Sunteți pe pagina 1din 5

Enhanced Model Predictive Voltage Control of

Four-Leg Inverters with Switching Frequency


Reduction for Standalone Power Systems
V. Yaramasu1, B. Wu1 , M. Rivera2 and J. Rodriguez 2
(1) Ryerson University, Toronto, ON M5B 2K3 CANADA
Phone: 416-979-5000 x 6484, Fax: 416-979-5280, Email: vyaramas@ee.ryerson.ca,
URL: http://www.ee.ryerson.ca/
(1) Universidad Tecnica Federico Santa Maria, Av. Espana 1680, Valparaiso, CHILE
Phone: +56-32-265-4000, Fax: +56-32-265-4000, Email: Marco.Rivera@usm.cl, URL: http://www.usm.cl

Abstract—This paper presents a model predictive voltage higher prediction horizon is necessary to improve the control
control strategy with a prediction horizon of two samples for performance [9], [10]. The output voltage control for a two-
enhanced output voltage control of three-phase four-leg inverters level four-leg inverter using one step prediction has been
used for standalone power systems. The four-leg inverter is
developed to deliver symmetrical sinusoidal three-phase volt- analyzed in [11]. In [12], two-step prediction for a three-
ages, irrespective of the arbitrary consumer load profiles. The phase UPS system is presented. However, this method requires
proposed controller uses the discrete-time model of the inverter greater computational capacity to handle the large amount
and RLC filter for two-step prediction of output voltage with of predictions. In this paper, the concept of FCS-MPC with
fewer commutations so as to reduce the switching frequency. two-step prediction has been extended to four-leg inverters
The control method chooses a switching state that minimizes
the error between the output voltage and its reference. The to improve the output voltage regulation while reducing the
proposed controller offers an excellent tracking reference with number of computations and converter switching frequency
less harmonic distortion voltage for balanced, unbalanced and for balanced, unbalanced and nonlinear loads.
nonlinear loading conditions. The feasibility of the proposed
control scheme has been verified by MATLAB/Simulink. II. F OUR -L EG I NVERTER M ODEL
The typical standalone hybrid power system, which consists
I. I NTRODUCTION of PV arrays, PMSG WECS, battery banks, a three-phase four-
leg inverter, an RLC filter, and arbitrary loads is shown in
The four-leg inverter with output RLC filter is proved to
Fig. 1. The DC/DC converter-1 and AC/DC converter perform
be the best candidate to provide transformerless neutral con-
maximum power point tracking to extract maximum possible
nection and symmetrical sinusoidal voltages to the consumers
energy from the sun and wind respectively. The loads are
in the standalone power systems. Voltage control methods of
unknown and can be single-phase or three-phase, balanced
four-leg inverters using hysteresis regulators [1]; open-loop
or unbalanced, linear or non-linear. The connection format of
feed forward controllers [2]; linear PID controllers [3] with
a four-leg inverter is similar to the conventional three-phase
external voltage and internal current control loops; and pole
inverter with the fourth leg connected to the neutral point of
placement controllers [4] have been analyzed before. Most
the load. The fourth leg increases switching states from 8 (2 3 )
of these complex control methods require 3D-space vector
to 16 (24 ) and thus offers control flexibility and improved
modulation [5], which is quite complicated, time consuming,
output voltage quality. The voltage in any leg x of the inverter,
and non-intuitive for software and hardware implementation.
measured from the negative point of the DC-link (N ) can be
The finite control set model predictive control (FCS-MPC) has
expressed in terms of switching states as,
found recent applications in power electronics. This method is
an attractive alternative to classical control methods, due to its vxN = Sx vdc , x = u, v, w, n. (1)
simple concept, fast dynamic response, and easy inclusion of Hence, the voltage applied to the output RLC filter, in terms
nonlinearities & constraints in the controller design. Moreover, of these inverter voltages is:
this scheme does not require internal current control loops
and modulators and thus greatly reduces the complexity. This vyn = vyN − vnN = (Sy − Sn ) vdc , y = u, v, w. (2)
control technique has successfully been applied to a wide The differential equations for the output filter, in terms of
range of power converters [6]–[8]. The FCS-MPC methods voltage and current vectors, are described as follows:
usually consider one step horizon because the number of di
switching states and the order of the load models are re- vo = v − Lf − Rf i, (3)
dt
duced. However, in the case of complex systems and special dvo
applications such as voltage control in standalone systems, a i = io + Cf , (4)
dt
15th International Power Electronics and Motion Control Conference, EPE-PEMC 2012 ECCE Europe, Novi Sad, Serbia

