1504 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 53, NO.

5, OCTOBER 2006

Interfacing Renewable Energy Sources to the

Utility Grid Using a Three-Level Inverter
Salvador Alepuz, Member, IEEE, Sergio Busquets-Monge, Student Member, IEEE, Josep Bordonau, Member, IEEE,
Javier Gago, David González, Member, IEEE, and Josep Balcells, Member, IEEE

Abstract—This paper presents a novel approach for the con- losses. Hence, the trend is to migrate from low-voltage to
nection of renewable energy sources to the utility grid. Due medium-voltage power systems, working with increased rated-
to the increasing power capability of the available generation voltage wind generators and series connection of solar panels
systems, a three-level three-phase neutral-point-clamped voltage-
source inverter is selected as the heart of the interfacing system. in photovoltaic power systems.
A multivariable control law is used for the regulator because Multilevel converters, and specifically three-level converters,
of the intrinsic multivariable structure of the system. A current are a good tradeoff solution between performance and cost in
source (playing the role of a generic renewable energy source) high-voltage and high-power systems. The main advantages
is connected to the grid using a three-level inverter in order to of three-level inverters are reduced voltage ratings for the
verify the good performance of the proposed approach. Large-
and small-signal d–q state-space averaged models of the system switches, good harmonic spectrum (making possible the use of
are obtained and used to calculate the multivariable controller smaller and less expensive filters), and good dynamic response
based on the linear quadratic regulator technique. This controller [7]. However, the control complexity increases compared to
simultaneously regulates the dc-link voltage (to operate at the conventional voltage-source inverters (VSI). The topology em-
maximum power point of the renewable energy source), the mains ployed in this paper is the three-level three-phase neutral-
power factor (the power is delivered to the grid at unity power
factor), and the dc-link neutral-point voltage balance. With the point-clamped (NPC) inverter [8]. In this topology, the dc-link
model and regulator presented, a specific switching strategy to neutral-point voltage needs to be regulated. Otherwise, higher
control the dc-link neutral-point voltage is not required. The device voltage stress and output voltage distortion will occur.
proposed controller can be used for any application, since its The objective of this paper is to verify the good performance
nature makes possible the control of any system variable. The of the three-level inverter as an interface between a generic re-
good performance of the presented interfacing solution in both
steady-state and transient operation is verified through simula- newable energy source (wind or photovoltaic power source) and
tion and experimentation using a 1-kW neutral-point-clamped the utility grid using a multivariable linear quadratic regulator
voltage-source-inverter prototype, where a PC-embedded digital (LQR) controller for the system regulation.
signal processor board is used for the controller implementation. A summary of our previous work with this converter and
Index Terms—Distributed power generation, grid interface, controller follows. In [9], a procedure to model three-level con-
multilevel conversion, multivariable control, photovoltaic power verters is proposed, and the model obtained with this procedure
system, three-level inverter, wind power system. is analyzed. In [10] and [11], a multivariable LQR regulator for
the comprehensive control of the three-level inverter with an
I. I NTRODUCTION LC filter and resistive load is proposed. This control approach
is also applied in [12] to simulate a photovoltaic power system
I N RECENT years, there has been an increasing interest
in renewable-energy sources such as wind power systems
or photovoltaic power systems (for grid-connected systems
connected to the utility grid using a three-level inverter. Finally,
the dissertation in [13] shows that the proposed model and
[1]–[5] or stand-alone systems). In parallel, the progress in control approach for the three-level converter can be used for
technology has led to more powerful systems. Wind generator any application, since the multivariable structure allows the
prototypes have reached 4.5 MW in a single turbine [6], and control of any variable.
photovoltaic power systems of 1 MW (peak) and more are Other models and control approaches have been presented
becoming popular. for the three-level converter. The model in [14] lacks a global
The best solution to increase the power in power systems description of the dc-link side and the ac-side dynamics. The
is to step up the voltage in order to limit currents and reduce controller is designed using the small-signal model. The model
in [15] is based on line-to-line switching functions. Information
about the ac neutral-point voltage and dc-link neutral-point
Manuscript received March 30, 2005; revised June 20, 2005. Abstract publ-
ished on the Internet July 14, 2006. Parts of this paper were presented at IEEE voltage is not included within the model. This information can
IECON’02 Seville, Spain, November 5–8, 2002, EPE-PEMC’02, Cavtat and be useful to determine ac-side common-mode voltages that
Dubrovnik, Croatia, September 9–11, 2002, and IEEE PESC’03, Acapulco, cause motor-bearing currents [16]. Controllers are designed
Mexico, June 15–19, 2003. This work was supported by the Ministerio de
Ciencia y Tecnología, Madrid, Spain under Grants DPI2000-0006-P4-02 and using the d–q transformation, and the dc-link unbalance is
TIC2002-03036. compensated by means of the switching strategy [17], [18].
The authors are with the Department of Electronic Engineering, Universitat Some applications can be found in literature considering the
Politècnica de Catalunya, 08028 Barcelona, Spain (e-mail: Sergio.Busquets@
upc.edu). connection of a three-level converter to the grid; for instance, in
Digital Object Identifier 10.1109/TIE.2006.882021 static VAr compensation (SVC) applications [14] or as part of a

0278-0046/$20.00 © 2006 IEEE

ALEPUZ et al.: INTERFACING RENEWABLE ENERGY SOURCES TO THE UTILITY GRID 1505

Fig. 1. Block diagram of a renewable power system for direct connection to

the ac mains.

superconducting magnetic energy storage (SMES) system [15].

