Download as pdf or txt
Download as pdf or txt
You are on page 1of 8

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/352992745

Sensorless Predictive Direct Power Control with On-line Inductance


Estimation for Grid-connected PV Applications

Conference Paper · May 2021

CITATION READS

1 165

4 authors:

Mostafa Ahmed Mohamed Abdelrahem


Technische Universität München Technische Universität München
40 PUBLICATIONS 233 CITATIONS 158 PUBLICATIONS 1,916 CITATIONS

SEE PROFILE SEE PROFILE

Ibrahim Harbi Ralph Kennel


Technische Universität München Technische Universität München
37 PUBLICATIONS 184 CITATIONS 799 PUBLICATIONS 16,133 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Mostafa Ahmed on 06 July 2021.

The user has requested enhancement of the downloaded file.


Sensorless Predictive Direct Power Control with On-line
Inductance Estimation for Grid-connected PV Applications
Mostafa Ahmed1,2 , Mohamed Abdelrahem1,2 , Ibrahim Harbi1 , Ralph Kennel1
1
Institute for Electrical Drive Systems and Power Electronics (EAL), Technical University of
Munich (TUM), Munich, Germany.
2
Electrical Engineering Department, Faculty of Engineering, Assiut University, Assiut, Egypt.

Corresponding author: Ibrahim Harbi, ibrahim.harbi@tum.de

Abstract
In this study, predictive direct power control technique is adopted for two-stage grid-connected PV systems.
The system consists of a photovoltaic (PV) source, boost converter, two-level inverter, and filter for the
grid connection. The proposed control strategy eliminates all the grid voltage sensors by employing an
extended Kalman filter (EKF). To enhance the robustness of the system against parameters variation,
the system’s inductance is included in the estimation process. Furthermore, the calculation burden
associated with the conventional predictive direct power control technique is reduced using a simplified
control methodology. The maximum power point tracking (MPPT) operation is accomplished using the
perturb and observe (P&O) method. The proposed control algorithm is validated via simulation results
under different atmospheric and operating conditions.

