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

net/publication/335478856

Affine discretization methods for the digital resonant control of uninterruptible

power supplies

Article in Journal of the Franklin Institute · August 2019

DOI: 10.1016/j.jfranklin.2019.08.037

CITATIONS READS
3 480

5 authors, including:

Jeferson Vieira Flores Charles Lorenzini

Universidade Federal do Rio Grande do Sul Universidade Federal do Rio Grande do Sul
97 PUBLICATIONS 533 CITATIONS 12 PUBLICATIONS 155 CITATIONS

SEE PROFILE SEE PROFILE

Luís Fernando Pereira Aurelio Salton

Universidade Federal do Rio Grande do Sul Universidade Federal do Rio Grande do Sul
115 PUBLICATIONS 1,677 CITATIONS 65 PUBLICATIONS 271 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Aurelio Salton on 24 August 2020.

The user has requested enhancement of the downloaded file.

 Affine discretization methods for the digital resonant
control of uninterruptible power supplies

Guilherme Keiela , Jeferson Vieira Floresb,∗, Charles Lorenzinib , Luı́s F. A.

Pereirab , Aurélio T. Saltonb
a Instituto Federal Rio Grande do Sul - Campus Farroupilha, RS, Brazil
b School of Engineering, Universidade Federal do Rio Grande do Sul, RS, Brazil

Abstract

This paper addresses the discrete-time design and performance evaluation of

finite-gain multiple resonant controllers for uninterruptible power supplies -
UPS. Two discretization methods that preserve the affinity with respect to
time-varying parameters are considered to obtain the discrete-time UPS uncer-
tain model. Based on a state-feedback formulation for the proposed controller, a
systematic approach for the robust tunning of controller gains by the solution of
a convex optimization problem subject to linear matrix inequality constraints is
derived. Experimental results in a 3.5 kVA inverter are obtained to compare the
effects of distinct sampling frequencies and plant discretization methods with
respect to the IEC 62040-3 performance parameters.
Keywords: Finite-gain resonant controller, uninterruptible power supplies,
harmonic rejection, discretization methods, robust control, linear matrix
inequality.

1. Introduction

Uninterruptible power supplies (UPS) are electronic power devices designed

to provide energy with continuity and quality to critical loads. These systems
are essentially composed by three stages: an AC-DC converter at the input, a

∗ Corresponding author.
Email address: [email protected] (Jeferson Vieira Flores)

Preprint submitted to Journal of The Franklin Institute March 22, 2019

5 battery bank and a DC-AC converter at the output stage, which is responsible
for the supplied voltage. UPS control aims to regulate the inverter output
voltage in order to ensure closed-loop stability and performance against load
variations and disturbances. In this context, usual performance parameters
are individual harmonic distortion (IHD), total harmonic distortion (THD) and
10 recovering time, with the admissible values imposed by standards such as the
IEC 62040-3 [1].
In the last decades, several strategies have been developed to the pulse width
modulation (PWM) control of DC-DC and DC-AC converters, e.g. proportional-
integral-derivative (PID) controller [2] or nonlinear control, such as sliding-mode
15 [3, 4] and model predictive control [5]. More recently, authors considered ac-
tive disturbance rejection control (ADRC) and type-2 fuzzy logic techniques
[6, 7] to improve the overall system performance under time-varying parame-
ter/uncertainties. When the aim is to track or reject periodic exogenous signals,
a fundamental result in Control Theory is the internal model principle (IMP) [8],
20 which provides sufficient conditions to ensure periodic reference tracking or dis-
turbance rejection with zero steady-state error. For instance, when dealing with
a sinusoidal signal, the IMP results in controllers with infinite gain at the signal
fundamental frequency [9], usually referred as resonant controllers [10]. When
dealing with DC-AC converters, researches have expanded this controller to a
25 multiple resonant structure [11, 12] in order to reject periodic disturbances at
harmonic frequencies resulting from nonlinear loads. Depending on the number
of compensated harmonics, this approach may result in a high order controller
with a high number of tunning parameters [13]. A systematic way to design
these parameters is to employ robust control techniques, where the controller
30 design is carried out by the solution of an optimization problem subject to lin-
ear matrix inequality (LMI) conditions [14]. Both stability and performance
constraints can be written directly in the form of LMIs [15] with the associated
optimization problem solved efficiently by numerical packages.
With the rise of digital control, designers must deal with both continuous-
35 time (plant) and discrete-time (controller) signals in the same control loop. In

2
 this context, literature offers two main approaches to digital controller design
known as design by emulation (DE) and discrete-time design (DD) [16]. The
main idea behind DE based techniques is to design a continuous-time controller
which satisfy a prescribed performance and then to converted it into a digital
40 controller for the practical implementation. Although it can benefit from a wide
range of continuous-time design methods, the closed-loop stability/performance
will depend on the controller discretization method employed and the adequate
choice of sampling-time. For instance, an extensive analysis of discretization
effects in the performance of resonant controllers is presented in [17], proving
45 that accuracy of the resonant peak is very sensitive to the method and sampling
rate employed, specially for high order harmonics. On the other hand, DD
firstly obtains a discrete-time model of the plant to directly design a digital
controller from it, which can guarantee prescribed properties such as phase
margin and performance if an appropriate discretization method is used [18]. In
50 [19], the authors presented the state-space formulation of the discretized system
for several discretization methods in the context of linear-parameter-varying
systems and provided guidelines to choose an adequate sampling-time such that
the approximation error is minimized and stability is preserved.
The UPS modeling proposed in [13] assumes the load at the inverter out-
55 put as an unknown parameter, yielding an uncertain continuous-time system.
The use of DD for this class of systems is not straightforward since discretiza-
tion methods such as zero-order hold (ZOH) and Tustin do not preserve an
affine dependence with the uncertain parameter, compromising the equivalence
between the polytopic description of the continuous and discrete-time system
60 [20]. This assumption is paramount when considering robust controller design
based on LMI constraints. Through the power series approximation of the plant
matrices discretized by ZOH, it is shown in [21] that this equivalence which
equivalence? holds for a sufficiently small sampling time, which is, neverthe-
less, an approximated result. When it comes to discrete-time design of resonant
65 controllers to single and three phase UPS, we can point out [22, 23, 24]. In
[22], assuming the UPS time-invariant nominal model and a zero-order hold