A discussion about the application of multilevel converters to
wind power systems can be found in [19]. A photovoltaic power
source is connected to the grid in [20] and [21] using cascaded
full-bridge topologies.
In this paper, a multivariable state-space model [22] for the
complete system (dc link + three-level VSI + ac filter + grid)
is obtained using line switching functions and the d–q transfor-
mation. This model describes the open-loop dynamic behavior
of the complete system. The steady-state solution and the small-
signal model are found from this model, which incorporates Fig. 2. Generic renewable energy source connected to the grid using a three-
all the state variables, ac currents, and dc-link voltages. In this level VSI. (a) Circuit diagram with NPC topology. (b) Equivalent switching
approach, emphasis is put onto the multivariable structure of the model.
system. The regulator can be implemented using any linear or
nonlinear multivariable control law. Hence, this approach does
not use the switching strategy for dc-link regulation. by the inverter. The power management block controls the dc
An optimal linear-adaptive regulator has been chosen for the voltage to maximize the power extracted from the renewable
controller using the LQR control technique [23]. Integral action energy source. For instance, in photovoltaic systems, it is well
has been incorporated within the LQR regulator in order to known that the dc voltage seen by the solar cells is controlled
complement the proportional action of this type of controller using maximum-power-point-tracking (MPPT) algorithms.
[24]. The proposed regulator allows controlling any converter
state variable in small- and large-signal operation. It will be ap-
plied to control the relevant variables related to any renewable B. System Model
energy source connected to the grid using a three-level inverter:
The inverter in Fig. 1 is the key element in the system, since
dc-link voltage, dc-link neutral-point voltage balance, and
this block controls the operating dc voltage at the renewable
ac-mains power factor.
energy source and defines how the energy is delivered to the ac
This paper is organized as follows. Section II describes the
mains. Fig. 2 presents the circuit diagram for the connection of a
system, as well as the proposed state-space model for the sys-
generic renewable energy source to the ac mains. The renewable
tem. Section III describes the calculation of the LQR controller.
energy source and additional components are represented as
Section IV describes the system implementation. Section V
a current source connected to the dc side of the inverter. The
presents the simulation and experimental results, and this paper
inverter is connected to the ac mains through an inductive
is concluded in Section VI.
filter. Inductors in Fig. 2 represent the inductance from both
the output filter and the distribution line. Finally, the network
II. S YSTEM D ESCRIPTION AND M ODELING is assumed to be a voltage source (power sink in this case) per
phase. Line switching functions have been used instead of line-
A. System Block Diagram
to-line switching functions for the switching model in Fig. 2(b).
Fig. 1 represents a block diagram of a renewable energy- A general model for the three-level converter is presented
management system. The renewable energy source (solar cells in [9] and [25]. This model describes the general dynamics of
or wind generator) is directly connected to the ac mains. The the dc-side and the ac-side including the dynamics of dc-link
power flow is always from the renewable source to the ac mains, neutral-point (o) voltage and the ac neutral-point (N) voltage.
and no batteries are required. Hence, the inverter is operating in This model is adapted to the application studied here. The mod-
one quadrant. The control of the dc voltage seen by the renew- eling procedure described in [9], [25], and [26] is applied to the
able energy source and the ac mains power factor is performed system in Fig. 2, yielding the large-signal nonlinear averaged
1506 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 53, NO. 5, OCTOBER 2006

state-space model given in (1). There are four independent state This state-space model is valid for large-signal operation, and
variables in (1): it is nonlinear since the control variables are found inside the
1) the dc-link voltage (νpn = νp − νn ) and the dc-link volt- state matrix. From the large signal model in (1), assuming a
age balance (νo = νp + νn ); symmetric modulation pattern, balanced dc-link voltages and
2) two line currents (inductor currents) in the d–q domain an ac-mains unity power factor, the steady-state expressions in
(iyd , iyq ). These state variables are the transformed com- (2) can be obtained from (1), forcing time derivatives to zero
ponents of the abc-frame inductor currents (ia , ib , ic ). and substituting all variables by the steady-state values
Zero components of ac voltages (νy0 ) and line currents
(iy0 ) are zero, since symmetric abc voltages and balanced Vyd wr · L · Idc · Vyq
Dpd = Dnd = − 2 +V2
ac operation are considered. Vpn Vyd yq

