STATE FEEDBACK CONTROLLER DESIGN

APPLIED TO QUADRATIC BOOST CONVERTER

USED IN PHOTOVOLTAIC ARRAY MPPT
Gabriel Morais, Ernane A. A. Coelho Ernane A. A. Coelho
NUPEP NUPEP
UFU UFU
Uberlândia, Brazil Uberlândia, Brazil
[email protected], [email protected] [email protected], [email protected]

Eric Nery Chaves Ghunter Paulo Viajante

NuPSE NuPSE
IFG IFG
Itumbiara, Brazil Itumbiara, Brazil
[email protected], [email protected] [email protected], [email protected]

Rafael Nielson
Enel Goiás
Goiânia, Brazil

Abstract — This paper presents the controller design that will VOUT 1
be used in the internal loop of the maximum power point tracking  (1)
(MPPT) algorithm to control the output voltage of a PV VIN 1 D
(photovoltaic) array. The PV system is intended to be connected to
the single-phase power grid through the DC-DC Quadratic Boost D is the duty cycle of the converter switch, 0 < D < 1.
converter at the interface between the PV panels and the DC bus. However, the Boost converter has performance limitations when
The proposed controller uses the state feedback technique and, in scanned in regions close to D = 1 [2]. In this context, the
its design, it will be demonstrated the mathematical modeling of Quadratic Boost converter becomes an option to be able to raise
the DC-DC Quadratic Boost converter, the step-by-step voltages at higher values without considerable loss of efficiency,
determination of the discrete gains that feed the plant, and since the conversion factor of the input voltage, VIN, to the
computational simulation results that show the viability of the output voltage, VOUT, is greater than that of the Boost converter
control technique in question and corroborate its use. [3]. The static gain of the DC-DC Boost Quadratic converter is
shown in (2).
Keywords—Maximum Power Point Tracking, Photovoltaic
Generation, Quadratic Boost Converter, State Feedback Control. VOUT 1
 (2)
I. INTRODUCTION VIN (1  D )2
The efficiency in the conversion of electrical energy in
Typically, this converter is used to obtain a stabilized output
distributed generation systems composed by photovoltaic panels
voltage, even with variations of the input voltage or load [4].
is limited to approximately 14 to 16% in the best conditions of
However, for maximum power extraction applications in
irradiance and temperature [1]. This efficiency depends on the
photovoltaic arrays, the input voltage is the variable to be
point of the IxV curve, characteristic of PV (photovoltaic)
controlled, and the output voltage is considered a constant value,
panels, in which the system is operating. This point of operation
since the output of the converter will be connected in parallel to
depends on the voltage and current at the output terminals of the
a fixed DC bus. Thus, the circuit is analyzed as shown in Figure
PV array, and since the load fed by the photo-generated energy
1.
has an eminently changeable profile, it is necessary to interface
this load with the PV array through a DC-DC converter. It can be noted that the input of the converter is a resistor of
variable resistance, RPV, that models the variations of voltage
A DC-DC converter typically used in this application is the
and current provided by the photovoltaic array, depending on the
Boost converter in which the static gain is defined by:
local irradiance and temperature. The mathematical modeling of
this converter is necessary so that controllers can be designed for
the variables of the internal loop (voltage Vc1 or current iPV) of The chosen states are the currents in each inductor and the
the algorithms that aim to extract the maximum power of the PV voltages in each capacitor. Thus,
array. T
x  iL1 iL2 vC1 vC2  (5)

The system input vector u(t) is the VDC voltage at the output
of the DC-DC Quadratic Boost converter.
A. Analysis of voltages and currents in the first stage of
operation

Fig. 1. Modified Structure of the Quadratic Boost Converter for application in

photovoltaic systems.