3
 discretization, a suboptimal H∞ multiple resonant controller is derived, while
in [23] a two-step procedure based on frequency response analysis is employed
to tune proportional-resonant controllers. Authors in [24] compared the effects
70 of two plant discretization methods in the design of digital infinite-gain resonant
controllers applied to UPS, however no experimental results were provided.
This work addresses the DD of finite-gain multiple resonant controllers con-
sidering the Forward Euler and Adams-Bashforth discretization methods, which
maintain the affinity with respect to the time-varying uncertainty. A novel state-
75 space formulation of digital finite-gain multiple resonant controllers is proposed
and the controller design problem is converted to the synthesis of a discretized
plant and controller state-feedback. In particular, the proposed control struc-
ture explicitly considers the additional states derived from the Adams-Bashforth
discretization method as degrees of freedom available for feedback. Then, a sys-
80 tematic procedure is derived to determine the gains associated to state-feedback
by the solution of a convex optimization problem under LMI constraints that
guarantee closed-loop stability and performance. Simulation and experimen-
tal results in a 3.5kVA commercial UPS are considered to evaluate the effects
of continuous-time discretization methods under the IEC 62040-3 performance
85 requirements. As a benchmark, all results are compared to the DE proposed
in [13], where the Tustin pre-warping approximation is considered to digitally
implement a multiple resonant controller designed in continuous-time.
Notation: N is the set of natural numbers, R is the set of real numbers,
R+ is the set of non-negative real numbers, C is the set of complex numbers,
90 Rn denotes the n-dimensional Euclidean space, Rn×m is the set of n × m real
matrices, 0 is the null matrix of appropriate dimension, I is the 2×2 identity
matrix, k·k is the Euclidean vector norm. For a real matrix Q, Q0 denotes its
transpose. Moreover Q > 0 (Q < 0) means Q is symmetric and positive definite
(negative definite).

4
95 2. UPS Model

In a UPS output stage, the supplied voltage is regulated by a DC-AC in-

verter associated to an output filter. In this work, this stage is composed of a
PWM controlled single-phase half-bridge inverter in series with an LC filter, as
illustrated in Fig. 1, where RLf represents the inductor internal resistance, and
100 Lf and Cf the filter inductance and capacitance, respectively. Effects of both
linear and nonlinear loads on the inverter output are represented by the parallel
connection of the admittance Yo and a periodic disturbance current source id .

S1 LC filter Load
C1
R Lf Lf
Vcc u
S2 iL f
+ 1
id
Cf vout Yo
C2
−

Figure 1: UPS output stage: half-bridge inverter with LC filter.

If the switching frequency is sufficiently high, an average modeling technique

can be applied such that the UPS dynamics is essentially determined by its
output filter. Hence, the following state-space average model can be considered
[13]: 
ẋ(t) = Ac (Yo )x(t) + Bc u(t) + Ec id (t)

(1)
y(t) = Cc x(t)


where,    
R KPWM
− Lf − L1f
Ac (Yo ) =  Lf , Bc =  Lf  ,
1
Cf − YCo (t)
f
0 (2)
h i0 h i
Ec = 0 − C1f , Cc = 0 1 .
In this case, x(t) = [iLf (t) vout (t)]0 is the state vector, where iLf denotes the
inductor current and vout the capacitor voltage, which is also assumed to be the

5
105 plant output y(t). The effects of the half-bridge inverter are now described by a
static gain KPWM multiplying the control signal u(t). Since the load connected
to the UPS output is usually unknown and time-varying, we considered Yo (t)
as an uncertain parameter restricted to the interval Ymin ≤ Yo (t) ≤ Ymax . Note
that in this formulation Ac (Yo ) is an affine matrix function of Yo (t).

110 3. Discrete-time Plant Model

Model-based DD assumes the existence of a discrete-time model – usually

obtained by discretization or system identification methods – describing the
continuous-time plant as accurately as possible. In the UPS context, the uncer-
tain model (1)-(2) is well established in the literature and can be used to obtain
an equivalent discrete-time state-space model. Hence, considering a sampling-
period T and assuming x(k) = x(t)|t=kT , k ∈ N, the aim of this section is to
determine a discrete-time approximation of system (1) in the form

xd (k + 1) = A(Yo )xd (k) + Bu(k) + Eid (k)
(3)
y(k) = Cxd (k)


where the state xd (k) ∈ Rn and matrices A(Yo ), B, E and C depend on the
discretization method employed. The equivalence between (1) and (3) is given
by the solution of state equation (1) at t = (k + 1)T , which yields [25]
(k+1)T
Z
x((k + 1)T ) = x(kT ) + f (x, u, id , Yo )(τ )dτ (4)
kT

where f (x, u, id , Yo )(τ ) = Ac (Yo )x(τ ) + Bc u(τ ) + Ec id (τ ). Discretization meth-

ods are derived by using different approximations of the integral term in (4),
which represents the state evolution between two consecutive samples. Two of
these methods are detailed in the sequel.