The control variables (dpd , dpq , dnd , dnq ) are the trans- Vyq wr · L · Idc · Vyd
formed (into d–q domain) components of the duty ratios Dpq = Dnq = + 2 +V2
Vpn Vyd yq
(dap , dan , dbp , dbn , dcp , dcn ) of the line switching functions
(Sap , San , Sbp , Sbn , Scp , Scn ) shown in Fig. 2(b). The zero Vpn · Idc · Vyd
Iyd = 2 +V2
components of the duty ratios (dp0 , dn0 ) do not appear in (1) Vyd yq
because they are multiplied by the zero component of the ac
Vpn · Idc · Vyq
current (iy0 = 0, isolated ac neutral point). Iyq = 2 +V2
Vyd
The input variables are the d–q components (νyd , νyq ) of yq

the phase-to-neutral ac mains voltages (νa
N , νb
N , νc
N ) and V0 = 0. (2)

the current idc . The model assumes a relatively high switching
frequency, and no losses are considered The linearization of the large-signal model (1) around an
 dpd −dnd dpd +dnd  
  0 ωr  operating point results in the small-signal model in (3), shown
iyd 2·L 2·L iyd

d  iyq   −ωr dpq −dnq dpq +dnq  at the bottom of the page. Three additional state variables have
0   iyq 
 = −dpd +dnd −dpq +dnq 2·L 2·L
·  been included (I îyq , I ν̂pn , I ν̂0 ) in order to introduce an integral
dt pn 
ν 0 0  νpn
C
−dpd −dnd −dpq −dnq
C control of the state variables îyq , ν̂pn , ν̂0 , as detailed in [11].
ν0 0 0 ν0
C C
   
− L1 0 0 III. I NVERTER C ONTROL U SING THE LQR T ECHNIQUE
νyd
 0 − L1 0 
+ 2 · νyq  (1) The LQR technique determines a matrix [K] so that the
0 0 C idc cost function in (4) is minimized for a linear system whose
0 0 0
dynamics is defined by d/dt[x] = [A] · [x] + [B] · [u] by means
where of applying a control law [u] = −[K] · [x]. The cost function
νpn = νp − ν n ; in (4) specifies the cost of deviations in the state variables, as
ν0 = νp + ν n ; well as the cost of the control effort. This function can therefore
νyd , νyq as d–q components of νa
N , νb
N , νc
N ; potentially take into consideration the relative cost of the dif-
iyd , iyq as d–q components of ia , ib , ic ; ferent components in the system. For example, additional volt-
dpd , dpq as d–q components of dap , dbp , dcp ; age filtering requires more capacitance, and additional current
dnd , dnq as d–q components of dan , dbn , dcn . filtering requires more inductance. According to the relative

   Dpd −Dnd Dpd +Dnd   

îyd 0 ωr 2·L 2·L 0 0 0 îyd
 îyq   Dpq −Dnq Dpq +Dnq  
0   îyq 
   −ωr 0 2·L 2·L 0 0 
 ν̂pn   −Dpd +Dnd −Dpq +Dnq  
ν̂pn 
d    0 0 0 0 0  
 ν̂0  =  C
 −Dpd −Dnd
C
·  ν̂0 
0
−Dpq −Dnq

dt    0 0 0 0   
 C C  I îyq 
 I îyq   0 1 0 0 0 0 0  
 I ν̂pn  
  I ν̂pn 
0 0 1 0 0 0 0
I ν̂0 0 0 0 1 0 0 0 I ν̂0
 Vpn +V0 V0 −Vpn 
0 0  1 
2·L 2·L −L 0 0
 Vpn +V0 V0 −Vpn 
ˆ   0
 0 0 2·L 2·L  − L1 0  
  dpd  2 
 − ICyd Iyd I
− Cyq
Iyq
 ˆ   0 0 C 
ν̂yd
 C C   d   
+  − Iyd I I
− Cyq 
I · nd
+ 0 0 0  ·  ν̂yq  (3)
 − Cyd − Cyq   dˆpq   
 0C 0 0 0   0 0 0 îdc