1 Introduction most common techniques are the voltage oriented


control (VOC) and the direct power control (DPC)
Day after day, renewable energy resources (RESs) method [5]. The VOC technique depends on
are becoming more dominant in the energy market. the cascaded control loop structure, where the
Among these resources, photovoltaic (PV) energy outer loop or voltage loop provides the reference
provides numerous merits including availability, current for the inner loop or the current loop.
quietness, cleanliness, etc [1]. However, the PV Then, the calculated reference voltages are applied
source exhibits nonlinear output characteristics. to the inverter using a modulator. The DPC
Thus, maximum power point tracker (MPPT) should structure avoids the cascading of the VOC by
be adopted with the PV source to extract the comparing the instantaneous active and reactive
maximum power from it [2]. Many MPPT techniques power with their references. Hence, the hysteresis
are investigated in the literature, where the main commands are adopted with a look-up table for
concerns of these methods are efficiency, required switching states generations [6]. Recently, model
sensors, tracking speed, implementation, and predictive control (MPC) techniques became very
cost [3]. Generally, the PV systems can be popular. MPC techniques are nonlinear control
classified into two prime categories, which are the approaches, in which the control law can handle
single-stage and the two-stage topologies. The different constraints and objectives. Among MPC
two-stage topology involves the DC-DC converter techniques, the finite set model predictive control
stage (first stage), in which the MPPT function is (FS-MPC) is more widespread, in which the
achieved. The second stage (DC-AC or inversion discrete-time model of the system is developed
stage) secures the active and reactive power control for the sake of system’s control. Furthermore,
management [4]. the optimal switching state is selected according
to a quality function design. The quality function
Previously, several methods have been addressed can be designed to include more than one control
for active and reactive power control. However, the parameter [4, 7, 8].
In this paper, two-stage grid-connected PV system where L is the boost converter inductance, cdc is
is proposed, where the MPPT is executed the coupling capacitance, and d is the duty cycle.
using the perturb and observe (P&O) method.
The two-level inverter connects the first stage (boost
Furthermore, the active and reactive power control
stage) with the grid by processing the DC power
is implemented using a simplified control structure
from the PV source. With reference to Fig. 1, this
of the predictive direct power control to reduce
part from the system can be analyzed as
the computation burden. All the sensors of
the grid voltages are eliminated, where the diabc
vabc = uabc + Lf + Rf iabc , (4)
voltages are estimated using a powerful extended dt
Kalman filter (EKF) estimator. Moreover, the where vabc are the grid-side voltages, iabc are the
inductance is involved in the estimator to improve line currents, uabc are the output voltages of the
the system’s robustness. The proposed system and inverter, Lf is the filter inductance, and Rf is the
control strategy are validated using Matlab/Simulink filter resistance. Further, the output voltages of the
simulation results. The verification considers two-level inverter are extracted as [9]
different atmospheric and operating conditions. 1
uabc = vdc Tabc Sabc , (5)
3
2 Modeling of the grid-connected where vdc is the DC-link voltage, Sabc is the
PV system switching state vector, and Tabc is the transformation
matrix, which can be expressed as [9]
The studied PV system is composed of a PV source,  
boost converter, two-level inverter, and filter with 2 −1 −1
grid connection. The details of the proposed control Tabc = −1 2 −1 . (6)
topology and components are illustrated in Fig. 1. −1 −1 2
The PV source can be described by the single diode In the stationary reference frame (α-β), (4) can be
equivalent model as follows rewritten as
v +i R
( pv pv s ) vpv + ipv Rs diαβ
ipv = iph − io [e nNs vt − 1] − , (1) vαβ = uαβ + Lf + Rf iαβ , (7)
Rsh dt
where iph is the photovoltaic current, n is the where vαβ and iαβ are the voltages and currents in
ideality factor of the diode, io is the diode saturation αβ reference frame. Furthermore, the active and
current, Rs is the series resistance, Rsh is the reactive power in the same reference frame are
shunt resistance, vt is the thermal voltage, Ns is given by
the number of cells, ipv is the PV output current, P = 32 (vα iα + vβ iβ ),
and vpv is the PV output voltage. (8)
Q = 32 (vβ iα − vα iβ ).
The PV array is followed by a boost converter to
adjust the array voltage and perform the MPPT 3 Classical predictive direct power
operation. The behavior of the boost converter is control for the PV system
specified by the actions of its switch, thus the state
space model of the boost is derived as The conventional predictive direct power control
technique is executed based on the discrete-time
ẋ = Ax + Bu,
(2) model of the system, thus modifying (8) gives [10]
y = Cx + Du,
Pp = 32 [vα (k + 1)iα (k + 1) + vβ (k + 1)iβ (k + 1)],
where x = [ipv vdc ]T is the state vector, u =
Qp = 23 [vβ (k + 1)iα (k + 1) − vα (k + 1)iβ (k + 1)],
[vpv iinv ]T is the input vector, and y = vdc is the
(9)
output. Further, A, B, C, and D are the system
where Pp and Qp are the predicted active and
matrices and are expressed as
reactive power, iα (k + 1) and iβ (k + 1) are the
0 − 1−d
  1 
L L 0 predicted currents. Reordering (7) gives
A = 1−d ,B = ,
cdc 0 0 − c1dc diα R 1
(3) dt = − Lff iα + Lf (vα − uα ),
diβ R 1
(10)
= − Lff iβ + Lf (vβ − uβ ).
 
C = 0 1 , D = 0, dt
RL filter
PV ipv DC-DC Converter Lf ia Rf

+
v cpv
pv Boost Converter
cdc +
vdc
ib
Grid
+- - ic
Two-level Inverter

Sboost iabc
ipv vpv Sinverter
vdcref
-
+ Gate Drive abc/
MPPT
i i
PI
Minimization of

the Cost Function


Sinverter vˆ 
EKF
u(k) u(k) v
Qref Pref dc vˆ 
Sector Selection Lˆ f
Reference Currents
i ref (k)
RVV Calculation
Estimation
i ref (k) vˆ (k)
i(k)
vˆ (k)
i(k)
Fig. 1: Topology of sensorless predictive direct power control for grid-connected PV system.