115 3.1. Forward Euler

Euler’s forward method [19] uses a rectangular approximation of the integral
in (4), which gives

x((k + 1)T ) 'x(kT ) + T Ac (Yo )x(kT ) + T Bc u(kT ) + T Ec id (kT ). (5)

6
 Considering xd (k) = x(k), then it follows that matrices A(Yo ), B, E and C are

A(Yo ) = I + T Ac (Yo ), B = T Bc ,
(6)
E = T Ec , C = Cc .
Based on the reasoning in [21], this method is equivalent to the ZOH dis-
cretization for sufficient small sampling periods.

3.2. Adams-Bashforth
Multi-step methods are an alternative to approximate the solution of (4)
by the use of information in the previous samples, leading to a smaller local
truncation error when compared to one step methods. The Adams-Bashforth
(AB) three-step method [19] approximates the solution (4) by
T
x((k + 1)T ) ' x(kT ) + (5f |(k−2) − 16f |(k−1) + 23f |(k) ) (7)
12
where f |(k) := f (x, u, id , Yo )(kT ). To represent the discretized system in the
form (3), it is considered the augmented state

xd (k) = [x(k)0 f |0(k−1) f |0(k−2) ]0

with associated matrices

   
23T −16T 5T 23T
I+ 12
Ac (Yo ) 12
I 12
I 12
Bc
   
A(Yo ) =  0  , B =  Bc ,
   
Ac (Yo ) 0
(8)
   
0 I 0 0
h i0 h i
E= 23T
12
E0c E0c 0 , C = Cc 0 0 .

Note that the multi-step characteristic of the AB method results in a discrete-

120 time state equation with augmented order when compared to the continuous-
time one. In the UPS case, two states are added for each step considered, i.e.
xd (k) is a vector with six elements.

Remark 1. In both Euler and AB methods, the affine dependence on the un-
certain parameter is preserved, which is not verified for other methods such as
125 ZOH and the trapezoidal approximation (Tustin). Although ZOH yields an ex-
act equivalence between x(t) and x(k) at the sampling instants, A(Yo ) has an

7
 exponential dependence on Yo , which is also propagated to matrices B(Yo ) and
E(Yo ). The same is verified for the Tustin method, where all discretized ma-
trices depend on the inverse of (I − A(Y0 )T /2) [19]. In addition, this method
130 (which one? or both of them? Strange, I thought they did) does not preserve the
original states (current and voltage), compromising in this way the application
of state-feedback controllers as considered in this work.

4. Controller Design

To start this section, a brief explanation on the discrete-time resonant con-

135 troller with finite-gain is presented and a respective state-space realization is
obtained.

4.1. Resonant Controller and Closed-loop system

Consider a closed-loop system where the plant (3) is controlled by a proportional-

resonant (PR) controller in the form:

u(z) α2 z + α1
Cpr (z) = = k̄e + 2 , (9)
e(z) z − 2z cos(ωd T )e−ξωr T + e−2ξωr T
p
where ωd = ωr 1 − ξ 2 and k̄e , α1 and α2 are free tuning parameters to be
determined. In this case, e(z) denotes the Z-transform of the tracking error
140 e(k) = r(k) − y(k), where r(k) is a given periodic reference signal.
Assuming ξ = 0 and the closed-loop system stable, then the IMP is satis-
fied since the frequency response of Cr (z) shows infinite magnitude (resonance
peak) at ωr , resulting in a zero steady-state error for sinusoidal references and
disturbances with that frequency. However, the digital implementation of infi-
nite gain resonant structures may lead to several problems, mostly related to
high sampling rates, numerical accuracy and high amplitude control signals [26].
In order to overcome some of these issues, a damping factor ξ was introduced
in (9), at the cost of a residual tracking error caused by the finite magnitude
at ωr . This control structure can also be adapted to address the problem of

8
multiple harmonic rejection at frequencies ωri , i = 1, . . . , h in the so-called
proportional-multiple-resonant (PMR) structure:
h
X α2i z + α2i−1
Crh (z) = ke + (10)
i=1
z 2 − 2z cos(ωdi T )e−ξi ωri T + e−2ξi ωri T

where α2i , α2i−1 and ke are free design parameter to be determined.

Considering a state-space framework, one possible realization of (10) is
    
A ... 0  Br1 


 r1


   . 
.. .. .. 

    . 
xr (k + 1) =  . . .  x (k) +  .  e(k)

r

(11)
   
 


 0 . . . A r h
B r h

  


yr (k) = Cr1 . . . Crh xr (k) + Dr e(k)




with xr (k) ∈ R2h being the PMR controller state vector and i = 1, . . . , h, with
h denoting the number of harmonics to be compensated. Controller matrices
are given by
   
1−β0i
β0i ωri β1i
A ri =  , Bri =  ωri ,
−ωri β1i β2i β1 i (12)
h i
Cri = kc2i−1 kc2i , Dr = ke .

 ξω 
ri
β0i = e−ξωri T sin(ωdi T ) + cos(ωdi T ) ,
ωdi
 sin(ω T ) 
di
β1i = e−ξωri T ,
ωdi
 ξω 
ri
β2i = e−ξωri T − sin(ωdi T ) + cos(ωdi T ) .
ωdi
This realization in of particular interest since the controller design is now
cast as a state feedback problem where the tuning parameters are gains kc2i−1 ,
kc2i and ke . Original variables α2i and α2i−1 can be recovered through:

α2i = (kc2i−1 − kc2i−1 β0i + kc2i β1i ωri )/ωri ,

α2i−1 = (kc2i−1 e−2ξωri T − kc2i−1 β2i − kc2i β1i ωri )/ωri .