 0

 dˆnq 
0 0 0

0 0 0
0 0 0 0 0 0 0
ALEPUZ et al.: INTERFACING RENEWABLE ENERGY SOURCES TO THE UTILITY GRID 1507

cost of capacitances and inductances, the weight of the voltage Weight Wd allows guaranteeing that the control variables
deviation and the current deviation can be adjusted. remain within their natural range (duty ratios in abc coordi-
∞ nates — dap , . . . , dcn — must be within 0 and 1).
 The control of the dc-link voltage is assured by the pro-
J= [x]T · [Q] · [x] + [u]T · [R] · [u] · dt. (4)
portional and integral action introduced by weights Wνpn and
0 WIνpn . Similarly, the control of the dc-link neutral-point volt-
The linear small-signal model in (3) is used to calculate the age balance is assured by weights Wνo and WIνo .
controller. The system is controllable [23], [27]–[29], and the To achieve unity power factor, the vector in d–q coordi-
observability does not need to be checked because complete nates related to the line currents, iy = (iyd , iyq ), must be in
state feedback is available. phase with the vector related to the phase-to-neutral ac mains
The LQR controller is calculated using the linear small- voltages, vy = (νyd , νyq ). This is expressed in the following
signal model. However, this regulator is effective for the control equation:
of the large-signal system due to the robustness of the LQR iyd νyd
technique [23]. = . (6)
iyq νyq
In our previous papers [10], [11], [25], [26], the inverter was
connected to a stiff voltage in the dc side, with an LC low pass To guarantee unity power factor, a proportional action to
filter and a resistive load on the ac side. The controlled variables both “d” and “q” components of the current perturbation is
were the dc-link neutral-point voltage balance and the load considered by means of weights Wiyd and Wiyq . It is intuitive
voltage. Here, a different set of variables will be controlled. to set these coefficients to the same value. However, an integral
action cannot be added to both components because the steady-
1) DC-link voltage (νpn ): The operating voltage of the
state values of the current in (2), with respect to which the
renewable energy source has to be controlled in order to
perturbations (small-signal model) are computed, have been
deliver the maximum power to the grid.
calculated assuming no losses. Therefore, the actual current
2) DC-link neutral-point voltage balance (νo ): In the NPC
vector is not expected to match exactly the vector calculated in
topology, it has to be guaranteed that all devices operate
(2). If we tried to force this equality, an instability would occur.
under the same stress, blocking half the dc-link voltage.
Instead, we will proceed as follows.
3) Mains power factor: The goal is to transfer the energy
First, axis “d” is chosen to be in phase with a virtual set
generated at the renewable energy source to the ac mains
of three-phase mains’ voltages, symmetric and balanced with
in the most efficient manner (unity power factor).
respect to phase “a.” This is done by sensing phase “a” of the
The cost function presented in (4) is defined with the expres- mains voltage and detecting the zero crossing. In the particular
sions in (5) case where the actual mains voltages are symmetric and bal-
anced, axis “d” and vector vy will be in phase (then νyq = 0).
Second, an integral action is introduced to iyq by means of
[x̂]T = [ îyd îyq ν̂pn ν̂0 I îyq I ν̂pn I ν̂0 ] WIiyq . This will ensure that the “q” component of the line
currents matches the calculated value. When a set of symmetric
[û]T = [ dˆpd dˆnd dˆpq dˆnq ]
and balanced mains voltages is considered, the value of iyq
 
Wiyd 0 0 0 0 0 0 should be zero, no matter whether we consider losses or not.
 0 Wiyq 0 0 0 0 0  If the voltages are slightly unsymmetrical and/or unbalanced,
 
 0 0 Wνpn 0 0 0 0  the value of iyq will be close to zero.
 
[Q] =  0 0 0 Wν0 0 0 0  The use of the cost function in (4) provides a tradeoff result
 
 0 0 0 0 WIiyq 0 0  for the control of all the selected variables, defining the trajec-
 
0 0 0 0 0 WIνpn 0 tory of the state variables accordingly. With this procedure, the
0 0 0 0 0 0 WIν0 LQR algorithm gives a control matrix [K], which can be seen
  as two different matrices: [KP ] proportional matrix, and [KI]
Wd 0 0 0
integral matrix (integral action).
 0 Wd 0 0 
[R] =  . (5) Different values for the controller weights have been tested
0 0 Wd 0
through simulation using MatLab and Simulink. The set of
0 0 0 Wd
values found to guarantee a fast response and zero steady-state
error are
These expressions are used for the LQR controller calcu-
lation performed using a built-in MatLab function. The LQR Wiyd = Wiyq = 1, Wνpn = Wν0 = 0.01,
control law provides essentially a multivariable proportional
regulator. The integral action that allows canceling steady-state WIiyq = 30 000, WIνpn = WIν0 = 30, and
errors has been included in the controller by adding three new Wd = 100.
state variables to the small-signal model in (3). These new
variables are the integral of the state variables îyq , ν̂pn , and ν̂0 . As described above, the LQR controller is calculated using
A weight for each variable is considered (Wiyd , Wiyq , Wνpn , the small signal model for a steady-state operating point. In this
Wνo , WIiyq , WIνpn , WIνo , Wd ). particular application, the operating point can be defined by the
1508 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 53, NO. 5, OCTOBER 2006

Fig. 3. Block diagram for the closed-loop configuration of the system.

