Modulated Model Predictive Speed Control for

PMSM Drives
Cristian Garcia and Jose Rodriguez Shafiq Odhano and Pericle Zanchetta S. Alireza Davari
Dept. of Engineering Dept. of Elec. and Electron. Eng. Faculty of Electrical Engineering
Universidad Andres Bello University of Nottingham Shahid Rajaee Teacher Training University
Santiago, Chile Nottingham, UK Tehran, Iran
cristian.garcia@unab.cl shafiq.odhano@nottingham.ac.uk davari@sru.ac.ir
jose.rodriguez@unab.cl pericle.zanchetta@nottingham.ac.uk

Abstract—Model predictive control (MPC) presents important

advantages in the control of the power converter and drives
such as, fast dynamic response and capability to include non-
linear constrains. These have positioned MPC as a powerful and PMSM
realistic control strategy, however, it also has disadvantages such
as variable switching frequency and parameter sensitivity. This
paper applied a modulated model predictive speed control that
guarantees a fix switching frequency and, thanks to disturbance
compensation, robustness to parameters variation. The strategy
is validated and compared to finite set model predictive speed
control through simulation results.
Index Terms—Model Predictive Control, variable speed drives,
dead-beat control.
Fig. 1. Two level voltage source inverter (2L-VSI). (a) Power circuit; (b)
I. I NTRODUCTION Voltage vectors.
Finite set model predictive control (FS-MPC) has
positioned itself as a feasible strategy for many power its switching harmonic spectrum is spread. This entails, among
electronic systems [1]. FS-MPC’s principle consist in other problems, difficulties in filter and thermal design [17]. It
predicting the behavior of the system based on the is for these reasons that different modulated model predictive
mathematical model and selecting in a cost function the strategies have been proposed to achieve a fixed switching
future state of the power converter that achieves the best frequency, [18]–[21].
performance. This is typically measured as reference tracking. This paper proposes a modulated model predictive speed
The capacity to include various control objectives and manage control (MMPSC). The proposed control strategy achieves a
non-lineal systems are attractive advantages, [2], [3]. That good dynamic and steady-state behavior, with a fix switching
is why FS-MPC has been successfully implemented in frequency, without weighting factors and without steady-state
several power converter topologies such as: neutral point speed tracking error.
clamped converters [4], cascade H-bridge converters [5],
matrix converters [6], flying capacitors converters [7], three- II. D RIVE M ODEL
phase two-level inverter [2], multilevel converters [8] and The proposed modulated model predictive control strategy
many others. is implemented for a permanent-magnet synchronous machine
FS-MPC has been widely implemented in electrical drives, (PMSM) fed by a two-level voltage source inverter (2L-VSI).
[9], [10]. Typically, the model predictive control (MPC) is This section presents the mathematical model of the 2L-VSI
used in the electrical sub-system, controlling the mechanical and the PMSM.
subsystem usually with a linear controller [9], [11]. In recent
A. Power Converter
works full predictive speed control has been proposed [12]–
[15]. However, these works consider complex cost functions. The converter generates the voltage to feed the stator
A simple full finite set model predictive speed control (FS- terminals of the machine, as shown in Fig. 1(a). The 2L-
MPSC) has been proposed in [16] for the control of a VSI generates eight voltage vectors, six active vectors and
permanent magnet synchronous machine (PMSM), achieving two zero vectors, as it is shown in Fig. 1(b). At any instant of
successful results. Nevertheless, due to the FS-MPC nature, it time, the voltage vector of the power converter in a stationary
has a spread switching harmonic spectrum. αβ-frame are,
FS-MPC does not guaranteed the same number of 2 h j 2π j 4π i
vsαβ = Vdc · 1 e 3 e 3 · S, (1)
semiconductor commutation in a fixed period. For this reason, 3
 Fig. 2. Scheme of Modulated Model Predictive Speed Control of a PMSM.

where Vdc is the dc-link voltage and S = [Sa Sb Sc ]T are the is to achieve a good tracking of the speed reference.
switching state of the converter. For that, this control loop used the mechanical model
Then, the power converter voltage in a synchronous dq- to determinate an adequate quadrature stator current
frame oriented with the rotor angle θr of the PMSM is, reference. Because the mechanical model requires the
load torque for an accurate estimation of the speed, a
vs = vsαβ · e−jθr . (2) Kalman filter is uses as disturbance observer for torque
B. Permanent Magnet Synchronous Machine estimation.
• Inner control loop: this stage achieves the control of
The model of the PMSM in a synchronous dq-frame
the currents in the electrical subsystem. In this case,
oriented with the rotor position angle θr is the following,
the proposed control strategy uses a cost function to
ẋ = f (x, u), (3) determinate the two active vectors whit minimum current
error. These vectors will be used in the next modulation
where, stage.
−R
Ls isd + ωr isq +
1
Ls vsd
• Modulation control: it receives the two active vectors
 s

  previously calculated and, by solving a system of
 −ωr isd − Rs isq − ψm ωr + 1 vsq equations, determines the duration for these two active
 

Ls Ls Ls
vectors and the zero vector that must be applied to
 
f (·) =  , (4)
achieve the control objective.
 