9
 ke +
Discrete-time
r(k) e(k) Multiple xr (k) u(k) u(t) x(t) x(k) y(k)
+ [kc1 kc2 . . . kc2h−1 kc2h ] + D/A UPS A/D C
− Resonant
Controller
+
[kp1 (kp2 − ke )]

+ Addit.
+ [kp3 . . . kpn ]
States

Figure 2: Closed-loop system.

Now, concatenating the UPS dynamics (3) and the controller in (11), the
augmented system is expressed by:

xa (k + 1) = Aa (Yo )xa (k) + Ba u(k) + Ea q(k)
(13)
y(k) = Ca xa (k)


with the augmented state xa (k) = [xd (k)0 xr (k)0 ]0 ∈ Rn+2h . Notice that xd (k) ∈
Rn represents the discretized plant states with n = 2 for the Euler method or
n = 6 for the AB. The vector of exogenous inputs becomes q(k) = [r(k) id (k)]0
and matrices are given by
 
A(Yo ) 0 0 ... 0
 
 −B C Ar1 0 ... 0
 r1 

.. ..

Aa (Yo ) =  −Br2 C 0 . ,
.
 
A r2
.. ..
 
 .. .. 

 . . . . 0  
−Brh C 0 ... 0 A rh (14)
h i0
Ba = B0 0 0 ... 0 ,
 0
0 B0r1 B0r2 ··· B0rh
Ea =   ,
E0 0 0 ... 0
h i
Ca = C 0 0 ... 0 .

Assume now that the control signal is given by

u(k) = yr (k) + Kp xd (k), (15)

10
 with Kp ∈ R1×n = [kp1 kp2 · · · kpn ], i.e., the PMR output (11) added to a plant
state-feedback. In this case, the closed-loop system assumes the form depicted in
Fig. 2, where the block “additional states” compute the vector [f |0(k−1) f |0(k−2) ]0
necessary to implement the control law with the Adams-Bashforth discretiza-
tion. In the Forward Euler method, these states are set to zero. Moreover, (15)
can be rewritten as an augmented state-feedback plus a feedforward term

u(k) = Kxa (k) + ke r(k), (16)

where
K = [kp1 (kp2 − ke ) . . . kpn kc1 kc2 . . . kc2h−1 kc2h ].

From (13) and (16), the closed-loop system becomes


xa (k + 1) = F(Yo )xa (k) + Gq(k)
(17)
y(k) = Ca xa (k)


with
F(Yo ) =Aa (Yo ) + Ba K,
 0
B0 ke B0r1 B0r2 ··· B0rh
G =  .
E0 0 0 ... 0
Next section presents an LMI framework to obtain the controller gains K
ensuring both the closed-loop robust stability and performance. To this end,
145 the PMR tunning problem is cast in a convex optimization problem subject to
LMI constraints, which can be solved by standard computational packages.

4.2. LMI Formulation

In order to satisfy the constraints imposed by the IEC 62040-3 standard, the
following performance criteria are also considered:
1) PC1: The closed-loop poles pi , i = 1, . . . , n + 2h must lie inside the
circular region described by

RCR = {pi ∈ C : |pi | ≤ ρ, ρ ∈ R+ } (18)

11
150 of the z-plane where 0 < ρ ≤ 1 is a fixed scalar directly related to the desired
transient dynamics [27].
2) PC2: Minimize the quadratic cost function
∞
2
X
J(z) := kz(k)k2 = z(k)0 z(k) (19)
k=0

where z(k) = Cz xa (k) + Dz u(k) is a performance output with matrices Cz

and Dz imposing weights to the energy of the augmented system states and/or
the control signal. This function is usually associated to the linear quadratic
155 regulator problem [28].
In the following theorem are presented conditions to ensure the closed-loop
robust stability and performance:

Theorem 1. For a given 0 < ρ ≤ 1, if there exists a symmetric positive definite

matrix Q ∈ R(n+2h)×(n+2h) , a matrix W ∈ R1×(n+2h) and a scalar γ ∈ R such
that  
−Q ∗ ∗
 
Aa (Ymin )Q + Ba W −Q ∗  < 0, (20)
 
 
Cz Q + D z W 0 −γ
 
−Q ∗ ∗
 
Aa (Ymax )Q + Ba W −Q ∗  < 0, (21)
 
 
Cz Q + Dz W 0 −γ
 
−ρ Q ∗
  < 0, (22)
Aa (Ymin )Q + Ba W −ρ Q
 
−ρ Q ∗
  < 0, (23)
Aa (Ymax )Q + Ba W −ρ Q
then the closed-loop system (17) is asymptotically stable for all Ymin ≤ Yo (t) ≤
Ymax , all eigenvalues of F(Yo ) lie inside the region (18) of the complex plane
2
160 and kz(k)k2 ≤ γxa (0)0 Q−1 xa (0).

The proof of theorem (1) is omitted since conditions (20) - (23) are standard
results in the discrete-time LMI framework [27, 28].

12
 Based on the previous theorem, feedback gain K can be determined by
solving the following optimization problem:

min γ subject to: Q > 0, (20) - (23) (24)

Q,W

where Q, W, γ are decision variables and K = WQ−1 . If the optimization

problem (24) results in an unfeasible solution, then the performance constraints
165 can be relaxed by choosing a value of ρ closed to 1 until a feasible solution is
reached.
h i
Remark 2. From K = k1 k2 · · · kn+2h obtained as the solution of the opti-
mization problem (24) it is not possible to simultaneously determine ke and kp2
since k2 = kp2 − ke . Thus, we can assume without loss of generality that kp2 = 0
170 such that ke = −k2 .