In general terms, the P&O-MPPT (perturb and observe

maximum power point tracking) algorithm used in this work
determines, dynamically, the voltage of the photovoltaic array
that would provide the maximum power available from the Fig. 2. Quadratic Boost Converter operation step 1.
environmental conditions at that time. From this value, a
controller seeks to change the duty cycle of the SW1 switch, so In Step 1, shown in Figure 2, the switch SW1 and the diode
that the input voltage Vc1 of the converter is maintained at the D2 conduct, and the diodes D1 and D0 are reverse polarized. The
reference voltage, given by the MPPT logic, at the same time linear equations that describe this step are:
that is boosted to the stabilized output voltage value VDC.
.
Therefore, the objective of this paper is to present the x  A1 x  B1u (6)
controller design that will be used in the internal loop of the
P&O-MPPT algorithm to control the output voltage of a y  C1 x  E1u (7)
photovoltaic array. The PV system is intended to be connected
to the single-phase power grid and has the DC-DC Quadratic From the analysis of circuits via Kirchhoff's Laws:
Boost converter at the interface between the PV panels and the
DC bus.  1 
 0 0
L1
0
For this same DC-DC Quadratic Boost converter in this type
of application (MPPT), in [5] a PID compensator was used,
 
added to the action of a Notch filter, and in [6] a controller based  1
on the internal model with a degree of freedom (IMC-1DOF),  0 0 0
L2 
both designed to control the input voltage of said DC-DC A1    (8)
 1 1 
 C 
converter. 0 0
C1RPV
In this paper, the proposed controller uses the state feedback  1 
technique and, in its design, will be demonstrated the  1 
mathematical modeling of the DC-DC Quadratic Boost
 0  0 0
converter, the step-by-step determination of discrete gains that  C2 
feed the plant and results of computational simulation that show
B1   0 0 0 0
T
the viability of the control technique in question and corroborate (9)
its use.
II. MATHEMATICAL MODELING OF QUADRATIC BOOST
C1  0 0 1 0 (10)

E1  0
CONVERTER
(11)
Since Quadratic Boost converter, in continuous conduction
mode (CCM) has two distinct operating stages - I, for switch B. Analysis of voltages and currents in the second stage of
SW1 conducting and II for switch SW1 blocked – it is needed to operation
find the state space mathematical model of each stage and
average them, considering the period in which each step is
active. The mathematical model by state space consists of
representing linear systems by two temporal equations, the
system and the output, namely:
.
x (t )  Ax (t )  Bu(t ) (3)
Fig. 3. Quadratic Boost Converter operation step 2
y(t )  Cx(t )  Eu(t ) (4)
 In Step 2, shown in Figure 3, the switch SW1 and the diode
X med   x1 x4 
T
D2 are blocked, and the diodes D1 and D0 conduct. The linear
x2 x3 (24)
equations that describe this step are: Where
.
x  A2 x  B2u VDC (1  D )2
(12) x1   (25)
RPV
y  C2 x  E2u (13)
VDC (1  D )3
From the analysis of circuits via Kirchhoff's Laws: x2   (26)
RPV
 1 1
 0 0 
L1 L1  x3  VDC (1  D ) 2 (27)
 
 1  x4  VDC (1  D) (28)
 0 0 0
L2 
A2    (14) Considering that the system will operate around the point of
 1 1  equilibrium, it is necessary to model the small variation
 C 0 
C1RPV
0  dynamics of the system around this point, those variations in
 1  the duty cycle and their effects on the input voltage. In this way,
 1 1  the states, the input vector and the duty cycle are analyzed as
C  0 0  containing their average value, plus a small signal of time-
 2 C2  varying perturbation. The model of small signals is applied to
the state and output equations, simplifications are made and the
T
 1  Laplace transform is applied to obtain (29).
B2  0  0 0 (15)
  X ( s )  ( sI  Amed ) 1  Bmed U ( s )  Bd D( s )  (29)
L2 ~ ~ ~

C2  0 0 1 0 (16)
 
