distributed Power Generation Units

The supervisory control uc is decomposed in two terms, ua and ub .

• The control term ua is given by

1
ua = − ẽT P2 B + Δua (8.182)
r

where assuming that the measurable elements of vector ẽ are {ẽ1 , e˜3 , . . . , e˜k }, the
term Δua is such that
⎛ ⎞
p11 ẽ1 + p13 ẽ3 + · · · + p1k ẽk
⎜p13 ẽ1 + p33 ẽ3 + · · · + p3k ẽk ⎟
− 1r ẽT P2 B + Δua = − 1r ⎜
⎝
⎟
⎠ (8.183)
··· ······
p1k ẽ1 + p3k ẽ3 + · · · + pkk ẽk

which finally stands for a product between the state vector elements {ẽ1 , e˜3 , . . . , e˜k }
and the elements of matrix P2 which is obtained from the solution of the previous
Riccati equation.
• The control term ub is given by

ub = −[(P2 B)T (P2 B)]−1 (P2 B)T CKoT P1 ê (8.184)

• ua is an H∞ control used for the compensation of the approximation error w

and the additive disturbance d̃. Its first component − 1r ẽT P2 B has been chosen so
as to compensate for the term 1r ẽT P2 BBT P2 ẽ, which appears in Eq. (8.181). By
including also the second component Δua , one has that ua is computed based on the
feedback only the measurable variables {ẽ1 , e˜3 , . . . , e˜k }, out of the complete vector
ẽ = [ẽ1 , ẽ2 , . . . , ẽn ]. Equation (8.182) is finally rewritten as ua = − 1r ẽT P2 B+Δua .
• ub is a control used for the compensation of the observation error (the control term
ub has been chosen so as to satisfy the condition ẽT P2 Bub = −ẽT CKoT P1 ê.

The control scheme is depicted in Fig. 8.15.

Substituting Eqs. (8.182) and (8.184) in V̇ and assuming that Eqs. (8.179) and
(8.180) hold, one gets

V̇ = − 21 êT Q1 ê + ẽT CKoT P1 ê − 21 ẽT Q2 ẽ+

+ 1r ẽT P2 BBT P2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ+
(8.185)
+ ẽT P2 Bua + ẽT P2 Bub + ẽT P2 B(w + d̃)+

+ γ11 θ̃˙fT θ̃f + tr[θ̃˙ g θ̃g ]

T
1
γ2

or equivalently,
8.4 Flatness-Based Adaptive Control of Distributed PMSGs 387

Fig. 8.15 The proposed output feedback-based adaptive fuzzy control scheme for distributed syn-
chronous generators

V̇ = − 21 êT Q1 ê − 21 ẽT Q2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ+
(8.186)
1 ˙T
θ̃ θ̃ + γ12 tr[θ̃˙ g θ̃g ]
T
+ ẽT P2 B(w + d̃ + Δua ) + γ1 f f

It holds that θ̃˙f = θ̇f − θ˙f∗ = θ˙f and θ̃˙g = θ̇g − θ˙g∗ = θ˙g . The following weights
adaptation laws are considered:

θ̇f = −γ1 Φ(x̂)T BT P2 ẽ

(8.187)
θ̇g = −γ2 Φ(x̂)T BT P2 ẽuT

where assuming N fuzzy rules and associated kernel functions the matrices dimen-
sions are θf ∈RN×1 , θg ∈RN×2 , Φ(x)∈R2×N , B∈R6×3 , P∈R6×6 and ẽ∈R6×1 .
The update of θf is a gradient type algorithm. The update of θg is also a gra-
dient type algorithm, where uc implicitly tunes the adaptation gain γ2 [34, 344].
Substituting Eq. (8.187) in V̇ gives
388 8 Synchronization and Stabilization of Distributed Power Generation Units

V̇ = − 21 êT Q1 ê − 21 ẽT Q2 ẽ −
1 T
2ρ 2
ẽ P2 BBT P2 ẽ + BT P2 ẽ(w + d + Δua )+
(8.188)
+ γ11 (−γ1 )ẽT P2 BΦ(x̂)(θf − θf∗ ) + γ12 (−γ2 )tr[uẽT P2 BΦ(x̂)(θg − θg∗ )]

or

V̇ = − 21 êT Q1 ê − 21 ẽT Q2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ + BT P2 ẽ(w + d̃ + Δua )+
+ γ11 (−γ1 )ẽT P2 BΦ(x̂)(θf − θf∗ ) + γ12 (−γ2 )tr[uẽT P2 B(ĝ(x̂|θg ) − ĝ(x̂|θg∗ )]
(8.189)
Taking into account that u ∈ R3×1 and ẽT PB(ĝ(x|θg ) − ĝ(x|θg∗ )) ∈ R1×3 it holds

V̇ = − 21 êT Q1 ê − 21 ẽT Q2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ + BT P2 ẽ(w + d̃ + Δua )+
+ γ11 (−γ1 )ẽT P2 BΦ(x̂)(θf − θf∗ ) + γ12 (−γ2 )tr[ẽT P2 B(ĝ(x̂|θg ) − ĝ(x̂|θg∗ ))u]
(8.190)
Since ẽT P2 B(ĝ(x̂|θg ) − ĝ(x̂|θg∗ ))u∈R1×1 it holds

tr(ẽT P2 B(ĝ(x|θg ) − ĝ(x|θg∗ )u) =

(8.191)
= ẽT P2 B(ĝ(x|θg ) − ĝ(x|θg∗ ))u

Therefore, one finally obtains

V̇ = − 21 êT Q1 ê − 21 ẽT Q2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ + BT P2 ẽ(w + d̃ + Δua )+
(8.192)
+ γ11 (−γ1 )ẽT P2 BΦ(x̂)(θf − θf∗ ) + γ12 (−γ2 )ẽT P2 B(ĝ(x̂|θg ) − ĝ(x̂|θg∗ ))u

Next, the following approximation error is defined

wα = [fˆ (x̂|θf∗ ) − fˆ (x̂|θf )] + [ĝ(x̂|θg∗ ) − ĝ(x̂|θg )]u (8.193)

Thus, one obtains

V̇ = − 21 êT Q1 ê − 21 ẽT Q2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ+
(8.194)
+BT P2 ẽ(w + d̃) + ẽT P2 Bwα