5. Numerical results

In this section, the proposed control strategy is applied to a 3.5 kVA half-
bridge inverter with parameters detailed in Table 1, which are based on a com-
mercial UPS manufactured by CP Eletrônica, a Schneider Electric affiliate. Both
175 linear (purely resistive) and nonlinear loads (composed of a rectifier bridge and
a parallel RC filter) with parameters determined according to IEC 62040-3 are
employed to assess the proposed controller performance. An harmonic analysis
of the nonlinear load indicates that the most significant disturbance components
are in third, fifth, seventh, and ninth harmonics [13]. In addition, the PWM
180 switching frequency is supposed to match the sampling frequency fs , which can
assume the values of 21.6, 10.8 or 5.4 kHz. All numerical results are compared
to the DE methodology proposed in [13], where the Tustin method with pre-
warping compensation is considered to obtain the discrete-time version of the
PMR controller.
185 The region of PC1 is chosen to result in a recovering time under 100 ms,
once in continuous-time systems poles are s = σ ± jωd it implies σ ≤ −40 for
DE or ρ ≤ eσT when mapped for DD. The value of ξ1 is set to zero in all cases

13
 ¯ i = 2, · · · , 4. Aiming
to ensure the nominal reference tracking, while ξi = ξ,
for a fair comparison, the cost function of PC2 has been adjusted to match
190 the maximum control signal obtained with the equivalent DE when supplying
the maximum nonlinear load. Table 2 resumes all parameters considered in the
design stage, where is assumed that Cz ∈ R1×n+2h = ψ[1 . . . 1] and Dz = 1.

Table 1: Inverter and load parameters.

Parameter Symbol Value
Apparent power S 3.5 kVA
Output power factor PF 0.7
Inverter gain KPWM 1
DC bus voltage Vcc 520 V
Output RMS voltage Vo 127 V
Frequency of the reference f 60 Hz
Filter capacitance Cf 300 µF
Filter inductance Lf 1 mH
Inductor resistance RLf 15 mΩ
Maximum load admittance Ymax 0.1519 S
Minimum load admittance Ymin 0.0001 S
Linear load resistance Rl 0.1519−1 Ω
Nonlinear load resistance Rnl 10.3924 Ω
Nonlinear load capacitance Cnl 12028 µF

5.1. Simulation results

Simulation results presented in the sequel are based on a co-simulation setup,

195 where the UPS output stage depicted in Fig. 1 is implemented in PSIM and the
control signal u is computed in Matlab/Simulink. The idea here is to evaluate
the steady-state performance in terms of THD under a nonlinear load with pa-
rameters described in Table 2 and for the frequencies considered in [13] (22kHz),
[22, 23] (10kHz) and 5.4kHz.

14
 Table 2: Design parameters

h=1 h=2 h=3 h=4

fs = 21.6 kHz DE DD Euler DD AB DE DD Euler DD AB DE DD Euler DD AB DE DD Euler DD AB
σ/ρ -100 0.9944 0.9967 -100 0.9954 0.9954 -100 0.9954 0.9954 -50 0.9977 0.9977
ξ¯ 0 0 0 0 0 0 0.01 0.01 0.01 0.004 0.004 0.004
ψ 5.00 5.00 5.00 5.00 5.30 4.00 5.00 5.50 4.80 1.80 1.80 1.80
fs = 10.8 kHz DE DD Euler DD AB DE DD Euler DD AB DE DD Euler DD AB DE DD Euler DD AB
σ/ρ -100 0.9908 0.9908 -100 0.9908 0.9908 -50 0.9954 0.9954 -40 0.9963 0.9963
ξ¯ 0 0 0 0 0 0 0.01 0.01 0.01 0.004 0.004 0.004
ψ 5.00 5.40 4.70 5.00 5.48 4.65 4.00 4.20 4.05 1.60 1.60 1.75
fs = 5.4 kHz DE DD Euler DD AB DE DD Euler DD AB DE DD Euler DD AB DE DD Euler DD AB
σ/ρ -100 0.9672 0.9834 -100 0.9908 0.9654 -80 0.9853 0.9853 -50 0.9907 0.9907
ξ¯ 0 0 0 0 0 0 0.01 0.01 0.01 0.002 0.002 0.002
ψ 0.80 1.60 2.40 0.10 0.40 0.40 0.15 0.10 0.10 0.01 0.01 0.01

200 Figure 3 summarizes the simulation results obtained for each design method.
One can see that no significant difference is noticed when the sampling frequency
is reduced from 21.6 to 10.8kHz, with a maximum THD variation under 0.2%
in the worst scenario (AB, h=1). However, the THD is increased significantly
when this frequency is further reduced to 5.4kHz, ranging from around 1.5%
205 (AB, h=4) to 8% (DE, h=1). Despite that variation, the THD limit of 8%
prescribed in the IEC62040-3 is satisfied for h = 3 in all methods. We also point
out that methods based on DD are less susceptible to frequency variations when
compared to DE.