values of dc-link voltage and current. Large-signal operation and the dc-link neutral-point (vNo ) or the common-mode load
control is achieved by using this small-signal calculated LQR voltage in ac-drive applications. In this paper, dp0 and dq0 have
controller to the whole range of operation (taking advantage of been set to the same constant value (0.8).
the LQR robustness). The required time by the DSP to acquire the input sig-
nals, process the abc-to-dq transformation, controller, dq-to-abc
IV. S YSTEM I MPLEMENTATION transformation to generate the new set of duty ratios in abc
coordinates, and deliver these duty ratios is 400 µs. The com-
The scheme in Fig. 3 is used for simulation and experimental putational effort with the proposed control scheme is around
testing purposes. 40 µs, and the rest is used by dSPACE1102 to acquire input
The three-level NPC VSI prototype has been implemented signals and to deliver output signals.
using IGBTs, and the ratings are 200 Vdc and 5 Adc. The pas- Sinusoidal duty ratios (dap , . . . , dcn ) exiting the d–q trans-
sive component values have been chosen to be C = 100 µF and formation block are the inputs of the PWM modulator block,
L = 5 mH. The renewable energy source electrical behavior which outputs the line switching functions (Sap , . . . , Scn ).
has been emulated using a dc power supply configured as a Classical carrier-based PWM modulation has been selected for
dc current source. The ac mains has been connected through the switching strategy. The switching frequency has been set to
an isolation transformer and an autotransformer in order to 9 kHz. This block is implemented using the DSP built-in PWM
regulate the ac mains voltage value. timers, which run at the switching frequency, regardless of the
The feedback block is implemented using Hall effect sen- control-system sampling time. A distributor block generates
sors (LEM LA25 and LEM LV25). Since only four analog then the control signals for the switches (S1 , . . . , S66 ) from the
inputs to the PC-embedded digital signal processor (DSP) board switching functions (Sap , . . . , Scn ). This block is implemented
(dSPACE1102) are available, and nine different variables have using a logical gate structure and a delay system to introduce
to be sensed, these variables had to be multiplexed. Each the blanking time.
variable is sampled for 400 µs out of every 1.6 ms. The last
sampled value is held for the following 1.2 ms until the next
V. S IMULATION AND E XPERIMENTAL R ESULTS
sampling period begins.
The abc-to-dq transformation, controller, dq-to-abc transfor- In the following, a set of simulations and experimental results
mation, and pulsewidth-modulation (PWM) modulator blocks will be presented to evaluate the goodness of the proposed
are performed by the DSP board. The transformations are syn- control approach.
chronized by detecting the zero crossing of voltage νa
N . The Fig. 4 shows the simulated and experimental phase-to-neutral
controller uses the sensed νyd , νyq , and idc , the Vpn command, voltage (νa
N ) and current (ia ) waveforms for phase “a” in
and (2) to calculate the steady-state value of all duty ratios and steady-state operation. It can be noted that the fundamental
state variables. The perturbations of the state variables are then component of voltage and current are in phase. In Fig. 4(b),
computed as the difference between the sensed and the steady- the displacement power factor is cos(ϕ) = 1. The measured
state values. The LQR-integral controller matrix [K] is then total harmonic distortion of the current is THD = 8.19%. The
multiplied by the vector of state-variable perturbations (and resulting power factor is PF = 0.994. The control approach
selected integrals of these values) to obtain the deviations of the presented is therefore guaranteeing a good displacement factor
duty ratios to be applied to the initially calculated steady-state and a low-current distortion. A discussion about the current
values. The zero component duty ratios (dp0 , dn0 ) are required distortion value is included at the end of Section V.
to generate the duty ratios in abc coordinates (dap , . . . , dcn ). A change in the operating point of the system has been tested.
dp0 and dn0 do not appear in the state-space model (1)–(3), and Figs. 5–7 show the evolution of several variables during a step
hence, they are not determined in the previous computational transient in the voltage command νpn from 100 to 80 V. The
process. Then, the values of dp0 and dn0 can be selected step in the voltage command occurs at time = 20 ms. In Fig. 5,
arbitrarily within a margin (duty ratios in abc coordinates must during the first 20 ms (initial steady state), both νp and νn are
be between 0 and 1). This degree of freedom to select the balanced, having a voltage of around 50 V each. During the
values of dp0 and dn0 can be useful to control other additional transient, both dc-link voltages remain fairly balanced because
variables; for instance, the voltage between the ac neutral-point of the controller action. The time required to reach the final
ALEPUZ et al.: INTERFACING RENEWABLE ENERGY SOURCES TO THE UTILITY GRID 1509

Fig. 5. DC-link voltages (νp , νn ) during a step transient in νpn (νpn =

100 V → 80 V, idc = 0.5 A).

Fig. 4. Phase-to-neutral voltage (νa
N ) and current (ia ) for phase “a”
(νpn = 100 V, idc = 2 A). (a) Simulation. (b) Experimental results. Fig. 6. Three-phase currents (ia , ib , ic ) during a step transient in νpn (νpn =
100 V → 80 V, idc = 0.5A).