DC PCC 4L–Inverter
PV Array DC/DC Converter-1 P
AC Filter
Su Sv Sw Sn
Rf Lf Unknown Load
Wind Turbine iu iou
u
Gear Box PMSG AC/DC Converter
v
Cdc
w
n
DC/DC Converter-2
Battery Banks
Su Sv Sw Sn vou
Cf
+ vnN

Figure 1. Standalone hybrid power system with four-leg inverter and output RLC filter.

where vo , io , v and i are output and inverter voltage and current io can be estimated with the Lagrange extrapolation
current vectors respectively. The system in (3) and (4) can be method of order 4 [8]:
represented in space state form as follows:
      io [k + 1] = 4 io [k] − 6 io[k − 1] + 4 io[k − 2] − io[k − 3]. (9)
v˙o v v
=A o +B , (5) C. Minimization of Cost Function
i̇ i i o
The cost function for N step prediction horizon can be
where,
    defined as follows:
0 1/Cf 0 −1/Cf
A= , B= . (6) 
N
−1/Lf −Rf /Lf 1/Lf 0 g= αn ||vo [k + n]∗ − vo [k + n]||, (10)
III. M ODEL P REDICTIVE VOLTAGE C ONTROL S CHEME n=1

The proposed model predictive voltage control scheme is where αn is an arbitrary weighting factor. The output voltage
shown in Fig. 3. This method uses the inherent discrete nature vector equals its reference when g = 0. Therefore, the
of the four-leg inverter and RLC filter to predict the output objective of the cost function considered in this paper is
voltage vector for a predefined horizon time k + N , in terms to achieve g value close to zero. The voltage vector that
of the measured voltage and current in time k, and it selects minimizes the cost function is chosen and then applied at the
a switching state based on the minimization of cost (quality) next sampling instant.
function for each sampling time. IV. S IMULATION R ESULTS
A. Discrete-Time Model for Predictive Control To validate the proposed control scheme, a simulation model
The cost function requires the predicted output voltage for the three-phase four-leg inverter with the parameters as
vector vo [k + N ] in discrete-time form. For this reason, the indicated in Table I, has been developed using MATLAB-
space-state system in (5) can be represented in discrete-time Simulink. Four different loading conditions are considered:
as follows: balanced loads, unbalanced loads, nonlinear, and hybrid, which
      is a combination of linear and nonlinear loads. The nonlinear
vo [k + N ] vo [k + N − 1] v[k + N − 1]
=Φ +Γ , load models used in the simulation are shown in Fig. 4. All the
i [k + N ] i [k + N − 1] io [k + N − 1]
results are presented in a peak per-unit system to simplify the
(7)
analysis. The following three variations of the MPC algorithms
where,
are used in the simulation:
Φ = eTs A , Γ = A−1 (Φ − I2x2 )B, (8)
1) MPC-1: Prediction horizon of N=1 and 16 switching
B. Two-Step Prediction Horizon states combinations.
For one-step prediction horizon, the 16 switching states at 2) MPC-2: Prediction horizon of N=2 and 256 switching
sampling time k are used to predict the inverter voltage v[k] as states combinations.
shown in Fig. 2a. For a prediction horizon of N =2, the number 3) MPC-3: Prediction horizon of N=2 and 16 switching
of feasible switching states K becomes 256, as shown in Fig. states combinations.
2b. To reduce the switching changes (switching frequency) and The phase-u output voltage with three MPC methods for
the computational burden caused by these 256 predictions, a a balanced loading condition is shown in Fig. 5. The MPC-2
modified two-step prediction is proposed, as shown in Fig. and MPC-3 track references very well compared to the MPC-
2c. The proposed predictive method uses 16 switching states 1. The output voltages, line and neutral currents with MPC-3
projected to the sampling time k + 2. The unknown output are shown in Fig. 6. The neutral current (i n = iou + iov + iow )

2
15th International Power Electronics and Motion Control Conference, EPE-PEMC 2012 ECCE Europe, Novi Sad, Serbia

v[k] v[k + 1] v[k + 2] v[k] v[k + 1] v[k + 2] v[k] v[k + 1] v[k + 2]


1 1

1 1 1

16 16 16
256 16
k k+1 k+2 k k+1 k+2 k k+1 k+2
(a) (b) (c)
Figure 2. Four-leg inverter output voltages using: (a) one-step prediction, (b) two-step prediction, (c) modified two-step prediction.