5.2. Experimental Results

210 All tests considered the experimental setup depicted in Fig. 4 and schemat-
ically illustrated in Fig. 5, also provided by CP Eletrônica. More specifically
the inverter switching module is composed of a Semikron SKD31F/08 three
phase rectifier bridge and a Semikron SKM400GB128D IGBT module with a
SKHI22AR driver. The output voltage and inductor current measurements
215 are obtained with an insulated transformer and a Hall effect sensor, respec-
tively. Data acquisition and control are performed by means of a dSPACE
DS1104 board with associated conditioning circuits. In addition, a Tektronix
TDS2024C oscilloscope with a Tektronix A622 AC/DC current probe and a

15
 18 21.6 kHz 18 21.6 kHz
10.8 kHz 10.8 kHz
16 5.4 kHz 16 5.4 kHz
14 14

12 12
THD (%)

THD (%)
10 10

8 8

6 6

4 4

2 2

0 0
0 1 2 3 4 0 1 2 3 4
Number of resonant modes (h) Number of resonant modes (h)

(a) DE (b) DD Euler

18 21.6 kHz
10.8 kHz
16 5.4 kHz
14

12
THD (%)

10

0
0 1 2 3 4
Number of resonant modes (h)

(c) DD AB

Figure 3: Simulation results - THD.

Fluke 43B Power Quality Analyzer are employed to record signal waveforms
220 and perform the harmonic analysis. For the UPS performance evaluation two
scenarios based on IEC 62040-3 have been considered: one to evaluate steady-
state performance by checking the harmonic distortion and another to verify
the closed-loop transient response. Based on the simulation results presented in
the previous section, only the frequencies of 10.8 and 5.4kHz will be considered.
225 In the first scenario, Figures 6 and 7 present a comparison of the output
voltage THD obtained by all controller designs for an increasing number of
compensated harmonics. One can readily notice that there is no significant
difference between the digital implementation methods at 10.8kHz, once their
THDs are similar with marginal differences associated to the experimental setup
230 EU REMOVERIA ESSA PARTE EM VERMELHO e a frase em azul depois
dessa também. In addition, an average increasing of near 1% can be noted

16
 Figure 4: Experimental test bench.

vac (t) Vcc PWM u(t) vout (t)

Rectifier Filter Load
Inverter

u(k)

x(k)
PC dSPACE

Figure 5: Schematic diagram of the experimental setup.

17
 when comparing these results with the simulations in Fig. 3. On the other
hand, at 5.4kHz the difference between DE and DD methods is significant when
only one or two harmonics are compensated. Comparing the DD controllers, AB
235 discretization led to better steady-state results due to the additional states (and
therefore degrees of freedom) used in the state-feedback approach. In Figure 8
is depicted the magnitude of Td (z) = y(z)/id (z), i.e., the closed-loop transfer
function from the disturbance signal id to the output voltage at each harmonic
component (doted lines). For h = 2, there exists a significant difference in the
240 magnitude of the first uncompensated harmonic (5th) between the methods,
resulting in a smaller THD for the DD. However, for h = 3 this difference is
small and occurs at uncompensated harmonics with less significance (7th and
9th), resulting in a similar THD for all methods.

11 DE
10 DD Euler
DD Adams
9
8
7
THD (%)

6
5
4
3
2
1
0
1 2 3
Number of compensated harmonics (h)

Figure 6: THD comparative for fs = 10.8 kHz.

18
 20 DE
19 DD Euler
18
17 DD Adams
16
15
14
13
THD (%)

12
11
10
9
8
7
6
5
4
3
2
1
0
1 2 3
Number of compensated harmonics (h)

Figure 7: THD comparative for fs = 5.4 kHz.

30 30

20 20

10 10
Magnitude (dB)

Magnitude (dB)

0 0

−10 −10

−20 −20

−30 −30
DE DE
DD Euler DD Euler
−40 DD AB −40 DD AB
1 2 3 4 1 2 3 4
10 10 10 10 10 10 10 10
Frequency (rad/s) Frequency (rad/s)

(a) h = 2 (b) h = 3

Figure 8: Magnitude frequency response from disturbance to the output at 5.4 kHz.

Table 3 displays THD and IHD measurements for resonant controllers with
245 h = 3 and h = 4 when fs = 5.4 kHz. Despite an almost perfect disturbance
attenuation at the compensated harmonics, the infinite gain resonant controller
violates the standard IHD limits for uncompensated (7th and 9th) harmonics.

19
 This is explained from Bode’s integral formula [29]: the attenuation of distur-
bances at some frequencies leads to an unavoidable amplification at other regions
250 of the spectrum which, in this case, comprehends the higher order harmonics.
On the other hand, this can be relaxed by the introduction of a damping factor
at the compensated harmonics, at the expense of an increase in the tolerable
THD/IHD.

Table 3: Steady-state comparative of resonant and finite-gain resonant controllers with fs =

5.4 kHz.
h=3 h=4
harmonics IEC ξi = 0 ∀i ¯ i = 2, 3
ξ1 = 0, ξi = ξ, ξi = 0 ∀i ¯ i = 2, 3, 4
ξ1 = 0, ξi = ξ,
THD [%] 8.0 6.9 7.2 4.2 4.0
IHD3 [%] 5.0 0.1 2.4 0.1 1.5
DE

IHD5 [%] 6.0 0.0 5.0 0.0 2.7

IHD7 [%] 5.0 6.3 4.1 0.0 0.4
IHD9 [%] 1.5 2.0 1.5 3.7 1.1
THD [%] 8.0 5.9 7.0 3.7 4.0
IHD3 [%] 5.0 0.1 4.8 0.2 1.8
DD Euler

IHD5 [%] 6.0 0.1 4.7 0.3 2.3

IHD7 [%] 5.0 5.4 0.3 0.2 0.5
IHD9 [%] 1.5 1.1 1.5 2.9 1.0
THD [%] 8.0 5.7 7.2 4.0 3.9
DD Adams