The predicted currents are calculated as available switching actions (8 states) of the two-
Ts Rf Ts level inverter according to a cost function design.
iα (k + 1) = (1 − Lf )iα (k) + Lf (vα (k) − uα (k)),
Ts Rf
However, the proposed and modified predictive
Ts
iβ (k + 1) = (1 − Lf )iβ (k) + Lf (vβ (k) − uβ (k)), direct power technique identifies the sector of the
(11) reference voltage vector (RVV) as follows
where Ts is the sampling period. Further, the
Firstly, the RVV is calculated as
predicted grid-voltages can be obtained using linear
extrapolation as L̂f
uαref (k) = −Rf iα (k) − Ts (iαref (k + 1) − iα (k))
vα (k + 1) = 2 vα (k) − vα (k − 1), + v̂α (k),
(12) L̂f
vβ (k + 1) = 2 vβ (k) − vβ (k − 1). uβref (k) = −Rf iβ (k) − + 1) − iβ (k))
Ts (iβref (k
Similarly, the predicted references of active and + v̂β (k),
reactive power can be evaluated as (15)
where Rf is the filter resistance, L̂f is the estimated
Pref (k + 1) = 2 Pref (k) − Pref (k − 1),
(13) filter inductance, v̂α (k) and v̂β (k) are the estimated
Qref (k + 1) = 2 Qref (k) − Qref (k − 1).
grid voltages in α-β reference frame using EKF
Finally, the optimal voltage vector among the as will be investigated at later step in this section,
available switching states is selected according the iα (k) and iβ (k) are the injected grid-currents, and
cost function design as iαref (k + 1) and iβref (k + 1) are the predicted
g1 = |P (k+1)−Pref (k+1)|+|Q(k+1)−Qref (k+1)|. reference currents, where they can be estimated
(14) from the active (Pref ) and reactive (Qref ) power
references as
4 Proposed sensorless predictive iαref (k + 1) = 2 v̂α
2 (k)+v̂ 2 (k) Pref
3 v̂α + 2 v̂β
2 (k)+v̂ 2 (k)
3 v̂α
direct power technique with ∗ Qref ,
β β

inductance estimation v̂β v̂α


iβref (k + 1) = 23 v̂2 (k)+v̂ 2 (k) Pref −
2
2 (k)+v̂ 2 (k)
3 v̂α
α β β
The conventional predictive direct power control ∗ Qref .
selects the best switching state among the whole (16)
Secondly, the position (sector) of the RVV in the uncertainty and measurement noise are not known,
α-β frame can be computed as so the EKF is implemented as follows:

δ(k) = atan2(uβref (k), uαref (k)). (17) x̂(k + 1) = Ad x̂(k) + Bd u(k) + K(k)(y(k) − ŷ(k)),
ŷ(k) = Cd x̂(k) + Dd u(k),
Lastly, the cost function subjected to minimization (22)
is defined as where K(k) is the Kalman gain, x̂(k) and ŷ(k) are
g2 = |uα (k) − uαref (k)| + |uβ (k) − uβref (k)|, (18) the estimated quantities.
The design of EKF can be performed through
where uα (k) and uβ (k) are the voltage vectors
two stages of prediction and modification. The
in the selected sector and adjacent to the RVV
prediction phase includes the state vector prediction
specified from (17). Thus, using these two vectors
and the covariance matrix error prediction as follows
along with one zero voltage vector, the number of
required computation of the cost function decreases x̂− (k) = Ad x̂(k − 1) + Bd u(k − 1). (23)
from 8 for the conventional predictive direct power
technique to only 3 calculation for the proposed one. P− (k) = f (k)P(k − 1)f (k)T + Q, (24)
The grid voltages are estimated employing the where
EKF, thus the sensors required for the voltages are

eliminated (3 sensors) establishing reduced sensor f (k) = (Ad x(k) + Bd u(k))|x̂− (k) . (25)
∂x
requirement, which reduces greatly the cost with
reliability enhancement. The state space model for The correction stage is developed as
voltage estimation can be obtained as
K(k) = P− (k)CT − T −1
d (Cd P (k)Cd + R) . (26)
ẋ = Ax + Bu + w,
(19)
y = Cx + Du + v, x̂(k) = x̂− (k) + K(k)(y(k) − Cd x̂− (k)). (27)

where x = [iα iβ vα vβ Lf ]T is the state vector, u = P(k) = P− (k) − K(k)Cd P− (k). (28)
[(v̂α − uα ) (v̂β − uβ )]T is the input, y = [iα iβ ]T is
the measurement, w is the system uncertainty with
covariance matrix Q, and v is the measurement 5 Simulation results and
noise with covariance matrix R. Further, A, B, C,
discussion
and D are the inverter system matrices, and based
on (10) they are defined as The PV system is composed of two stages, which
 Rf 
1
0
 are the boost and the inversion stages. The
−L 0 0 0 0 Lf
MPPT is managed at the boost section, where the
 f  0 L1f 
 