Denoting the aggregate approximation error and disturbances vector as

w1 = w + d̃ + wα + Δua (8.195)

the derivative of the Lyapunov function becomes

V̇ = − 21 êT Q1 ê − 21 ẽT Q2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ + ẽT P2 Bw1 (8.196)

which in turn is written as

8.4 Flatness-Based Adaptive Control of Distributed PMSGs 389

V̇ = − 21 êT Q1 ê − 21 ẽT Q2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ+
(8.197)
+ 21 ẽT PBw1 + 1 T T
w B P2 ẽ
2 1

Lemma: The following inequality holds

1 T
2
ẽ P2 Bw1 + 21 w1T BT P2 ẽ − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ
(8.198)
≤ 21 ρ 2 w1T w1

Proof : The binomial (ρa − ρ1 b)2 ≥ 0 is considered. Expanding the left part of the
above inequality one gets

ρ 2 a2 + ρ12 b2 − 2ab ≥ 0⇒ 21 ρ 2 a2 + 2ρ1 2 b2 − ab ≥ 0 ⇒

(8.199)
ab − 2ρ1 2 b2 ≤ 21 ρ 2 a2 ⇒ 21 ab + 21 ab − 2ρ1 2 b2 ≤ 21 ρ 2 a2

The following substitutions are carried out: a = w1 and b = ẽT P2 B and the
previous relation becomes
1 T T
w B P2 ẽ
2 1
+ 21 ẽT P2 Bw1 − 1 T
2ρ 2
ẽ P2 BBT P2 ẽ
(8.200)
≤ 21 ρ 2 w1T w1

The above relation is used in V̇ , and the right part of the associated inequality is
enforced
1 1 1
V̇ ≤ − êT Q1 ê − ẽT Q2 ẽ + ρ 2 w1T w1 (8.201)
2 2 2
Thus, Eq. (8.201) can be written as

1 1
V̇ ≤ − E T QE + ρ 2 w1T w1 (8.202)
2 2
where    
ê Q1 0
E= , Q= = diag[Q1 , Q2 ] (8.203)
ẽ 0 Q2

Hence, the H∞ performance criterion is derived. For ρ sufficiently small Eq.

(8.201) will be true and the H∞ tracking criterion will be satisfied. The smallest
value of ρ for which the Riccati equation given in Eq. (8.180) admits as solution
a positive definite matrix P2 is the one that provides maximum robustness to the
control loop. In that case, the integration of V̇ from 0 to T gives
T T T
0 ||E||2 dt + 21 ρ 2 0 ||w1 ||2 dt ⇒
V̇ (t)dt ≤ − 21 0
T T
2V (T ) − 2V (0) ≤ − 0 ||E||2Q dt + ρ 2 0 ||w1 ||2 dt ⇒ (8.204)
T T
2V (T ) + 0 ||E||2Q dt ≤ 2V (0) + ρ 2 0 ||w1 ||2 dt
390 8 Synchronization and Stabilization of Distributed Power Generation Units
∞
It is assumed that there exists a positive constant Mw > 0 such that ||w1 ||2 dt ≤
T 0
Mw . Therefore for the integral 0 ||E||2Q dt one gets
 ∞
||E||2Q dt ≤ 2V (0) + ρ 2 Mw (8.205)
0
∞
Thus, the integral 0 ||E||2Q dt is bounded and according to Barbalat’s Lemma

limt→∞ E(t) = 0 ⇒
limt→∞ ê(t) = 0 (8.206)
limt→∞ ẽ(t) = 0

Therefore limt→∞ e(t) = 0.

8.4.5 Simulation Tests

The performance of the proposed adaptive fuzzy control scheme for distributed syn-
chronous generators was tested through simulation experiments. The state feedback
gain was K∈R6×2 . The basis functions used in the estimation of fi (x̂, t), i = 1, . . . , 3
x̂−cj 2
and gij (x̂, t), i = 1, . . . , 3 j = 1, . . . , 3 were μAj (x̂) = e( σ ) , j = 1, . . . , 2. Since
for each unknown function fi (x̂, t) and gij (x̂, t) there are 6 inputs and each one of them
consists of 3 fuzzy sets, for the approximation of the functions fi (x̂, t) i = 1, . . . , 3,
and gij (x̂, t) i = 1, . . . , 3, j = 1, . . . , 3 there will be 729 fuzzy rules of the form:

Rl : IF x̂1,1 is Al1 AND x̂1,2 is Al2 · · ·

(8.207)
AND x̂3,2 is THEN fˆ l is bl i

The aggregate output of the neuro-fuzzy approximator (rule-base) is

27 
ˆ l 3 μl (x̂i )
fˆi (x̂, t) = l=1 fi
27 4
i=1 Ai
(8.208)
l=1 i=1 μA (x̂i )
l
i

The estimation of the control input gain functions ĝij (x̂, t) i = 1, . . . , 3 was
derived in a similar way. The sampling period was taken to be 0.01 sec. In the begin-
ning of the training of the neuro-fuzzy approximators their weights were initialized
to zero. Moreover, the elements of the system’s state vector were also initialized to
zero. The positive definite matrices P1 ∈R6×6 and P2 ∈R6×6 stem from the solution of
the algebraic Riccati equations Eqs. (8.179) and (8.180), for Q1 and Q2 also positive
definite.
The approximations fˆ and ĝ were used in the derivation of the control law, given
by Eq. (8.137). To show the disturbance rejection capability of the proposed adaptive
8.4 Flatness-Based Adaptive Control of Distributed PMSGs 391

(a) (b)
0.9 0.55

0.8
ω 1 (p.u.)

u1 (p.u.)
0.7 0.5
0.6

0.5
0 2 4 6 8 10 12 14 16 18 20 0.45
t (sec) 0 5 10 15 20
t (sec)
0.6
0.9

0.8

u2 (p.u.)
0.55
ω 2 (p.u.)

0.7
0.5
0.6

0.5 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)

Fig. 8.16 a Tracking of reference set-point 1 by the state variables ωi , i = 1, . . . , 2 of the distrib-
uted power generators’ model b Control inputs ui , i = 1, . . . , 2 applied to the individual power
generators

fuzzy controller, at the beginning of the second half of the simulation time additive
sinusoidal disturbances of amplitude equal to 10 % of the mean value of the control
inputs were applied to the model of the coupled synchronous generators.
The performance of the differential flatness theory-based adaptive fuzzy control
loop was tested in the case of tracking of different reference setpoints. The obtained
results are depicted in Figs. 8.16, 8.17, 8.18, 8.19, 8.20 and 8.21. The state variables
were measured in the per unit (p.u.) system. It can be observed that the proposed
adaptive fuzzy control scheme succeeded fast and accurate tracking of all these
setpoints.
The RMSE (root mean square error) of the examined control loop is also cal-
culated (assuming the same parameters of the controller) in the case of tracking of
the previous setpoints 1 to 5. The results are summarized in Table 8.2. The tracking
accuracy of the control method was remarkable despite the fact that (i) the dynamic
model of the power generators was completely unknown, (ii) only output feedback
was used in the implementation of the control scheme. It can be also seen that the
transient characteristics of the control scheme are quite satisfactory.