steady state is long (more than 100 ms), which is due to the
significant output capacitance of the dc power supply employed
to emulate the renewable energy source. A shorter settling time
can be achieved if the controller is designed to admit larger ac
currents. However, ac currents should be limited according to
the semiconductor devices’ ratings and filter inductor satura-
tion. Hence, there is a tradeoff between the speed of the system
response and the maximum ac current values.
Fig. 6 presents the evolution of the three-phase currents.
Initially, the currents (and therefore the output power)
significantly increase to discharge the dc-link capacitors from
νpn = 100 to 80 V. As νpn attains the steady-state value, the
currents decrease to their steady-state value. As mentioned
above, for a given step in the voltage command νpn , if a faster
response is desired, higher current values will result.
Fig. 7 shows the phase-to-neutral voltage and current for
phase “a.” It can be observed that both variables are fairly in Fig. 7. Phase-to-neutral voltage (νa
N ) and current (ia ) for phase “a” during
phase during all the step transient. a step transient in νpn (νpn = 100 V → 80 V, idc = 0.5 A).
1510 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 53, NO. 5, OCTOBER 2006

in the current distortion from the mentioned above THD =

6.87% (unbalanced mains) to THD = 3.31% (balanced mains).
Moreover, the THD value decreases as the power
delivered to the utility grid increases: A THD = 3.39%
was obtained assuming a DSP computational time of 200 µs,
unbalanced mains voltages, and an output power of 500 W; a
THD = 0.93% was obtained for the same previous conditions
but balanced voltages.

VI. C ONCLUSION
An interfacing system for the connection of renewable en-
ergy sources to the utility grid by means of a three-level NPC
Fig. 8. Inverter efficiency versus input power. VSI has been presented. The controller of the system is based
on the multivariable LQR control technique. For the controller
calculation, a comprehensive model for the system is developed
using line switching functions. The good performance of the
system in both steady state and transient operation has been
verified through simulation and experimentation using a 1-kW
prototype. The proposed controller for the inverter has proven
to guarantee a nearly unity mains power factor, a tight regu-
lation of the dc-link voltage, and dc-link neutral-point voltage
balance, even with unbalanced ac mains voltages. Additionally,
a good agreement has been found between the simulations and
the experimental results, verifying that the developed model
properly describes the system performance.

R EFERENCES
[1] Q. Chongming and K. M. Smedley, “Three-phase grid-connected
inverters interface for alternative energy sources with unified constant-
frequency integration control,” in Proc. IEEE-IAS Annu. Meeting, 2001,
pp. 2675–2682.
[2] D. M. Baker, V. G. Agelidis, and C. V. Nayer, “A comparison of tri-level
Fig. 9. Unbalanced three-phase phase-to-neutral ac mains voltages. and bi-level current controlled grid-connected single-phase full-bridge
inverters,” in Proc. IEEE Int. Symp. Ind. Electron., 1997, pp. 463–468.
[3] F. J. ter Reide and J. Wanner, “High-performance inverters for grid-
Fig. 8 presents the inverter efficiency at different input power connected PV applications,” in Proc. IEEE Photovolt. Spec. Conf., 1994,
levels calculated as Pout /Pin [refer to Fig. 2(a)]. The efficiency pp. 913–916.
[4] A. Woyte, R. Belmans, and J. Nijs, “Islanding of grid-connected ac mod-
in all measurements was above 90%. ule inverters,” in Proc. IEEE Photovolt. Spec. Conf., 2000, pp. 1683–1686.
In order to verify the robustness of the proposed controller, [5] C. J. Hatziadoniu, F. E. Chalkiadakis, and V. K. Feiste, “A power con-
a slightly unbalanced ac mains voltages (see Fig. 9), created ditioner for a grid-connected photovoltaic generator based on the 3-level
inverter,” IEEE Trans. Energy Convers., vol. 14, no. 4, pp. 1605–1610,
by means of the autotransformer, have been applied in all Dec. 1999.
experiments. Results already shown in Figs. 4–7 confirm the [6] A. Wobben, “4.5 MW turbine,” in Proc. Eur. Wind Energy Conf. and
good operation of the controller, even under unbalanced ac Exhib., Madrid, Spain, Jun. 16–19, 2003.
[7] J. Rodríguez, J. S. Lai, and F. Z. Peng, “Multilevel inverters: A survey
mains voltages. of topologies, controls, and applications,” IEEE Trans. Ind. Electron.,
A more detailed analysis of the ac current distortion shows vol. 49, no. 4, pp. 724–738, Aug. 2002.
[8] A. Nabae, I. Takahashi, and H. Akagi, “A new neutral-point-clamped
that part of this distortion is due to the relatively long time PWM inverter,” IEEE Trans. Ind. Appl., vol. IA-17, no. 5, pp. 518–523,
required by the DSP used in the experiments to generate Sep./Oct. 1981.
the new set of duty ratios (400 µs) and due to the long time [9] S. Alepuz, J. Bordonau, and J. Peracaula, “Dynamic analysis of three-
level voltage-source inverters applied to power regulation,” in Proc. IEEE
the sensed variables’ values are held (1.2 ms). This has been Power Electron. Spec. Conf., 1999, pp. 721–726.
verified through simulation. The total harmonic distortion of the [10] S. Alepuz, J. Salaet, A. Gilabert, J. Bordonau, and J. Peracaula, “Control
simulated current waveform (ia ) in Fig. 4(a) is THD = 8.87%, of three-level VSLs with a LQR-based gain-scheduling technique applied
to dc-link neutral voltage and power regulation,” in Proc. IEEE Conf. Ind.
which is reasonably close to the experimental result. If the Electron. and Instrum., 2002, pp. 914–919.
simulation is performed reducing the DSP computational time [11] ——, “Optimal regulator with integral action and gain-scheduling for the
from 400 to 200 µs, the resulting current distortion is then comprehensive control of three-level NPC VSI,” in Proc. IEEE Power
Electron. Spec. Conf., 2003, pp. 1420–1425.
reduced to THD = 6.87%. [12] S. Alepuz, A. Gilabert, E. Argüelles, J. Bordonau, and J. Peracaula,
On the other hand, part of the current distortion is due to the “A new approach for the connection of a three-level inverter to the
unbalance in the set of three-phase mains voltages. Assuming power grid for applications in solar energy conversion,” in Proc. IEEE
Conf. Ind. Electron. and Instrum., 2002, pp. 3285–3290.
a DSP computational time of 200 µs and balanced ac mains [13] S. Alepuz, “Contribution to the control of three-level inverters,” Ph.D.
voltages, the simulation results present an additional reduction dissertation, Tech. Univ. Catalonia, Barcelona, Spain, Dec. 2004.
ALEPUZ et al.: INTERFACING RENEWABLE ENERGY SOURCES TO THE UTILITY GRID 1511