Rf
− well-known perturb and observe (P&O) algorithm

 0 0 0 0
A= Lf ,B =  0 ,
  
 0 0 is executed for power harnessing. P&O is a
0 0 0 0  
0 0

0 0 0 0 0 simple technique to implement. Furthermore, it
0 0 is independent of the PV array parameters [11].
  The second level (inverter) is responsible for
1 0 0 0 0 the active and reactive power control, which is
C= , D = 0. (20)
0 1 0 0 0 developed by sensorless predictive direct power
control with reduced computational burden in
this study. The proposed methodology makes
Therefore, the discrete model can be expressed as benefit from integrating the EKF with the predictive
x(k + 1) = Ad x(k) + Bd u(k) + w(k), technique to eliminate all the grid-voltage sensors.
(21) Moreover, the inductance of the system is included
y(k) = Cd x(k) + Dd u(k) + v(k),
in the estimation process to enhance the robustness
where Ad = I + ATs , Bd = BTs , Cd = C, Dd = D, of the direct power method. Thus, it is expected to
and I is the identity matrix. Normally, the system improve the power quality injected into the grid. The
G=800 W/m2 G=1000 W/m2
G=600 W/m2
10
G=400 W/m2

T=25 oC
0
500
400
300
200

40

20

0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4

Fig. 2: Performance of P&O algorithm at different atmospheric conditions of radiation.

Tab. 1: Parameters of PV grid-connected system. Figure 3 shows the active power, reactive power,
reference currents and actual currents in αβ
Parameter Value
reference frame, and the abc currents for the
PV array power (kW) 15
proposed sensorless predictive power strategy with
Boost inductance L (mH) 5
and without inductance estimation, respectively. In
DC-link capacitance cdc (µF) 1000
the first interval of simulation results, the inductance
Filter inductance Lf (mH) 12
is underestimated to half of the nominal value, and
Filter resistance Rf (Ω) 0.25
in the second interval the inductance of the model
DC-link reference voltage vdcref (V) 700
equals to that of the nominal one. Finally, the
Grid-frequency ω (rad/s) 2π × 50
inductance of the model is 1.5 times of the actual
Grid line-line voltage v (V) 400
value. The simulation results in the first period with
Sampling time Ts [µs] 40
the proposed methodology show an improved active
and reactive power tracking in comparison with the
model without inductance estimation, where the
details of the studied system is presented in Table 1.
active power shows large spikes. Furthermore, the
reactive power shows higher peak-to-peak ripples.
Figure 2 shows the behavior of the P&O method Thus, the αβ and the abc currents exhibit a distorted
under different atmospheric conditions of radiation. sinusoidal current injection into the grid. To be
The studied conditions are 400 W/m2 , 600 W/m2 , specific, in the first period, the THD of the currents
800 W/m2 , and 1000 W/m2 , respectively. The with the proposed sensorless predictive power
results show the extracted PV power, the PV array control is 3.64%, where the value without inductance
voltage, and the PV array current, where the P&O estimation is 6.90% leading to deterioration of the
tracks the available power at every step change of injected currents. Furthermore, the THD goes
the radiation. The PV array voltage changes in a beyond the standard IEEE values [12].
narrow range. However, the PV current behaves
like staircase due the large dependency on the The estimated grid voltages in αβ reference frame
radiation [5]. One can say that the PV power’s and the inductance of the system are presented in
variation is mainly related to the PV current when Figure 4, where the grid voltages are estimated with
the radiation is subjected to change. high accuracy leading to enablement of sensorless
10 10
0 0

2 2
0 0
-2 -2

20 20
0 0
-20 -20

20 20
0 0
-20 -20

20 20
0 0
-20 -20
0 0.05 0.1 0.15 0.2 0.25 0.3 0 0.05 0.1 0.15 0.2 0.25 0.3

Fig. 3: Simulation results of the grid-connected PV inverter (from top): Injected active power, reactive power, α
reference frame current, β reference frame current , and the abc currents for: (a) Sensorless predictive power
control with inductance estimation, and (b) without inductance estimation.