8.5 Control and Synchronization of Distributed Inverters

8.5.1 Outline

Next, the problem of decentralized control for stabilization and synchronization of

distributed DC power generation units is examined. The integration of distributed DC
392 8 Synchronization and Stabilization of Distributed Power Generation Units

(a) (b)
0.6
1

0.9

u1 (p.u.)
0.55
ω (p.u.)

0.8
0.5
1

0.7

0.6 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)

1 0.6

0.9 0.55

u2 (p.u.)
ω (p.u.)

0.8 0.5
2

0.7 0.45

0.6 0.4
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)

Fig. 8.17 a Tracking of reference set-point 2 by the state variables ωi , i = 1, . . . , 2 of the distrib-
uted power generators’ model b Control inputs ui , i = 1, . . . , 2 applied to the individual power
generators

(a) (b)
1 0.55
0.9
ω 1 (p.u.)

0.8
u1 (p.u.)

0.7 0.5

0.6
0.5
0 2 4 6 8 10 12 14 16 18 20 0.45
0 5 10 15 20
t (sec) t (sec)
1 0.55
0.9
ω 2 (p.u.)

u2 (p.u.)

0.8
0.5
0.7
0.6
0.5 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)

Fig. 8.18 a Tracking of reference set-point 3 by the state variables ωi , i = 1, . . . , 2 of the distrib-
uted power generators’ model b Control inputs ui , i = 1, . . . , 2 applied to the individual power
generators

power generation units in microgrids, such as photovoltaics and fuel cells requires the
interfacing of such devices with the electricity network through inverters [28, 259,
317, 324, 392, 393]. In this direction, several results on inverters’ control have been
obtained. In [386], control of photovoltaic inverters is analyzed in fault ride-through
situations and voltage sags conditions. In [223], predictive control is proposed for
voltage source inverters. A discrete-time model of the inverters is used to compute
8.5 Control and Synchronization of Distributed Inverters 393

(a) (b)
1 0.55
0.9

u1 (p.u.)
ω 1 (p.u.)

0.8
0.5
0.7
0.6
0.5 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)
1 0.6
0.9
ω 2 (p.u.)

u2 (p.u.)
0.55
0.8
0.7
0.5
0.6
0.5 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)

Fig. 8.19 a Tracking of reference set-point 4 by the state variables ωi , i = 1, . . . , 2 of the distrib-
uted power generators’ model b Control inputs ui , i = 1, . . . , 2 applied to the individual power
generators

(a) (b)
1 0.55
0.9
ω 1 (p.u.)

u1 (p.u.)

0.8
0.5
0.7
0.6
0.5 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)
1 0.6
0.9
0.55
ω 2 (p.u.)

u2 (p.u.)

0.8
0.7
0.5
0.6
0.5 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)

Fig. 8.20 a Tracking of reference set-point 5 by the state variables ωi , i = 1, . . . , 2 of the distrib-
uted power generators’ model b Control inputs ui , i = 1, . . . , 2 applied to the individual power
generators

the control law while Lyapunov stability analysis is also provided. In [164], a con-
trol method is proposed for the amplitude and frequency of the output of voltage
inverters, in the case of distributed (grid-tied) functioning of the inverters, as well
as in the case of stand-alone functioning (islanding). In [413], sliding mode con-
trol is proposed for active and reactive power regulation of a three-phase inverter
394 8 Synchronization and Stabilization of Distributed Power Generation Units

(a) (b)
1 0.55
0.9

u1 (p.u.)
ω 1 (p.u.)

0.8
0.5
0.7
0.6
0.5 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)
1 0.6
0.9

u2 (p.u.)
ω 2 (p.u.)

0.55
0.8
0.7
0.5
0.6
0.5 0.45
0 2 4 6 8 10 12 14 16 18 20 0 5 10 15 20
t (sec) t (sec)

Fig. 8.21 a Tracking of reference set-point 6 by the state variables ωi , i = 1, . . . , 2 of the distrib-
uted power generators’ model b Control inputs ui , i = 1, . . . , 2 applied to the individual power
generators

Table 8.2 RMSE of the power generator’s state variables

Parameter ω1 ω̇2 ω2 ω̇2
RMSE1 0.0035 0.0002 0.0034 0.0002
RMSE2 0.0123 0.0545 0.0118 0.0602
RMSE3 0.0035 0.0020 0.0035 0.0020
RMSE4 0.0031 0.0020 0.0026 0.0020
RMSE5 0.0034 0.0003 0.0033 0.0002
RMSE6 0.0035 0.0003 0.0033 0.0002

connected to an imbalanced grid. In [9], sensorless control is implemented for volt-

age inverters functioning within an islanded-distributed power generation system.
The Kalman Filter is used to compensate for DSP calculation delays. In [245], con-
trol that compensates for load imbalances is developed for a three-phase inverter in
an autonomous functioning mode.
Another problem that arises is the synchronization between multiple parallel
inverters. In the recent years there has been significant effort in solving control and
synchronization problems for voltage inverters, so as to enable connection of a large
number of power sources to the grid [512]. In [205] a state-space model is devel-
oped for the dynamics of distributed power generation units connected to the grid
through inverters. Stability analysis is performed and a simple state feedback control
scheme is proposed. In [258] decentralized frequency control of inverters is pro-
posed. Information about the total active power deficiency of the microgrid is shared
between the local units. Frequency control is performed to achieve a cooperative
8.5 Control and Synchronization of Distributed Inverters 395