[14] G. Cho, G. Jung, N. Choi, and G. H. Cho, “Analysis and controller design Josep Bordonau (S’87–M’89) was born in Sitges,
of static VAr compensator using three-level GTO inverter,” IEEE Trans. Spain. He received the M.Sc. and Ph.D. degrees in
Power Electron., vol. 11, no. 1, pp. 57–65, Jan. 1996. electrical engineering (with honors) from the Univer-
[15] N. Celanovic, D. Lee, D. Borojevic, and F. C. Lee, “Control design of sitat Politècnica de Catalunya, Barcelona, Spain, in
three-level voltage source inverter for SMES power conditioning system,” 1984 and 1990, respectively.
in Proc. IEEE Power Electron. Spec. Conf., 1999, pp. 613–618. He has been Lecturer and Assistant Professor with
[16] H. Zhang, A. von Jouanne, S. Dai, A. K. Wallace, and F. Wang, “Multi- the Department of Electronic Engineering, Universi-
level inverter modulation schemes to eliminate common-mode voltages,” tat Politècnica de Catalunya, where he has been an
IEEE Trans. Ind. Appl., vol. 36, no. 6, pp. 1645–1653, Nov./Dec. 2000. Associate Professor since 1991. He has been active
[17] N. Celanovic and D. Borojevic, “A comprehensive study of neutral-point in more than 25 research projects with international
voltage balancing problem in the three-level neutral-point-clamped volt- companies and institutions. He has authored more
age source inverters,” in Proc. IEEE Appl. Power Electron. Conf. and than 50 journal and conference papers and is the holder of a European patent.
Expo., 1999, pp. 535–541. His fields of interest are multilevel conversion and ac power conversion applied
[18] D. H. Lee and F. C. Lee, “An analysis of mid-point balance for the neutral- to renewable energy systems and energy-management systems.
point clamp three-level VSI,” in Proc. IEEE Power Electron. Spec. Conf.,
1998, pp. 193–199.
[19] J. Bordonau, M. Cosan, D. Boroyevich, H. Mao, and F. C. Lee,
“A state-space model for the comprehensive dynamic analysis of
three-level voltage-source inverters,” in Proc. IEEE Power Electron. Spec.
Conf., 1997, pp. 942–948.
[20] P. Dorato, C. Abdallah, and V. Cerone, Linear-Quadratic Control. An
Introduction. Englewood Cliffs, NJ: Prentice-Hall, 1995.
[21] K. Ogata, Modern Control Engineering, 3rd ed. Englewood Cliffs, NJ: Javier Gago was born in Barcelona, Spain, in 1965.
Prentice-Hall, 1997. He received the Master of Engineering and Doc-
[22] L. H. Hansen, L. Helle, F. Blaabjerg, E. Ritchie, S. Munk-Nielsen, toral degrees from the Universitat Politècnica de
H. Bindner, P. Sørensen, and B. Bak-Jensen, Conceptual Survey of Gen- Catalunya, Barcelona, Spain, in 1991 and 2003,
erators and Power Electronics for Wind Turbines, Dec. 2001, Roskilde, respectively.
Denmark: Risø National Laboratory. [Online]. Available: www.risoe.dk/ Since 1993, he has been an Assistant Profes-
rispubl/VEA/ris-r-1205.htm sor with the Department of Electronic Engineering,
[23] L. Tolbert and F. Peng, “Multilevel converters as a utility interface for Barcelona. His research interests are electromagnetic
renewable energy systems,” in Proc. IEEE Power Eng. Soc. Summer compatibility, HF modeling of analog and digital
Meet., 2000, pp. 1271–1274.
ICs, printed circuit boards (PCBs), modeling, and
[24] H. Ertl, J. Kolar, and F. Zach, “A novel multicell dc–ac converter for
instrumentation.
applications in renewable energy systems,” IEEE Trans. Ind. Electron.,
vol. 49, no. 5, pp. 1048–1057, Oct. 2002.