4L-Inverter AC filter Table I


Load
i[k] Rf Lf io [k] F OUR - LEG I NVERTER AND L OAD PARAMETERS

u, v, w Variable Description Value


vdc [k] vo [k] Cf vdc DC-link voltage 515 [V ]
Cdc DC-link capacitor 1000 [µF ]
n
An DC-link noise amplitude 2 [V ]
fn DC-link noise frequency 100 [Hz]
8 S[k] fo∗ Reference output frequency 50 [Hz]

Minimization of vo∗ Peak reference output voltage 220 2 [V ][1.0 pu]

cost function g vo∗ [k + N] ioB Peak base load current 22 2 [A][1.0 pu]
with (10) Rf Filter resistor 0.05 [Ω]
Lf Filter inductor 2 [mH]
Cf Filter capacitor 80 [µF ]
16N vo [k + N ] vdc [k]
Prediction of R Load resistance 10 [Ω]
vo [k]
output voltage v o L Load inductance 1 [mH]
io [k]
C Load capacitance 120 [µF ]
with (7) and (8) i[k]
Ts Sampling time 50 [µs]

Figure 3. Block diagram of predictive voltage control for four-leg inverter.

u L
is observed to be zero, as shown in Fig. 6c. For the unbalanced u
loading condition, a step change in phase-u, v, w from 1pu, R
n
C R
0pu, 1pu to 2pu, 1pu, 0.75pu, respectively, is applied at time
L
t = 4.5π. This is the typical case for most of the standalone
power systems. The phase-u output voltage with three MPC n
methods is shown in Fig. 7. The MPC-2 and MPC-3 track Figure 4. Nonlinear load models: (a) nonlinear RL load, (b) nonlinear RLC
references with much less error than the MPC-1. The load load.
step change does not perturb the output voltage control, as
observed in Fig. 7. The output voltages, line, and neutral
currents with MPC-3 are shown in Fig. 8. The neutral current
in , which is sinusoidal in nature, flows through the fourth leg shown in Fig. 12. The neutral current i n , which is nonlinear
as shown in Fig. 8c. The load model, shown in Fig. 4a, is in nature, flows through the fourth leg as shown in Fig. 12c. A
used to simulate non-linear loading condition. A step change detailed measurement of error voltage, e v and % THD for the
from no-load to nonlinear load is applied at time t = 4.5π. output voltage are summarized in Table IV for the three MPC
The MPC-1 generates more ripple than MPC-2 and MPC-3, methods. The MPC-2 and MPC-3 provide a higher quality
as shown in Fig. 9. The non-linear current flowing through power supply in a typical standalone power system compared
the fourth leg does not affect the output voltage control, as to the MPC-1. Moreover, compared to the MPC-2, the MPC-3
shown in Fig. 10. The hybrid load with linear load on phase- reduces the number of calculations and switching frequency,
u, nonlinear RL load (Fig. 4a) on phase-v, and nonlinear as shown in Fig. 13. Thus this method is most suitable
RLC load (Fig. 4b) on phase-w is considered. The output for hardware implementation in current microprocessors. The
voltage control is shown in Fig. 11 with three MPC methods. average switching frequency is obtained by counting each ON-
The output voltages, line and neutral currents with MPC-3 are OFF change for each switch within one signal period.