frequency recovery. In [105] a Lyapunov theory-based feedback controller is

proposed for voltage inverters connected to the grid. In [167] stability of parallel
DC sources connected to the grid through inverters is investigated and simple feed-
back control is proposed using the equivalence of the inverters’ model to that of
a synchronous generator. In [200] it is considered that each inverter is equivalent
to an electronic circuit with oscillatory dynamics and thus the problem of control
of parallel inverters becomes a problem of synchronization of coupled electrical
oscillators. In [268] Lie algebra is used to transform the nonlinear model of three-
phase photovoltaic units into linear systems and based on the linearized description
a state feedback controller is designed. In [511] the concept of synchronverters is
introduced. This is based on the equivalence between the voltage inverter’s model
and that of a synchronous generator. It is shown that the controller is in principle
a power controller with integrated capability of voltage and frequency regulation.
In [49] Lie algebra-based input-output feedback linearization is used to solve the
synchronization problem for nonlinear inverters connected in parallel to the grid.
It is considered that the power exchange of each local inverter towards the grid is
modeled equivalently by the dynamics of a synchronous generator. In [396] a feed-
back control and synchronization method for distributed DC power generation units
is developed using the equivalence between the inverter’s model and the model of
the synchronous power generator. In [397, 398] stability conditions are provided for
distributed DC power generation units using a Lyapunov function approach.
By solving the problem of synchronization between DC distributed power gen-
eration units which are connected to the microgrid through voltage inverters a more
efficient operation of the power grid can be achieved in terms of power quality and
stability [167]. To this end, the present section is proposing a nonlinear feedback
control method for parallel inverters which is based on differential flatness theory.
Moreover, it is proposing a nonlinear filtering method under the name Derivative-
free nonlinear Kalman Filtering which enables to estimate the non-measurable state
vector elements of each local inverter, as well as disturbance terms affecting each
inverter (the latter can be due to their interaction with other power generation units
in the grid or due to external perturbations and modeling uncertainty) [352, 366,
374, 375].
In the following section, the problem of control and synchronization of parallel
inverters connected to the grid is primarily examined. As far as the power exchange
with the grid is concerned, it is shown that the dynamics of each inverter can be
written in a form that is equivalent to the model of the synchronous power generator.
Using the latter description one can compute the active power that each local inverter
should inject to the network so as to remain synchronized with the rest of the grid. The
frequency of this virtual generator model is next used to compute the output voltage
setpoints for the electrical part of each local inverter. Thus, finally the synchronization
problem of each local inverter is turned into a problem of nonlinear feedback control
for the associated inverter’s electrical model. It is also shown that in the proposed
decentralized control scheme each local controller processes measurements not only
coming from the individual inverter but also coming from other inverters which are
connected to the grid.
396 8 Synchronization and Stabilization of Distributed Power Generation Units

8.5.2 Dynamic Model of the Inverter

The dynamic, model of the three-phase inverter, given in Chap. 1 is reviewed here
⎛ ωLf VLq (iI2 +iI2q ) ⎞
1 pf VLd +qf VLq
⎛ ⎞ ωVLq + 1
i
Cf Id
− Cf VL2 +VL2q
+ ωCf VLq − (VL2 +VL2q )
d

⎜ ⎟
⎜ ⎟
d d
VLd
⎜ VLq ⎟ ⎜−ωVLd + − 1 pf VLq −qf VLd
− ωCf VLd +
ωLf VLd (iI2 +iI2q ) ⎟
⎜ ⎟=⎜ ⎟
1
i d
d
⎝ iId ⎠ ⎜
Cf Iq Cf VL2 +VL2q (VL2 +VL2q ) ⎟+
dt ⎜ d d
⎟
⎜ ωiIq − 1
VLd ⎟
iIq ⎝ Lf ⎠
− ωiId − L1f VLq
⎛ ⎞
0 0
 
⎜ 0 0 ⎟ VId
⎜
+ ⎝ 1 0⎠ ⎟
Lf VIq
0 L1f
(8.209)
while the measurement equation of the inverter’s model is
⎛ ⎞
      VLd
y1 VLd 1000 ⎜ ⎜ VLq ⎟
⎟
= = (8.210)
y2 VLq 0 1 0 0 ⎝ i Ld ⎠
i Lq

and by using the state variables notation x1 = VLd , x2 = VLq , x3 = iLd and x4 = iLq
one has
⎛ p x +q x ωL x (x 2 +x 2 )
⎞
⎛ ⎞ ωx2 + C1f x3 − C1f f x12 +x2f 2 + ωCf x2 − f(x22 +x3 2 ) 4
x1 ⎜ 1 2 1 2 ⎟
⎜ −qf x1 ωLf x1 (x32 +x42 ) ⎟
⎜x2 ⎟ ⎜−ωx1 + C1 x4 − C1 pf xx22 +x − ωC x + ⎟
⎜ ⎟=⎜ 2 f 1 (x12 +x22 ) ⎟
⎟+
d f f
⎝x3 ⎠ ⎜
1 2
dt
⎜ ωx − 1
x ⎟
x4 ⎝ 4 Lf 1 ⎠
− ωx3 − Lf x2 1 (8.211)
⎛ ⎞
0 0
 
⎜ 0 0 ⎟ u1
+⎜⎝ Lf 0 ⎠ u2
1 ⎟

0 L1f

while the measurement equation of the inverter’s model is

⎛ ⎞
      x1
y1 VLd 1000 ⎜⎜x2 ⎟
⎟
= = ⎝ (8.212)
y2 VLq 0100 x3 ⎠
x4
8.5 Control and Synchronization of Distributed Inverters 397

thus, the inverter’s model is written in the nonlinear state-space form

ẋ = f (x) + G(x)u
(8.213)
y = h(x)

where f (x)∈R4×1 , G(x)∈R4×1 and h(x)∈R2×4 .

8.5.3 The Synchronization Problem for Parallel Inverters

8.5.3.1 Equivalence Between Inverters and Synchronous Generators

By applying appropriate control inputs, inverters can be made to function in a man-

ner equivalent to synchronous generators. Actually, with the regulation of the active
power produced by the inverters one can succeed the synchronization between the
voltage output of the inverters and the voltage of the grid [513–515]. This synchro-
nization enables better integration of DC renewable energy units to the grid.
The functioning of the i-th inverter is shown to be equivalent to a synchronous
generator with turn speed denoted as ωi [49, 396–398, 511, 512]. In this modelling
approach one has
Δδi = ωi − ωd = −kpi (Pim − Pid ) (8.214)

where Pim is the measured active power of the i-th power generation unit and Pid
is the desirable active power. Coefficient kpi is a “droop” gain which is practically
computed by dividing the range of variation of the inverter’s frequency (ωmax –ωmin )
by the maximum active power Pimax that the inverter can produce. Equation (8.214)
denotes that when representing the inverter as an equivalent synchronous generator
the turn speed should be proportional to the active power that the inverter provides.
Next, a relation is obtained for the measured active power Pim and the real active
power Pi of the inverter Pi . It holds that Pim (s) = e−sτpi P(s) (delay relation in fre-
quency domain) which after intermediate operations gives the differential equation

τpi Ṗim = −Pim + Pi (8.215)