IHD3 [%] 5.0 0.1 3.4 0.3 1.3

IHD5 [%] 6.0 0.1 5.9 0.6 2.5
IHD7 [%] 5.0 5.2 1.5 0.1 0.3
IHD9 [%] 1.5 1.0 1.4 3.4 1.5

UPS transient performance (with h = 4 and fs = 5.4kHz) was evaluated with

255 respect to additive load steps and compared to most restrictive profile defined
in the IEC 62040-3. Fig. 9 presents a comparison of instantaneous percent
deviation obtained for additive steps from: (a) 20% to 100% of linear load and
(b) 25% to 100% of the nonlinear reference load. In terms of recovery time for
sudden load changes, all methods considered showed a relatively similar behavior
260 with a settling time under the 80ms specified in the optimization problem.
Finally, Figs. 10 and 11 show THD/IHD readings from the power quality

20
 30
Undervoltage transient limit DE
DD Euler
20 DD Adams
Nominal value (%)

10

−10

−20

Overvoltage transient limit

−30
−1 0 1 2
10 10 10 10
Transient duration (ms)
(a) Linear load

30
Undervoltage transient limit DE
DD Euler
20 DD Adams
Nominal value (%)

10

−10

−20

Overvoltage transient limit

−30
−1 0 1 2
10 10 10 10
Transient duration (ms)
(b) Nonlinear load

Figure 9: Transient responses for additive load steps.

21
 analyzer when h = 1 and h = 4 (fs = 5.4kHz). As expected, the multiple
resonant strategy clearly leads to a significant attenuation of the output voltage
THD in all cases. In Fig. 12 the output voltage and current waveforms for the
265 case with the smaller THD reading is presented, i.e. AB method and h = 4.

(a) DE (b) DD Euler (c) DD Adams

Figure 10: Output voltage THD for h = 1

(a) DE (b) DD Euler (c) DD Adams

Figure 11: Output voltage THD for h = 4

6. Conclusion

This paper compared the effect of plant discretization methods in the digital
resonant control of UPS systems. In particular, since the plant continuous-time
model is uncertain, only discretization methods that preserve an affine depen-
270 dence with the uncertain parameter between continuous and discrete-time sys-
tems have been considered. Whereas the Euler Forward discretization method is

22
 Figure 12: Output voltage and current waveforms for h = 4 and fs = 5.4kHz.

somewhat standard in control literature, multi-step methods such as the Adams-

Bashforth have not yet been properly explored. Control of UPS poses itself as
an important benchmark due to the time-varying characteristic of the output
275 load and the strict performance limits imposed by the IEC 62040-3 standard.
Towards this end, a novel state-space formulation for the discrete-time finite-
gain resonant controller was proposed and a robust control design method based
on LMI conditions was presented to ensure both closed-loop stability and per-
formance. Experimental results showed that when the sampling rate is reduced,
280 the plant discretization method plays an important role in the steady-state
tracking performance, in some cases reducing the THD in about 8% nao seria
6%?) when compared to DE. Also, the effect of finite-gain controllers in the IHD
was illustrated, leading to a reduction in the amplification of uncompensated
harmonics (the so-called waterbed effect). As future work one can mention the
285 application of multi-step methods to other problems which require affinity with
the time-varying parameter such as gain scheduling or sampled-data control.

23
 Acknowledgment

This study was financed in part by the Coordenação de Aperfeiçoamento

de Pessoal de Nı́vel Superior – Brasil (CAPES) – Finance Code 001, and by
290 CNPq – Brazil, under grants 305886/2015-0 (J. V. Flores), 302917/2015-2 (L.
F. A. Pereira) and 309272/2015-7 (A.T. Salton). Authors also acknowledge
the support of CP Eletrônica (a Schneider Electric affiliate) on the research of
advanced control techniques applied to UPS systems.

References

295 [1] IEC 62040-3, Uninterruptible Power Systems (UPS) - Part 3: Method of
specifying the performance and test requirements, IEC, Switzerland, 2011
(2011).

[2] G. Willmann, D. F. Coutinho, L. F. A. Pereira, F. B. Libano, Multiple-loop

H-infinity control design for uninterruptible power supplies, IEEE Trans-
300 actions on Industrial Electronics 54 (3) (2007) 1591–1602 (June 2007).

[3] S. Jung, Y. Tzou, Discrete sliding-mode control of a pwm inverter for sinu-
soidal output waveform synthesis with optimal sliding curve, IEEE Trans-
actions on Power Electronics 11 (4) (1996) 567–577 (July 1996).

[4] J. M. Olm, D. Biel, E. Fossas, R. Cardoner, Fixed frequency sliding mode-

305 based robust inversion with a full-bridge current dc-link buck-boost, Jour-
nal of the Franklin Institute 351 (1) (2014) 123 – 138 (2014).

[5] L. Cavanini, G. Cimini, G. Ippoliti, Model predictive control for pre-

compensated power converters: Application to current mode control, Jour-
nal of the Franklin Institute 356 (4) (2019) 2015 – 2030 (2019).

310 [6] G. Wu, L. Sun, K. Y. Lee, Disturbance rejection control of a fuel cell power
plant in a grid-connected system, Control Engineering Practice 60 (2017)
183 – 192 (2017).

24
 [7] D. Xu, Y. Dai, C. Yang, X. Yan, Adaptive fuzzy sliding mode command-
filtered backstepping control for islanded pv microgrid with energy storage
315 system, Journal of the Franklin Institute 356 (4) (2019) 1880 – 1898 (2019).