3
15th International Power Electronics and Motion Control Conference, EPE-PEMC 2012 ECCE Europe, Novi Sad, Serbia

vou (pu) vou (pu)

1 1

∗ ∗
vou MPC–2 MPC–3 vou MPC–2 MPC–3

MPC–1 MPC–1

0 0
Step Change in Load
4π 4.2π 4.4π 4.6π 4.8π t(s) 4π 4.2π 4.4π 4.6π 4.8π t(s)

Figure 5. Comparison of three MPC methods for balanced loads. Figure 9. Comparison of three MPC methods for nonlinear loads.
vo (pu) vo (pu)
vou vov vow vou vov vow
1 1
0 0
−1 −1
(a) (a)
io (pu) io (pu)
iou iov iow iou iov iow
1 1
0 0
−1 −1
(b) (b)
in (pu) in (pu)
0.5 0.5
0 0
−0.5 −0.5
Step Change in Load
0 2π 4π 6π t(s) 0 2π 4π 6π t(s)
(c) (c)
Figure 6. Simulation results for MPC–3 with balanced loads: (a) output Figure 10. Simulation results for MPC–3 with non-linear load: (a) output
voltages, (b) output currents, and (c) neutral current. voltages, (b) output currents, and (c) neutral current.
vou (pu) vou (pu)

1 1

∗ ∗
vou MPC–2 MPC–3 vou MPC–2 MPC–3

MPC–1 MPC–1

0 0
Step Change in Load
4π 4.2π 4.4π 4.6π 4.8π t(s) 4π 4.2π 4.4π 4.6π 4.8π t(s)

Figure 7. Comparison of three MPC methods for unbalanced loads. Figure 11. Comparison of three MPC methods for hybrid loads.
vo (pu) vo (pu)
vou vov vow vou vov vow
1 1
0 0
−1 −1
(a) (a)
io (pu) io (pu)
iou iov iow iou iov iow
1 1
0 0
−1 −1

(b) (b)
in (pu) in (pu)
0.5 0.5
0 0
−0.5 −0.5
Step Change in Load
0 2π 4π 6π t(s) 0 2π 4π 6π t(s)
(c) (c)
Figure 8. Simulation results for MPC–3 with unbalanced loads: (a) output Figure 12. Simulation results for MPC–3 with hybrid load: (a) output
voltages, (b) output currents, and (c) neutral current. voltages, (b) output currents, and (c) neutral current.

4
15th International Power Electronics and Motion Control Conference, EPE-PEMC 2012 ECCE Europe, Novi Sad, Serbia

Table II
C OMPARISON OF OUTPUT VOLTAGE ERROR ev AND %THD FOR THREE The main limitation to implementing Model based Predic-
MPC METHODS . tive Control (MPC) algorithms is the computational effort
required. Evidently, this issue is daily becoming less important,
Before Load Step After Load Step
due to the development of new and faster processors. This
MPC–1 MPC–2 MPC–3 MPC–1 MPC–2 MPC–3
ev (V ) 9.26 6.29 7.52 9.26 6.29 7.52
issue, the comparison with classical techniques and the effect
Balanced Load
% THD 3.54 1.99 2.63 3.54 1.99 2.63 of parameter variations will all be part of a future work.
ev (V ) 10.54 6.36 7.82 8.49 5.68 6.79
Unbalanced Load
% THD 4.04 2.02 2.71 3.15 2.05 2.64 ACKNOWLEDGMENTS
ev (V ) 11.99 7.46 8.39 12.01 7.57 9.56 The authors wish to thank the financial support from the
Nonlinear Load
% THD 4.79 2.18 2.79 4.77 2.65 3.69
Natural Sciences and Engineering Research Council of Canada
ev (V ) 10.62 6.46 7.67 10.9 6.10 7.81
Hybrid Load (NSERC) through Wind Energy Strategic Network (WESNet)
% THD 4.08 2.09 2.65 4.22 1.99 2.67
Project 3.1, from the Universidad Tecnica Federico Santa
Maria and Basal Project FB0821.
5568 5480 5506 MPC–1
5109 5315
Switching Frequency [Hz]