Thus the inverter’s dynamics is expressed as

Δδ̇i = Δωi
(8.216)
τpi Ṗim = −Pim + Pi

By deriving Eq. (8.214) with respect to time one obtains

Δω̇i = −kpi Ṗim + kpi Ṗid (8.217)

398 8 Synchronization and Stabilization of Distributed Power Generation Units

Moreover, by solving Eq. (8.215) with respect to Ṗim one has

Ṗim = − τ1p Pim + 1

τ pi
Pi (8.218)
i

By substituting Eq. (8.218) into (8.217) the following relation is obtained

Δω̇i = −kpi (− τ1p Pim + 1

τ pi
Pi ) + kpi Ṗid (8.219)
i

and by considering that Ṗid = 0 one gets

kpi kpi
Δω̇i = τ pi
Pim − τ pi
Pi (8.220)

or equivalently
Ji Δω̇i = Pim − Pi (8.221)

where the virtual moment of inertia is defined as Ji = τpi /kpi . Additionally, from Eq.
(8.214) one has

ωi − ωd − kpi Pid = −kpi Pim ⇒Pim = − k1p ωi + 1

kpi
ωd + Pid ⇒
i
(8.222)
Pim = − k1p Δωi + Pid
i

By substituting Eq. (8.222) into Eq. (8.221) the following relation is obtained

Ji Δω̇i = − k1p Δωi + Pid − Pi (8.223)

i

The damping coefficient Dpi = k1p is defined. Using this coefficient in Eq. (8.222)
i
gives
Ji Δω̇i = −Dpi Δωi + Pid − Pi (8.224)

which is the equation that describes the rotational motion of a synchronous genera-
tor. In ideal conditions there is no interaction (power exchange) between distributed
power units connected to the same electricity grid. However, frequently such inter-
action exists and in the latter case Eq. (8.224) should be enhanced by including an
interaction term
n
Ji Δω̇i = −Dpi Δωi + (Pid − Pi ) + j=1,j=i G ij sin(δi − δj ) (8.225)

where δi is the virtual turn angle that is associated with the i-th power generation unit
(inverter). About the coupling coefficients G ij these are functions of the conductance
of the grid line which connects the i-th to the j-th power generation unit, as well as
of the grid voltage that is measured at points i and j respectively.
Thus, finally the dynamics of the i-th power generation unit is described as
8.5 Control and Synchronization of Distributed Inverters 399

Δδ̇i (t) = Δωi (t)

(8.226)
Ji Δω̇i (t) = −Dpi Δωi (t) + (Pid (t) − Pi (t)) + Nj=1 G ij sin(δi − δj )

In the design of the control and synchronization system for the distributed power
generation units, it is considered that the parameters Ji , Dpi and G ij are either known
or can be computed from measurements. Moreover, it is considered that the i-th
local controller not only processes measurements coming from the associated power
generation unit, but also uses measurements coming from the other power units which
are connected to the grid (that is the virtual turn angles δj ).

8.5.3.2 Control for Parallel Inverters Connected to the Grid

In Eq. (8.226), it has be shown that under certain conditions, the dynamics of the
inverter becomes equivalent to that of the synchronous power generator. Then one
has that the dynamics of the inverter is composed of two parts (i) the rotation part
and (ii) the electrical part.
The virtual rotation part is given by:

Δδ̇i (t) = Δωi (t)

(8.227)
Ji Δω̇i (t) = −Dpi Δωi (t) + (Pid (t) − Pi (t)) + Nj=1,j=i G ij sin(δi − δj )

Ji is the virtual inertia, Dpi is the virtual damping coefficient, Pid is the desirable
active power of the i-th unit and Pi and the measured active power, G ij is a coupling
coefficient between the i-th and the j-th power generation unit and sin(δi − δj ) is
a synchronization index between the i-th and the j-th power generation unit [49,
396–398, 511, 512].
The electrical part has been defined in Chap. 6 and is given by:
⎛ ωLf VLq (iI2 +iI2q ) ⎞
⎛ ⎞ 1 pf VLd +qf VLq
ωVLq + 1
i
Cf Id
− Cf VL2 +VL2q
+ ωCf VLq − d
(VL2 +VL2q )
VLd ⎜ ⎟
⎜ VLq ⎟ ⎜ ωLf VLd (iI2 +iI2q ) ⎟
d d
1 pf VLq −qf VLd
⎜ ⎟=⎜ −ωVLd + − − ωCf VLd + ⎟
1
i d
⎟+
⎝ iId ⎠ ⎜
d
Cf Iq Cf VL2 +VL2q (VL2 +VL2q )
dt ⎜ d d
⎟
⎝ ωiIq − L1f
VLd ⎠
iIq
− ωiId − L1f VLq
⎛ ⎞
0 0
 
⎜ 0 0 ⎟ VId
⎜
+ ⎝ 1 0⎠ ⎟
Lf VIq
0 L1f
(8.228)
The synchronizing control approach for the i-th inverter makes use of Eq. (8.227)
and of the linearized inverter model given in Eq. (8.228). First, the value of Pi , that
is the active power that the i-th inverter should inject to the grid, is found from
the solution of the control problem of Eq. (8.227). Subsequently Pi is used in the
computation of the solution of the control problem of Eq. (8.228). This is depicted
in Fig. 8.22.
400 8 Synchronization and Stabilization of Distributed Power Generation Units

Fig. 8.22 Control loops for the virtual synchronous generator model and for the electrical part of
the inverter

The synchronizing control method can be generalized to the case of multiple

inverters connected to the grid (see Fig. 8.23). It can be shown again that the model
of N distributed power units which are connected to the grid through inverters is
also differentially flat. The latter is proven following the procedure described in
Sect. 6.2.3 and by defining as flat output a generalization of the vector described
in Eq. (6.24), which now becomes Y = [y11 , y21 , y31 , y12 , y22 , y32 . . . , y1N , y2N , y3N ] or
Y = [δ 1 , VL1d , VL1q , δ 2 , VL2d , VL2q , . . . , δ N , VLNd , VLNq ]. All state variables and control
inputs for the model of the N coupled inverters can be expressed as functions of the
aforementioned flat output Y and of its derivatives.
Using the previous flat output definition, and the state variables z1i = y1 , z2i = ẏ1 ,
z3 = y2 , z4i = ẏ2 , z5i = y3 , z6i = ẏ3 , one has:
i