[25] S. Alepuz, J. Bordonau, and J. Peracaula, “A novel control approach of
three-level VSIs using a LQR-based gain-scheduling technique,” in Proc.
IEEE Power Electron. Spec. Conf., 2000, pp. 743–748.
[26] S. Alepuz, J. Salaet, A. Gilabert, J. Bordonau, and J. Peracaula, “Control
of three-level VSI using LQR-based gain-scheduling technique for the
regulation of the dc-link and the output voltages,” in Proc. Eur. Power
Electron. and Motion Control Conf., 2002, pp. 161–172.
David González (M’96) was born in Barcelona,
[27] W. L. Brogan, Modern Control Theory. Englewood Cliffs, NJ: Prentice-
Spain, in 1968. He received the Master of Engi-
Hall, 1991.
neering and Doctoral degrees from the Universitat
[28] J. Slotine and W. Li, Applied Nonlinear Control. Englewood Cliffs, NJ:
Politècnica de Catalunya, Barcelona, Spain, in 1993
Prentice-Hall, 1991.
[29] K. Aström and B. Wittenmark, Adaptive Control. Reading, MA: and 2001, respectively.
Addison-Wesley, 1989. Since 1995, he has been an Assistant Profes-
sor in the Department of Electronic Engineering,
Barcelona. His research interests are electromagnetic
compatibility in industrial electronics, HF modeling
Salvador Alepuz (M’03) was born in Barcelona, of switched converters, and power conversion and
Spain, in 1967. He received the M.Sc. and Ph.D. power quality.
degrees in electrical and electronic engineering from
the Universitat Politècnica de Catalunya (UPC),
Barcelona, Spain, in 1993 and 2004, respectively.
Since 1994, he has been an Associate Professor
with the Mataró School of Engineering, Universitat
Politècnica de Catalunya, Barcelona. His main re-
search interests include modulation and control of
multilevel converters.
Josep Balcells (M’95) was born in Vila-Rodona,
Spain, in 1949. He received the Master of Engi-
neering and the Ph.D. degrees from the Universitat
Sergio Busquets-Monge (S’99–M’01–S’04) was Politècnica de Catalunya (UPC), Barcelona, Spain,
born in Barcelona, Spain. He received the degree in in 1975 and 1983, respectively.
electrical engineering from the Universitat Politèc- From 1975 to 1986, he was an Associate Profes-
nica de Catalunya, Barcelona, Spain, in 1999, and the sor. From 1978 to 1986, he was the Head of R&D in
M.S. degree in electrical engineering from Virginia Power Electronics with AGUT SA (now a GE Group
Polytechnic Institute and State University, Blacks- Company). Since 1986, he has been Professor of
burg, VA, in 2001. He is currently working toward Electronics Engineering Department at UPC. He has
the Ph.D. degree in electrical engineering at the led several research projects funded by the Spanish
Universitat Politècnica de Catalunya. Ministry of Ciencia y Tecnologia and also by the European Union in the
From 2001 to 2002, he was with Crown Interna- V Frame Program. His research interests are electromagnetic compatibility
tional, Inc. He is currently an Assistant Professor (EMC) in power systems, measurement and filtering of disturbances produced
with the Universitat Politècnica de Catalunya. His research interests include by power converters, and design of EMC-compliant power converters.
multilevel conversion, power-converter modeling, control and automated de- Dr. Balcells has been serving as an Associate Editor of the IEEE
sign, power-factor correction, and EMI suppression techniques. TRANSACTIONS ON INDUSTRIAL ELECTRONICS since 2003.