Where
E2  0 (17) Bd  [ A1  A2 ] X med  [ B1  B2 ]VDC (30)
The converter can be represented by matrices obtained Considering that there is no variation of the input U(s) of
through the average between the matrices of each step, ~
considering the time within the conducting period in which each the system, U ( s )  0 . Writing the transfer function of the
one is valid. So, variations of the states in relation to the variation of duty cycle,
Amed  A1D  A2 (1  D) (18) it is obtained (32):
~
Bmed  B1D  B2 (1  D) X ( s)
 ( sI  Amed ) 1  Asub X med  BsubVDC  (31)
(19)
~
Cmed  C1D  C2 (1  D) (20) D( s )
Where
Emed  E1D  E2 (1  D) (21)
Asub  A1  A2 (32)
Since this system is nonlinear, it is necessary to perform its
linearization by choosing a point of operation in a steady state in Bsub  B1  B2 (33)
which the derivatives of the state equations are zero. Hence, the
system represented in the form of equation (3) can be Being this system of fourth-order and one input, the
represented by equation (22). expression (31) is a 4x1 column matrix. Desiring to control the
disturbance of the input voltage of the converter, here
.
X med  Amed X med  BmedU med (22)
~
represented by v C 1 , in relation to the disturbance in the duty
. ~
After some algebra in (22), considering X med  0 in steady cycle d , the desired transfer function is
state, it is obtained (23): a2 s 2  a1s  a0
G~ ( s)  (34)
X med   Amed 1BmedU med (23) v C1 b4 s 4  b3s 3  b2 s 2  b1s  b0
~
d
Thus, the average state vector becomes (24):
Where
 a2  VDC (1  D)( RPV C2 L2 ) (35) III. OPEN-LOOP SYSTEM ANALYSIS
A mathematical software allows to obtain the system
a1  VDC (1  D )3 L2 (36) response to the unitary step, shown in Figure 4, where one can
se a certain oscillation in high frequency, caused by two open-
a0  2VDC (1  D) RPV (37) loop complex conjugated poles that have a high imaginary part,
in addition to the inversion of the temporal response, caused by
b4  RPV C1L1C2 L2 (38) two zeros in the right half-plane. In addition, the settling time
(time required for the signal to enter the ± 2% of steady state
value for the last time) is around 1ms.
b3  L1L2C2 (39)

b2  RPV [C1L1  C2 L2  C1L2 (1  D )3 ] (40)

b1  L1  L2 (1  D ) 2 (41)

b0  RPV (42)

C. Presentation of the electrical parameters

The inductance and capacitance values available for carrying
out the work are such that L1 = 982 mH, L2 = 1986 mH, C1 = C2
= 4.77 μF. The DC bus at the output of the converter is such that Fig. 4. Open-loop system response to unitary step.
VDC = 100 V. The average duty cycle that raises the maximum
power voltage from 61.2 V to 100 V, through (2), is D = 0.212. IV. DISCRETE INTEGRAL CONTROL WITH STATE FEEDBACK
The value of RPV, being a model for the photovoltaic
Figure 5 shows the discrete block diagram of the system
arrangement, is given by (43).
consisting of integral controller + plant + state feedback.
VMPP 2
RPV 
r(k) e(k) e(k+1) u(k) x(k) y(k)
(43) +_ +
+ Ke +_ Bx +
+
Z 1 C
PMPP Z 1 Ax