⎛ i⎞ ⎛ ⎞⎛ i⎞ ⎛ ⎞
ż1 0100000 z1 000
⎜żi ⎟ ⎜0 0 0 0 0 0 0⎟ ⎜zi ⎟ ⎜1 0 0⎟ ⎛ i ⎞
⎜ 2i ⎟ ⎜ ⎟ ⎜ 2⎟ ⎜ ⎟
⎜ż ⎟ ⎜0 0 0 1 0 0 0⎟ ⎜zi ⎟ ⎜0 0 0⎟ v1i
⎜ i⎟ = ⎜
3 ⎟ ⎜ 3i ⎟ + ⎜ ⎟⎝ ⎠
⎜ż ⎟ ⎜0 0 0 0 0 0 0⎟ ⎜z ⎟ ⎜0 1 0⎟ v2i (8.229)
⎜ 4i ⎟ ⎜ ⎟ ⎜ 4i ⎟ ⎜ ⎟ v
⎝ż5 ⎠ ⎝0 0 0 0 0 1 0⎠ ⎝z5 ⎠ ⎝0 0 0⎠ 3
ż6i 0000000 z6i 001

where the control inputs of this model are defined as

N
v1i = 1
Ji
[−Dpi Δωi (t) + (Pid (t) − Pi (t)) + j=1 G ij sin(δi − δj )] (8.230)

and using Eq. (6.53) it holds for the i-th inverter’s model
8.5 Control and Synchronization of Distributed Inverters 401

Fig. 8.23 Distributed DC power generation units connected through inverters to the grid

v2i = Lf2 h1 i (x) + Lga Lf h1i (x)u1i + Lgb Lf h1i (x)u2i

(8.231)
v3i = Lf2 h2 i (x) + Lga Lf h2i (x)u1i + Lgb Lf h2i (x)u2i

If the coupling coefficients G ij , i = 1, . . . , n, j = 1, . . . , n are known then the

solution of the control problem for Eq. (8.227) can be obtained, by communicating
to each power generation unit i the rotation angles of the rest N − 1 power generation
units so as to compute the sum Nj=1 G ij sin(δi − δj ). In such a case, and by denoting
x1,i = δi , x2,i = ωi the control input (power) applied to the i-th power generation
unit is
Pi = −Ji ẍ1,i
d
− Dp1 x2 + Pid + Nj=1 G ij sin(x1,i − x1,j )+
(8.232)
+ Ji Kdi (ẋ1,i − ẋ1,i
d
) + Ji Kpi (x1,i − x1,i
d
)

8.5.4 State and Disturbances Estimation of Parallel Inverters

with Nonlinear Kalman Filtering

A state estimator for each local power generation unit can be also designed in the form
of a disturbance observer. It is considered that the linearized model of the i-th inverter
is affected by additive input disturbances (which are considered to be modelling
uncertainties and external perturbations such as load changes, power injected due to
402 8 Synchronization and Stabilization of Distributed Power Generation Units

the connection of other power generation units to the grid, faults in the grid etc.).
With reference to the state-space description given in Eq. (8.229), the disturbance
terms can describe both modelling uncertainties and external perturbations.

z̈1i = v1i + d̃1i

z̈3i = v2i + d̃2i (8.233)
z̈5i = v3i + d̃3i

The disturbances’ dynamics can be represented by the n-th order derivative of the
disturbances variables together with the associated initial conditions (however, since
the disturbances are going to be estimated by the Kalman Filter, the knowledge of
the initial conditions finally becomes obsolete). Thus the additive disturbances are
equivalently described in the form d̃1(n) = fd1 , d̃2(n) = fd2 and d̃3(n) = fd3 .
Without loss of generality it is assumed that in the relation d̃i(n) = fdi , the deriv-
ative’s order is n = 2. The state vector is extended by including as additional state
variables the disturbances and their derivatives. Thus, one has z1i = x1 , z2i = ẋ1 ,
zi = x , zi = ẋ , zi = x , zi = ẋ , zi = d̃ , zi = d̃˙ , zi = d̃ , zi = d̃˙ , zi = d̃ and
3 2 4 2 5 3 6 3 7 1 8 1 9 2 10 2 11 3
i
z12 = d̃˙ 3 .
It holds that ż1i = z2i , ż2i = z7 + v1i , ż3i = z4i , ż4i = z9 + v2i , ż5i = z6i , ż6i = z11
i
+ v3i ,
ż7 = z8 , ż8 = fd1 , ż9 = z10 , ż10 = fd2 , ż11 = z12 , ż12 = fd3 . Therefore, one has the
i i i i i i i i i

system’s dynamics in the extended spate-space form

że = Ae ẑe + Be ve (8.234)

where the extended control inputs vector is ve = [v1i , v2i , v3i , fd1 , fd2 , fd3 ]T and
⎛ ⎞ ⎛ ⎞ ⎛ ⎞
010000000000 000000 100
⎜0 0 0 0 0 0 1 0 0 0 0 0 ⎟ ⎜1 0 0 0 0 0 ⎟ ⎜0 0 0 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 1 0 0 0 0 0 0 0 0 ⎟ ⎜0 0 0 0 0 0⎟ ⎜0 1 0 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 0 0 0 0 1 0 0 0 ⎟ ⎜0 1 0 0 0 0 ⎟ ⎜0 0 0 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 0 1 0 0 0 0 0 0 ⎟ ⎜0 0 0 0 0 0⎟ ⎜0 0 1⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 0 0 0 0 0 0 1 0 ⎟ ⎜ 0 0 1 0 0 0 ⎟ T ⎜0 0 0 ⎟
⎜
Ae = ⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎟ Be = ⎜0 0 0 0 0 0⎟ Ce = ⎜0 0 0⎟ (8.235)
⎜0 0 0 0 0 0 0 1 0 0 0 0 ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 0 0 0 0 0 0 0 0⎟ ⎜0 0 0 1 0 0 ⎟ ⎜0 0 0 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 0 0 0 0 0 1 0 0 ⎟ ⎜0 0 0 0 0 0⎟ ⎜0 0 0 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎜0 0 0 0 0 0 0 0 0 0 0 0⎟ ⎜0 0 0 0 1 0 ⎟ ⎜0 0 0 ⎟
⎜ ⎟ ⎜ ⎟ ⎜ ⎟
⎝0 0 0 0 0 0 0 0 0 0 0 1⎠ ⎝0 0 0 0 0 0⎠ ⎝0 0 0 ⎠
000000000000 000001 000

For the extended state-space description of the system the state observer becomes

ẑ˙e = Ao ẑe + Bo ve + Kf (Co ze − Co )ẑe (8.236)