operation. The estimated inductance diverges a stability is preserved during abrupt variation of
little bit from the reference value, especially at the inductance. The proposed methodology can
high inductance values, where the maximum error be used as a back-up control in case of sensors’
between the actual and the estimated inductance failure.
is about 3.7%. The estimation process of the
inductance plays an important role in case of control References
techniques that rely on the system’s parameters
like the current research point. Therefore, the [1] Z. Qian, O. Abdel-Rahman, H. Hu, and
robustness of the system is improved and this I. Batarseh, “An integrated three-port inverter
reflects on the power quality injected into the grid. for stand-alone pv applications,” in 2010 IEEE
Energy Conversion Congress and Exposition.
IEEE, 2010, pp. 1471–1478.
6 Conclusion
[2] A. Loukriz, M. Haddadi, and S. Messalti,
Sensorless predictive direct power control “Simulation and experimental design of a
technique is proposed in this study, where the new advanced variable step size incremental
voltage sensors of the grid are eliminated and EKF conductance mppt algorithm for pv systems,”
is utilized to estimate theses voltages. Furthermore, ISA transactions, vol. 62, pp. 30–38, 2016.
the inductance of the system is involved in the
[3] A. Belkaid, I. Colak, and O. Isik, “Photovoltaic
estimation for robustness and power quality
maximum power point tracking under fast
enhancement. The calculation burden of the
varying of solar radiation,” Applied energy, vol.
conventional predictive direct power technique is
179, pp. 523–530, 2016.
further reduced by identifying the sector of the
reference voltage from the commanded reference [4] M. Ahmed, M. Abdelrahem, and R. Kennel,
currents. The results show that the proposed “Highly efficient and robust grid connected
methodology has a fast transient behavior, and photovoltaic system based model predictive
robust. Moreover, no need for parameters tuning or control with kalman filtering capability,”
predefined switching tables, even more the system’s Sustainability, vol. 12, no. 11, p. 4542, 2020.
400
200
0
-200
-400

20

15

5
0 0.05 0.1 0.15 0.2 0.25 0.3

Fig. 4: Estimation results using EKF for: (a) The αβ grid-voltages, and (b) system’s inductance.

[5] M. Ahmed, M. Abdelrahem, I. Harbi, and windkraftanlagen: Eine einführung,”


R. Kennel, “An adaptive model-based mppt in Elektrische Antriebe-Regelung von
technique with drift-avoidance for grid- Antriebssystemen. Springer, 2015, pp.
connected pv systems,” Energies, vol. 13, 1540–1614.
no. 24, p. 6656, 2020. [10] P. Cortes, J. Rodríguez, P. Antoniewicz, and
M. Kazmierkowski, “Direct power control of an
[6] I. Hammoud, K. Morsy, M. Abdelrahem, and afe using predictive control,” IEEE Transactions
R. Kennel, “Efficient model predictive power on Power Electronics, vol. 23, no. 5, pp. 2516–
control with online inductance estimation for 2523, 2008.
photovoltaic inverters,” Electrical Engineering,
vol. 102, no. 2, pp. 549–562, 2020.
[11] M. Ahmed, M. Abdelrahem, R. Kennel, and
[7] Y. Zhang, W. Xie, and Y. Zhang, “Deadbeat
C. M. Hackl, “Maximum power point tracking
direct power control of three-phase pulse-width
based model predictive control and extended
modulation rectifiers,” IET Power Electronics,
kalman filter using single voltage sensor for
vol. 7, no. 6, pp. 1340–1346, 2014.
pv systems,” in 2020 IEEE 29th International
[8] Y. Zhang, Y. Peng, and C. Qu, “Model Symposium on Industrial Electronics (ISIE).
predictive control and direct power control IEEE, 2020, pp. 1039–1044.
for pwm rectifiers with active power ripple
minimization,” IEEE Transactions on Industry
Applications, vol. 52, no. 6, pp. 4909–4918, [12] R. Langella, A. Testa, and E. Alii, “Ieee
2016. recommended practice and requirements for
harmonic control in electric power systems,”
[9] C. Dirscherl, C. Hackl, and K. Schechner, University of Campania âLuigi Vanvitelliâ:
“Modellierung und regelung von modernen Caserta, Italy, 2014.

View publication stats

You might also like