The photovoltaic array used provides a voltage of 62.10 V K2

for a 540 W maximum power. Table 1 shows the specifications
of each of the two panels that were connected in series, valid
values for 1000 W/m² irradiance and 25 ° C.
Fig. 5. Block diagram of system (integral controller, state feedback and plant).
TABLE I. ELECTRICAL PARAMETERS OF EACH PHOTOVOLTAIC PANELL
Model Risen RSM60-6-270P
Discrete Integral Control consists in determining the values
Maximum Power 270 W of the row-matrix K and the constant Ke such that the closed-
Open-circuit voltage 37.90 V loop poles of the system are the ones that provide the desired
Open-circuit current 9.20 A dynamics for the output. Integral control is necessary because it
Voltage at maximum power 31.05 V raises the system type, zeroing the steady-state error for step
Current at maximum power 8.70 A inputs of plants that do not contain any integrators in the path of
Efficiency 16 % forward action, as it is in this case, evidenced by (36). As a
Thus, at the point of operation adopted, RPV = 7.14 Ω. The consequence of this control, there is the increase of the order of
parameters of the voltage perturbation transfer function in the the system by 1. One way to determine the K and Ke gains is by
input capacitor of the converter relative to the disturbance in the using the Ackermann Formula with some modifications for
switch duty cycle for the given electrical parameters are shown discrete systems. The step-by-step of the determination of (44)
in Table II. is shown in [7]. Equations (44) to (49) are made considering that
the systems are represented in the controller canonical form and
TABLE II. PARAMETERS OF THE TRANSFER FUNCTION GIVEN IN (36) that they are of 4th order.
a2 -6.498×10-6
1
a1 0.09719 ^ G  I 4 H 
[ K1,4 K e ]  [ K  [01,4 1]] 
CH 
a0 -1370 (44)
b4
b3
3.863×10-16
9.312×10-12
 CG
b2 1.744×10-7 Where
b1 0.00215
b0 8.696 ^ ^
K  [01,4 1]Wc 1 ( A) (45)
 ^ ^ ^ ^2 ^ ^3 ^ ^ 4 ^ inferior time than the reference change given by the MPPT
Wc  [ B AB A B A B A B] (46) algorithm operating at 50 Hz, i.e., less than 20 ms. The closed-
loop settling time of 1 ms was adopted. The response adopted
^ G H for the output had the Bessel Type dynamics, which presents
A   (47) little overshoot and adjustable settling time. Figure 6 shows the
01,4 0 Bessel response for systems of order 1 to 5 with settling time of
1 second.
^ 04,1  The continuous poles providing the 5th order dynamics
B  (48) shown in Figure 6 are -6.4480; -4.1104 ± j6.3142; -5.9268 ±
1  j3.0813 rad/s. They should be divided by 1 ms so that the
response will have a settling time of 1 ms. In addition, one must
^
have to map them to z-plane by applying (53) to find the discrete
 ( A) is the closed-loop discrete characteristic equation for characteristic closed-loop equation given by (49). The choice of
z = Â. G and H are the discrete system and input matrices, the sampling period T must be made in accordance with the
respectively, represented in the controller canonical form. Nyquist's Theorem, in other words, T ≤ 2Ta, where Ta is the
^ ^5 ^4 ^3 ^2 ^
smaller period of the signal. An empirical way of choosing T is
 ( A)  A  1 A   2 A  3 A   4 A 5 I 5 (49) so that it is 6 to 10 times smaller than the settling time of the
signal. Thus, T should be between 0.1 ms and 0.16 ms. However,
The coefficients α are given after determination of the being more conservative, a T of 0.0025 ms was chosen.
closed-loop poles. The discrete representation of the closed-loop
system is given by (50).

 x(k  1)   x 
 e(k  1)   t  e  (50)
   
Where

 G H 
t  (51)
 K1,4  K1,4 G  K e CG 1  K1,4 H  K eCH 
Fig. 6. Bessel Type Response for 1st to 5th order systems.
The eigenvalues of the matrix t shown in (51) are the poles
of the closed-loop system (CL). Knowing that K1x4 gains are z  esT (53)
found for systems represented in the controller canonical form,
a similarity transformation is needed to find K2, which are the In this way, after the due calculations, the values of the array-
corresponding K1x4 values that bring the current system line K2 and the constant Ke are obtained: K2 = [0.06777; -
represented in any other form to have the CL poles given by the 0.07129; 0.002312; -0.001159] and Ke = -0.0001127.
eigenvalues of t. This is done through (52).
VI. COMPUTATIONAL SIMULATION RESULTS
1 1
K 2  K1,4 tWc  (52) The MPPT algorithm used was the Perturb and Observe
(P&O) algorithm. To validate the operation of the control, a
Where t is the controllability matrix of the open-loop plant time-varying irradiance profile was created in the range of 200
represented in its original form and WC is the controllability W/m² to 1000 W/m², with a temperature of 25 °C, to be
matrix of the open-loop plant represented in the controller implemented in PV panels. Figure 7 shows the output dynamics
canonical form. compared to the reference voltage, given by the MPPT
algorithm.
The great advantage of this controller is to allow the closed-
loop poles, as long as they lead to a stable system, to be allocated
arbitrarily. However, one does not have control over the
saturation of the control effort. Empirically, it is known that
when designing a control that requires controlled response with
dynamics not much faster than the open loop one, the control
action will not saturate. Computational simulations are
necessary at the designing stage of performance criteria to
ensure that the control effort is within acceptable limits.
V. DEFINITION OF THE DESIRED PERFORMANCE OF THE
RESPONSE AND DETERMINATION OF FEEDBACK GAINS
The controller of this plant must be capable of stabilizing the Fig. 7. Comparison of the output signal with the reference.
input voltage of the Quadratic Boost DC-DC converter at an
 It is possible to note certain voltage variations at times
multiple of 0.1 seconds, which is when steps of 200 W/m² in the
irradiance were chosen to be made in simulation. It is worth
mentioning that these abrupt variations of climatic conditions
were simulated only to ascertain the effectiveness of the control
and they are hardly repeated in practice.
Figure 8 shows an enlargement of Figure 7 in the range of
0.06 seconds to 0.14 seconds. In it, it can be seen that the input
voltage dynamics meets the design criteria determined in section
V.
Fig. 10. Amplification of Figure 9 in the time range of 0.2 to 0.3 seconds.