8.5 Control and Synchronization of Distributed Inverters 403

where for matrices Ae and Ce holds Ao = Ae and Co = Ce while for matrix Bo differs
from Be in the elements of the 10th and 12th row, where all elements are set to zero.
For the extended state-space description of the system it is possible to perform
state estimation using the Derivative-free nonlinear Kalman Filter. This filter con-
sists of the standard Kalman Filter recursion on the linearized equivalent model
of the i-th inverter, and of an inverse transformation based on differential flatness
theory (as explained in Sect. 6.2.3) which enables to obtain estimates for the state
variables of the initial nonlinear model of the inverter (if the latter is necessary). In
the filter’s algorithm, the previously defined matrices Ae ,Be and Ce are substituted
by their discrete-time equivalents Aed , Bed and Ced . This is done through common
discretization methods. The recursion of the Kalman Filter in this case is [34, 341,
344, 367]:
measurement update:

Kf (k) = P− (k)CdT [Ced P− (k)CeTd + R(k)]−1

x̂(k) = x̂ − (k) + Kf (k)[Ced z(k) − Ced ẑ(k)] (8.237)
P(k) = P− (k) − K(k)Ced P− (k)

time update:
P− (k + 1) = Aed P(k)ATed + Q(k)
(8.238)
x̂ − (k + 1) = Aed x̂(k) + Bed v(k)

After identifying the disturbance terms, the control input of the inverter is modified
as follows:
v1i = z̈1 − kd1 (ż1 − ż1d ) − kp1 (z1 − z1d ) − ẑ7
v2i = z̈3 − kd2 (ż3 − ż3d ) − kp2 (z3 − z3d ) − ẑ9 (8.239)
v3i = z̈5 − kd3 (ż5 − ż5d ) − kp3 (z5 − z5d ) − ẑ11

The inclusion of the disturbance estimation terms ẑ7 , ẑ9 and ẑ11 in the feedback
control inputs enables to compensate for effects of the perturbations d̃1 , d̃2 and d̃3 .

8.5.5 Simulation Tests

The performance of the proposed distributed control scheme for the synchronization
of parallel inverters was tested through simulation experiments. A model of N = 3
distributed DC power generation units was considered, while each one of these units
was connected to the grid through an inverter (see Fig. 8.24). The power exchange
between each inverter and the grid was considered to be described by the model of
a synchronous generator.
The three interconnected inverters, shown in Fig. 8.24, are assumed to have differ-
ent model parameters which are described in Table 8.3. Moreover, the three invert-
ers are considered to be subjected to different perturbation inputs. Synchronization
404 8 Synchronization and Stabilization of Distributed Power Generation Units

Fig. 8.24 Distributed DC power generation units connected through inverters

Table 8.3 Parameters of the inverters

Inv1 Inv2 Inv3
Lf (mH) 10.5 10.3 10.1
Cf (mF) 0.04 0.03 0.02
pf (Kw) 21.1 22.3 23.6

means that given the desirable reference rotation speed ωi , one can compute the
amount of active power Pi contributed to the grid by each local inverter which in
turn results into ωi . Next, knowing Pi one can compute for the inner control loop of
Eq. (8.228) the associated voltage reference setpoints VLd and VLq and can solve the
control problem for the electrical part of the inverter’s dynamics. This means that
the synchronization problem of each local inverter is finally turned into a problem
of nonlinear feedback control for the associated electrical model of the inverter.
The main relation that the section considers for synchronization of the invert-
ers with a reference frequency of the grid is Eq. (8.227). According to this relation
the objective is that all inverters finally attain the same frequency ωi . By know-
ing the active power Pi which results into virtual rotation speed ωi , one can com-
pute also the setpoints for the output voltages VLd and VLq (see Fig. 8.22) which
have to be used in the control problem of the inverter’s electrical part described by
Eq. (8.228). One can compute these setpoints through Eqs. (1.98) and (1.99). In
8.5 Control and Synchronization of Distributed Inverters 405

(a) (b)
1.5 1

0.8

sync error DG1 −DG2 (p.u.)

0.6

0.4
1
Gen 1ω (p.u.)

0.2

−0.2
0.5
−0.4

−0.6

−0.8

0 −1
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.25 a Angular speed of power generation unit 1 (red line setpoint, blue line real value), b
synchronization error between power generation units 1 and 2

(a) (b)

d/dt d1 (p.u.)
1 1 1
1
d1 (p.u.)
d/dt V L (p.u.)

0.8 0.5 0.5 0.5

V L (p.u.)

0.6
q

0 0 0
0 10 20 30 40 0 10 20 30 40
d

0.4
−0.5 time time
0.2
d/dt d2 (p.u.)

1 1
d2 (p.u.)

0 −1
0 2 4 0 2 4
0.5 0.5
time time
1 1 0 0
0 10 20 30 40 0 10 20 30 40
d/dt V L (p.u.)

0.5 0.5 time time

V L (p.u.)

d/dt d3 (p.u.)

1 1
q

d3 (p.u.)

0 0
d

0.5 0.5
−0.5 −0.5

−1 −1 0 0
0 2 4 0 2 4 0 10 20 30 40 0 10 20 30 40
time time time time

Fig. 8.26 Inverter of power generation unit 1 a Voltage vector components VLd and VLq and their
derivatives, (red line setpoint, blue line real value, green line estimated value) b estimation (green
line) of disturbance inputs (blue line) and of their derivatives with the use of the Derivative-free
nonlinear Kalman Filter

Eq. (1.98) the active power should be given the value that is computed from the
solution of the control problem for the virtual synchronous generator model of Eq.
(8.227). Additionally, in Eq. (1.99) a reference value for the reactive power can be
used. The currents iL,d and iL,q are considered to be measurable. Thus one finally
has a set of two equations with unknowns the output voltages VLd and VLq . Solving
this system with respect to VLd and VLq provides reference setpoints for the output
voltages which finally result in the synchronization of the inverters with the grid.
406 8 Synchronization and Stabilization of Distributed Power Generation Units

(a) (b)
300 1
pf

0.9 qf

200
0.8

0.7
100

pf − qf (p.u.)
0.6
Vabc

0 0.5

0.4
−100
0.3

0.2
−200
0.1

−300 0
1 1.2 1.4 1.6 1.8 2 5 10 15 20 25 30 35 40
time (sec) t

Fig. 8.27 Inverter of power generation unit 1 a Three-phase voltage variables VL , (red line Va , blue
line Vb , green line Vc ) b active and reactive power of the inverter

(a) (b)
1.5 1

0.8
sync error DG2 −DG3 (p.u.)