3 2 Bm
2Jm ψm p isq − Jm ωr
 
 
  The stages mentioned above will be described in more detail
ωr below.

and, A. Outer Control Loop

This outer loop uses a predictive dead-beat control to
x = [isd isq ωr θr ]T , establish a quadrature stator current reference starting from the
u = [vsd vsq ]T . mechanical equation. The following equation is considered:
The parameters of the machine are Rs stator resistor, Ls stator dωm
Jm = Te − TL − Bm ωm , (5)
inductance, ψm the magnitude of the flux generated by the dt
rotor magnet, p number of pole pairs, Jm inertia and Bm the where ωm is the mechanical speed, Te is the electrical torque
friction of the machine. The values of these parameters are of the PMSM and TL is the load torque. Considering that the
shown in the Table I. electrical torque of the PMSM is,
3
III. P ROPOSED C ONTROL S TRATEGY Te = pψm · isq , (6)
2
The proposed control strategy takes the FS-MPSC concept and solving for the speed derivative:
[16], where a full predictive speed control is achieved, and dωm 3 1 Bm
tries to solve the problem of spread switching harmonic = pψm · isq − TL − ωm . (7)
dt 2Jm Jm Jm
spectrum and ripple using a modulated model predictive
control. The control strategy scheme is presented in Fig. 2. The second order derivative of the speed is obtained by
This is composed by three main stages: derivation of equation (7),
• Outer control loop: it corresponds to the control of the d 2 ωm 3 disq 1 dTL Bm dωm
= pψm − − . (8)
mechanical subsystem of the machine. The objective dt2 2Jm dt Jm dt Jm dt
The second order Taylor discretization of the mechanical Eq (vs,i ) = i∗q − ipq (vs,i ), (16)
speed is then:
where i∗d and i∗q are the direct and quadrature current
T2 references, respectively. vs,i is the voltage vector of the 2L-
k+1 k
+ Tsω · ω̇m k + sω · ω̈m k ,
 
ωm = ωm (9)
2 VSI, with i ∈ {0, . . . , 7}. ipd (vs,i ) and ipq (vs,i ) are the direct
where the Tsω is the downsampling period of the outer control and quadrature current predictions based on the load model,
loop. The derivative of the quadrature stator current in (8) is respectively.
discretized with the forward-Euler method:
C. Modulated Model Predictive Control
disq ik+1 k
sq − isq 0
= . (10) The two active vectors vopt , vopt with zero vector v0 are
dt Tsω used to synthesize the optimal voltage actuation for the next
Then, using equations (7), (8) and (10) in equation (9) and sampling period time. By modulating between these three
considering ωm k+1
= ω ∗ and ik+1
sq = i∗sq , where ω ∗ and i∗sq voltage vectors, the current error can be made to average zero
are the speed and quadrature current references respectively, in a single sampling period, [14]. Then, the problem is reduced
and that the load torque is invariant in a sampling time, it is to obtaining the three times for which each of the respective
possible to solve equation (9) in order to obtain a quadrature three voltage vectors must be applied to average zero current
stator current reference: error, τj with j ∈ {0, 1, 2}. For this, the following system of
1 equations must be solved,
i∗sq = − 2 2 k
(2Jm ωm − 2Jm 2 ∗
ω + Bm 2 2 k
Tsω ωm
(Jm KT Tsω ) 
2
(11)
P
τj · Ed,j = 0,
2 k

+ Bm T̂L Tsω p − 2Bm Jm Tsω ωm − 2Jm T̂L Tsω p




 j=0
+ iksq Jm
2
KT Tsω − Bm iksq Jm KT Tsw
2
),




2

P
2
where KT = 3p2Jψmm . τj · Eq,j = 0, (17)
j=0
The reference obtained in (11) is used as input in the inner




control loop. Equation (11) needs an estimation of the load


2

P
torque T̂L to be calculated. For this reason a Kalman filter is



 τj = Ts ,
used as a disturbance observer to estimate the load torque TL . j=0
More details of the KF implementations have been reported where Ed,j and Eq,j are the errors in the d and q axis as
by the authors in [16]. defined in (15) and (16), produced by the zero vector (v0 ),
B. Inner Control Loop the optimum vector (vopt ) and the second optimum vector
0
(vopt ), as obtained with the cost function (14).
The inner control loop corresponds to the electrical
Solving (17), the three times are obtained:
subsystem of the PMSM. Due to the discrete nature of the
control platform, a discrete model of the PMSM, delivered Ed,1 Eq,2 − Ed,2 Eq,1
τ0 = Ts , (18)
in (3)-(4), is presented below using a second order Taylor D
discretization, Ed,2 Eq,0 − Ed,0 Eq,2
τ1 = Ts , (19)
T2  D
xk+1 = xk + Ts · ẋk + s · ẍk ,