VII. CONCLUSION
Mathematical modeling using an average state space,
followed by linearization around an operating point, allowed one
to analyze the behavior of the system when subjected to
disturbances in the switch duty cycle. This model allowed to
identify the location of the poles and zeros in the open-loop of
the plant, evidencing that it has its own dynamics that cannot be
easily controlled by traditional techniques. From the
mathematical model, a discrete integral controller was designed
Fig. 8. Amplification of Figure 7 in the time range from 0.06 seconds to 0.14 by state feedback, which allowed to allocate the closed-loop
seconds.
poles where they guarantee a desired transient response and
which made it possible to bring to zero the steady-state error
Figure 9 shows the comparison of the maximum power that
between the given reference voltage by the MPPT algorithm and
could be delivered by the photovoltaic arrangement to that
the input voltage of the Quadratic Boost converter. The desired
extracted by the DC-DC converter. From it, one can see the
closed-loop poles were chosen from the standard Bessel Type
efficiency of the control in extracting the maximum power
response, which presents little overshoot and adjustable settling
available by the array.
time, and the responses obtained in a computational simulation
attest the location of these poles, thus validating the suggested
technique of control.

REFERENCES
[1] SNYMAN, D. B., ENSLIN, J. H. R., “An experimental evaluation of
MPPT converter topologies for PV installations”. Renewably Energy, vol.
3, issue 8, pp. 841-848, 1993.
[2] D. Maksimovic, S. Cuk, “Switching converters with wide DC conversion
range”, IEEE Transactions on Power Electronics, vol. 6, no.1, pp. 151–
157, janeiro 1991.
Fig. 9. Comparison of the maximum power available by the photovoltaic array [3] L. H. S. C. Barreto, Análise, projeto e desenvolvimento de conversores
to the extracted power. para a concepção de uma UPS online não isolada, Tese de Doutorado,
Universidade Federal de Uberlândia - UFU, Uberlândia-MG, Brasil,
2003.
Figure 10 shows an enlargement of Figure 9 in the period of
[4] M. L. Kumari, S. Bhattacharya and U. S. Triar, "Stabilization of boost
0.2 seconds to 0.3 seconds. The voltage reference step used in converter with output filter using LQR based state-feedback controller,"
the MPPT algorithm in simulation was 0.5 V. Representing a 2016 10th International Conference on Intelligent Systems and Control
small percentage - 0.8% - of the maximum power voltage VMPP (ISCO), Coimbatore, 2016, pp. 1-6.
- 61.2 V - it is noticed that, in steady state, there is little [5] R. A. S. Carvalho, L. S. Vilefort, F. V. R. Silva, L. C. G. Freitas, E. A. A.
oscillation around the maximum available power at the moment. Coelho, L. C. FREITAS, J. B. VIEIRA JR., “Estudo do Conversor Boost
However, for abrupt variations of irradiance, MPP tracking is Quadrático para Rastreamento de Máxima Potência em Sistemas
Fotovoltaicos Utilizando o Método Perturba & Observa”, Conferência de
slower. Estudos em Engenharia Elétrica, UFU, 2014.
[6] R. A. S. Carvalho, L. S. Vilefort, F. V. R. Silva, L. C. G. Freitas, E. A. A.
Coelho, L. C. FREITAS, J. B. VIEIRA JR., “Estudo do Conversor Boost
Quadrático para Rastreamento de Máxima Potência em Sistemas
Fotovoltaicos Utilizando o Método Perturba & Observa”, Conferência de
Estudos em Engenharia Elétrica, UFU, 2014.
[7] K. Ogata, Discrete-time Control Systems, Prentice-Hall International, 2nd
edition, New Jersey, 1995.