5025 4854 4645 MPC–2


4584 4595 4685 4649 MPC–3 R EFERENCES
4000 [1] X. Zhang, J. Wang, and C. Li, “Three-phase four-leg inverter based on
voltage hysteresis control,” in Proc. IEEE Int. Electrical and Control En-
gineering Conf. ICECE, Jun. 2010, pp. 4482–4485, Dhaka, Bangladesh.
[2] W. Muangjai and S. Premrudeepreechacharn, “Implementation of a
2000 carrier-based three-dimensional space vector PWM technique for three-
phase four-leg voltage source converter with microcontroller,” in Proc.
IEEE Conf. on Industrial Electronics and Applications, ICIEA, May
2009, pp. 837–841, Singapore.
0
Balanced Load Unbalanced Load Nonlinear Load Hybrid Load [3] I. Vechiu, O. Curea, H. Camblong, S. Ceballos, and J. Villate, “Digital
control of a three-phase four-leg inverter under unbalanced voltage
Figure 13. Comparison of switching frequency for three MPC methods. conditions,” in Proc. European Conf. on Power Electronics and Ap-
plications, EPE, Sep. 2007, pp. 1–10, Aalborg, Denmark.
[4] R. Nasiri and A. Radan, “Pole-placement control strategy for 4-leg
voltage-source inverters,” in Proc. IEEE Power Electronic and Drive
V. C ONCLUSION Systems and Technologies Conf. PEDSTC, Feb. 2010, pp. 74–79, Tehran,
Iran.
A finite control set model predictive voltage control strategy [5] M. Aissani and K. Aliouane, “Three-dimensional space vector modu-
lation for four-leg voltage-source converter used as an active compen-
for a standalone three-phase four-leg inverter with a prediction sator,” in Proc. IEEE Int. Symposium on Power Electronics, Electrical
horizon of two samples has been proposed to improve the Drives, Automation and Motion, SPEEDAM, Jun. 2010, pp. 1416–1421,
output voltage quality from the one-step prediction horizon. Pisa, Italy.
[6] P. Cortes, M. Kazmierkowski, R. Kennel, D. Quevedo, and J. Rodriguez,
This algorithm does not need internal current controllers and “Predictive control in power electronics and drives,” IEEE Trans. on
modulation stages. Industrial Electronics, vol. 55, no. 12, pp. 4312–4324, Dec. 2008.
[7] S. Kouro, P. Cortes, R. Vargas, U. Ammann, and J. Rodriguez, “Model
The proposed predictive control strategy is much simpler predictive control-A simple and powerful method to control power
and more intuitive than the classical methods previously pro- converters,” IEEE Trans. on Industrial Electronics, vol. 56, no. 6, pp.
posed. The proposed algorithm reduces the number of switch- 1826–1838, Jun. 2009.
[8] J. Rodriguez, J. Pontt, C. A. Silva, P. Correa, P. Lezana, P. Cortes, and
ing state calculations during each sampling time and thus U. Ammann, “Predictive current control of a voltage source inverter,”
offers a lower converter switching frequency. The algorithm IEEE Trans. on Industrial Electronics, vol. 54, no. 1, pp. 495–503, Feb.
tests each of the 16 possible switching states and then selects 2007.
[9] P. Correa, J. Rodriguez, I. Lizama, and D. Andler, “A predictive
a state that minimizes the cost function. control scheme for current-source rectifiers,” IEEE Trans. on Industrial
The proposed controller delivers high quality power to the Electronics, vol. 56, no. 5, pp. 1813–1815, May 2009.
standalone hybrid power system with stiff and symmetrical [10] T. Geyer, “Generalized model predictive direct torque control: Long
prediction horizons and minimization of switching losses,” in Proc.
three-phase sinusoidal voltages. It also offers lower values of Chinese Control Conf. CCC, Dec. 2009, pp. 6799–6804, Shanghai,
voltage error in tracking reference and less %THD for out- China.
put voltages for balanced, unbalanced and nonlinear loading [11] V. Yaramasu, J. Rodriguez, B. Wu, M. Rivera, A. Wilson, and C. Rojas,
“A simple and effective solution for superior performance in two-level
conditions. four-leg voltage source inverters: Predictive voltage control,” in Proc.
The proposed control can compensate for the effect of IEEE Int. Symposium on Industrial Electronics, ISIE, Jul. 2010, pp.
uncertainties in the load and DC-link voltage and, conse- 3127–3132, Bari, Italy.
[12] P. Cortes, J. Rodriguez, S. Vazquez, and L. Franquelo, “Predictive con-
quently, the load voltage waveform remains balanced. This trol of a three-phase UPS inverter using two steps prediction horizon,”
compensation has been achieved without any detriment to the in Proc. IEEE Int. Conf. on Industrial Technology, ICIT, Mar. 2010, pp.
transient and steady state operation. 1283–1288, Viña del Mar, Chile.

S-ar putea să vă placă și