0.6

0.4
1
Gen 2 ω (p.u.)

0.2

−0.2
0.5
−0.4

−0.6

−0.8

0 −1
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.28 a Angular speed of power generation unit 2 (red line setpoint, blue line real value), b
synchronization error between power generation units 2 and 3

Moreover, to estimate the nonmeasurable state variables in the inverter models as

well as to estimate and compensate for external perturbations affecting the inverter’s
dynamics the Derivative-free nonlinear Kalman Filter has been used.
The obtained results are depicted in Figs. 8.25, 8.26, 8.27, 8.28, 8.29, 8.30, 8.31,
8.32 and 8.33. The state variables of the distributed power generators are actually
measured in SI units. However, in the simulation experiments presented in Figs. 8.25,
8.26, 8.27, 8.28, 8.29, 8.30, 8.31, 8.32 and 8.33, these variables have been measured in
8.5 Control and Synchronization of Distributed Inverters 407

(a) (b)

d/dt d1 (p.u.)
1 1
1

d1 (p.u.)
1

d/dt V L (p.u.)
0.8 0.5 0.5
0.5
V L (p.u.)

0.6 0 0

d
0 0 10 20 30 40 0 10 20 30 40
q

0.4
−0.5
time time
0.2

d/dt d2 (p.u.)
1 1

d2 (p.u.)
0 −1
0 2 4 0 2 4
0.5 0.5
time time
1 1 0 0
d/dt V L (p.u.)

0 10 20 30 40 0 10 20 30 40
0.5 0.5
V L (p.u.)

time time

d/dt d3 (p.u.)
q

0 0 1 1

d3 (p.u.)
q

−0.5 −0.5 0.5 0.5

−1 −1 0 0
0 2 4 0 2 4 0 10 20 30 40 0 10 20 30 40
time time time time

Fig. 8.29 Inverter of power generation unit 2 a Voltage vector components VLd and VLq and their
derivatives, (red line setpoint, blue line real value, green line estimated value) b estimation (green
line) of disturbance inputs (blue line) and of their derivatives with the use of the Derivative-free
nonlinear Kalman Filter

(a) (b)
300 1
pf
0.9 qf

200
0.8

0.7
100
pf − qf (p.u.)

0.6
V abc

0 0.5

0.4
−100
0.3

0.2
−200
0.1

−300 0
1 1.2 1.4 1.6 1.8 2 5 10 15 20 25 30 35 40
time (sec) t

Fig. 8.30 Inverter of power generation unit 2 a Three-phase voltage variables VL , (red line Va , blue
line Vb , green line Vc ) b active and reactive power of the inverter

the per unit (p.u.) system. It can be noticed that the proposed control and state estima-
tion scheme achieved both satisfactory transients and good tracking performance of
the reference setpoints. Moreover, fast synchronization between the distributed power
generation units was achieved. It is noted that the proposed control and synchroniza-
tion approach is scalable and can be applied to a larger number of N interconnected
408 8 Synchronization and Stabilization of Distributed Power Generation Units

(a) (b)
1.5 1

0.8

sync error DG3 −DG1 (p.u.)

0.6

0.4
1
Gen 3 ω (p.u.)

0.2

−0.2
0.5
−0.4

−0.6

−0.8

0 −1
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
time time

Fig. 8.31 a Angular speed of power generation unit 3 (red line setpoint, blue line real value), b
synchronization error between power generation units 3 and 1

(a) (b)

d/dt d1 (p.u.)
1 1 1 1
d1 (p.u.)
d/dt V L (p.u.)

0.8 0.5 0.5 0.5

V L (p.u.)

0.6
d

0 0 0
0.4 0 10 20 30 40 0 10 20 30 40
d

−0.5 time time

0.2
d/dt d2 (p.u.)

1 1
d2 (p.u.)

0 −1
0 2 4 0 2 4
0.5 0
time time
1 1 0 −1
0 10 20 30 40 0 10 20 30 40
d/dt V L (p.u.)

0.5 0.5 time time

V L (p.u.)

d/dt d3 (p.u.)

1 1
q

0 0
d3 (p.u.)
q

0.5 0.5
−0.5 −0.5

−1 −1 0 0
0 2 4 0 2 4 0 10 20 30 40 0 10 20 30 40
time time time time

Fig. 8.32 Inverter of power generation unit 3 a Voltage vector components VLd and VLq and their
derivatives, (red line setpoint, blue line real value, green line estimated value) b estimation (green
line) of disturbance inputs (blue line) and of their derivatives with the use of the Derivative-free
nonlinear Kalman Filter

DC power generation units. The simulation diagrams show convergence of the virtual
turn speed ωi of each inverter, to the associated reference value ωi∗ .
Additionally, it is noted that the section’s results confirm the usefulness of the
synchronverters concept for control and stabilization of distributed power generation
units, e.g. in the case of renewable energy systems [511]. The methodology developed
in this section enables a power corporation to control a synchronverter in the same
8.5 Control and Synchronization of Distributed Inverters 409

(a) (b)
300 1
pf
0.9 qf

200
0.8

0.7
100

pf − qf (p.u.)
0.6
Vabc

0 0.5

0.4
−100
0.3

0.2
−200
0.1

−300 0
1 1.2 1.4 1.6 1.8 2 5 10 15 20 25 30 35 40
time (sec) t

Fig. 8.33 Inverter of power generation unit 3 a Three-phase voltage variables VL , (red line Va , blue
line Vb , green line Vc ) b active and reactive power of the inverter

Table 8.4 RMSE for the distributed inverters

RMSE1 RMSE2 RMSE3
ω 0.0225 0.0427 0.0199
VLd 0.0180 0.0008 0.0003
VLq 0.0246 0.0020 0.0010

way as controlling synchronous generators, which considerably facilitates the grid

connection of renewable energy and smart grid integration.
Finally, in Table 8.4, the RMSE of setpoint tracking by the state variables ω, VLd ,
VLq of the distributed inverters j = 1, 2, 3 is presented. It can be noticed that the
proposed control and state estimation scheme assures good tracking performance.
The presented simulation experiments demonstrated the efficiency of the control
method in tracking rapidly changing reference setpoints while also succeeding good
transients. The disturbances appearing in the simulation experiments could be met in
adverse operating conditions of the distributed power generation system. Even for the
latter case the good performance of the control loop is confirmed. Such disturbances
can be due to modelling errors (e.g. parametric changes in the inverters’ model) or
due to external perturbations (e.g. grid faults or disturbances due to the connection
or disconnection from the grid of power generation units).