[8] B. A. Francis, W. M. Wonham, The internal model principle for linear mul-
tivariable regulators, Applied Mathematics and Optimization 2 (2) (1975)
170–194 (June 1975).

[9] C. Lorenzini, J. V. Flores, L. F. A. Pereira, L. A. Pereira, Resonant-

320 repetitive controller with phase correction applied to uninterruptible power
supplies, Control Engineering Practice 77 (2018) 118 – 126 (2018).

[10] S. Fukuda, T. Yoda, A novel current-tracking method for active filters based
on a sinusoidal internal model [for pwm invertors], IEEE Transactions on
Industry Applications 37 (3) (2001) 888–895 (May 2001).

325 [11] A. G. Yepes, F. D. Freijedo, . Lopez, J. Doval-Gandoy, Analysis and design

of resonant current controllers for voltage-source converters by means of
nyquist diagrams and sensitivity function, IEEE Transactions on Industrial
Electronics 58 (11) (2011) 5231–5250 (Nov. 2011).

[12] A. Hasanzadeh, C. S. Edrington, B. Maghsoudlou, F. Fleming,

330 H. Mokhtari, Multi-loop linear resonant voltage source inverter controller
design for distorted loads using the linear quadratic regulator method, IET
Power Electronics 5 (6) (2012) 841–851 (July 2012).

[13] L. F. A. Pereira, J. Vieira Flores, G. Bonan, D. Ferreira Coutinho,

J. Gomes da Silva Junior, Multiple resonant controllers for uninterrupt-
335 ible power supplies – a systematic robust control design approach, IEEE
Transactions on Industrial Electronics 61 (3) (2014) 1528–1538 (Mar. 2014).

[14] L. A. Maccari, J. R. Massing, L. Schuch, C. Rech, H. Pinheiro, R. C. L. F.

Oliveira, V. F. Montagner, LMI-based control for grid-connected converters
with LCL filters under uncertain parameters, IEEE Transactions on Power
340 Electronics 29 (7) (2014) 3776–3785 (July 2014).

25
 [15] M. Chilali, P. Gahinet, H∞ design with pole placement constraints: an
LMI approach, Automatic Control, IEEE Transactions on 41 (3) (1996)
358–367 (Mar 1996).

[16] K. Åström, B. Wittenmark, Computer-Controlled Systems: Theory and

345 Design, Third Edition, Dover Books on Electrical Engineering, Dover Pub-
lications, 2013 (2013).

[17] A. G. Yepes, F. D. Freijedo, J. Doval-Gandoy, O. Lopez, J. Malvar,

P. Fernandez-Comesana, Effects of discretization methods on the perfor-
mance of resonant controllers, IEEE Transactions on Power Electronics
350 25 (7) (2010) 1692–1712 (July 2010).

[18] G. F. Franklin, M. L. Workman, D. Powell, Digital Control of Dynamic

Systems, 3rd Edition, Addison-Wesley Longman Publishing, Boston, MA,
USA, 1997 (1997).

[19] R. Toth, P. S. C. Heuberger, P. M. J. V. D. Hof, Discretisation of linear

355 parameter-varying state-space representations, IET Control Theory Appli-
cations 4 (10) (2010) 2082–2096 (Oct. 2010).

[20] S. Boyd, L. Vandenberghe, Convex optimization, Cambridge University

Press, Cambridge, 2004 (2004).

[21] L. A. Maccari, H. Pinheiro, R. C. Oliveira, e Vinı́cius F. Montagner, Robust

360 pole location with experimental validation for three-phase grid-connected
converters, Control Engineering Practice 59 (2017) 16 – 26 (2017).

[22] S. P. Ribas, L. A. Maccari, H. Pinheiro, R. C. de Leao Fontoura de Oliveira,

V. F. Montagner, Design and implementation of a discrete-time H-infinity
controller for uninterruptible power supply systems, IET Power Electronics
365 7 (9) (2014) 2233–2241 (Sept. 2014).

[23] R. E. Carballo, F. Botterón, G. G. Oggier, G. O. Garcı́a, Design approach

of discrete-time resonant controllers for uninterruptible power supply appli-

26
 cations through frequency response analysis, IET Power Electronics 9 (15)
(2016) 2871–2879 (Sept. 2016).

370 [24] G. Keiel, J. V. Flores, L. F. Pereira, A. T. Salton, Discrete-time mul-

tiple resonant controller design for uninterruptible power supplies, IFAC-
PapersOnLine 50 (1) (2017) 6717 – 6722, 20th IFAC World Congress (2017).

[25] P. Apkarian, On the discretization of LMI-synthesized linear parameter-

varying controllers, Automatica 33 (4) (1997) 655–661 (1997).

375 [26] R. Teodorescu, F. Blaabjerg, M. Liserre, P. C. Loh, Proportional-resonant

controllers and filters for grid-connected voltage-source converters, IEE
Proceedings - Electric Power Applications 153 (5) (2006) 750–762 (Septem-
ber 2006).

[27] D. Krokavec, A. Filasova, On pole placement LMI constraints in control

380 design for linear discrete-time systems, in: International Conference on
Process Control (PC), New York: IEEE, Strbske Pleso, 2013, pp. 69–74
(June 2013).

[28] E. Ostertag, Mono- and Multivariable Control and Estimation: Linear,

Quadratic and LMI Methods, 1st Edition, Springer Publishing Company,
385 Incorporated, 2011 (2011).

[29] M. M. Seron, G. C. Goodwin, J. Braslavsky, Fundamental Limitations in

Filtering Control, 1st Edition, Springer-Verlag New York, Secaucus, NJ,
USA, 1997 (1997).

27

View publication stats