(12)
2 Ed,0 Eq,1 − Ed,1 Eq,0
τ2 = Ts , (20)
where the second derivative of the state vector is obtained as, D
where,
ẍ = f˙(x, u), (13)
D =Ed,0 Eq,1 − Ed,1 Eq,0 − Ed,0 Eq,2 + Ed,2 Eq,0
and where Ts is sampling period of the inner control loop. (21)
First, the system’s response is predicted for each of the + Ed,1 Eq,2 − Ed,2 Eq,1 .
voltage vectors in the finite set of vectors produced by the Where sub-indexes 0, 1 and 2 refere to the zero vector (v0 )
power converter, assuming application of each vector for a and to the optimum vector (vopt ) and the second optimum
complete sampling period T s. The current error produced 0
vector (vopt ) found optimizing the value of the cost function
is evaluated by the quadratic cost function (14), in which (14).
the objective is a current reference tracking in dq-frame.
The active vector that minimizes the error (vopt ) and the IV. S IMULATION R ESULTS
0
active vector that produces the second smallest error (vopt ) The simulations of this work were performed using the
0
are identified. It must be noted that vopt and vopt are always software PLECS. The parameters of the permanent-magnet
adjacent vectors. synchronous machine used are presented in Table I. The inner
control loop of the modulated model predictive speed control
g(vs,i ) = Ed2 (vs,i ) + Eq2 (vs,i ), (14)
runs at a sampling time of Ts = 50[µs], while the outer speed
with, loop it is subsampled by a factor of eight, i.e. Tsω = 400[µs].
Ed (vs,i ) = i∗d − ipd (vs,i ), (15) The pulse-width modulation (PWM) has a carrier frequency
 220 220
215 215
210 210
205 205
200 200

35 35
30 30
25 25
20 20
15 15
10 10

10 10
5 5
0 0
-5 -5
-10 -10

40 40
20 20
0 0
-20 -20
-40 -40
0 0.002 0.004 0.006 0.008 0.010 0.012 0 0.002 0.004 0.006 0.008 0.010 0.012

Fig. 3. Steady-state condition: (a)-(e) Motor speed ω and reference speed ω ∗ ; (b)-(f) Quadrature stator current isq and reference quadrature stator current
i∗sq ; (c)-(g) Direct stator current isd and reference direct stator current i∗sd ; (c)-(h) Phase-a stator current isa .

TABLE I
PARAMETERS objectives, however, they have significant ripple around their
current references. This produce a distortion in the phase-a
Parameter Value Unit stator current, as it is shown Fig. 3(d).
Rs 0.369 [Ω] Fig. 4 shows the performance of the current tracking in αβ-
Ls 2.4 [mH]
ψm 0.129 [W b] frame for both methods. The result for the proposed MMPSC
Jm 1.916 · 10−3 [Kg · m2 ] is shown in Fig. 4(b), where its excellent performance is
Bm 4.64 · 10−3 [ N m · rad
s
] evident. Fig. 4(a) presents the behavior of the FS-MPSC, the
p 5
method achieves the a current with average value tracking the
reference,, however, its performance has an evident a large
switching ripple.
of fsw =10 kHz. The proposed method is compared to the The harmonic spectrum of phase-a stator current is
related finite-set model predictive speed control (FS-MPSC) presented in Fig. 5. Fig. 5(b) shows the spectrum of the
as previously proposed in [16]. In this case, the inner FS- MMPSC strategy, where concentrated spectral lines around
MPSC current loop runs at Ts = 17[µs]. of carrier frequency (fsw =10kHz) and its multiples are
The steady-state behavior of the proposed control scheme observable. The spectrum for the FS-MPSC method is
is presented in Fig. 3. The steady-state condition considers a presented in Fig. 5(a). This shows a switching harmonic
nominal reference speed and 90 % of the load torque. The spectrum that is distributed, which is typical of FS-MPC
results of the proposed method are the ones on the right, strategies. The total harmonic distortion (THD) for MMPSC
while FS-MPSC results are the ones on the left. The proposed is only 3.2%, while the THD of the FS-MPSC is 23.1%.
method has an excellent tracking of the reference speed, The dynamic behavior for the proposed control method
without an observable ripple and stationary error, as shown is evaluated in a speed reversal from the nominal speed
in Fig. 3(e). The current control of this method (Fig. 3(f) to negative nominal speed with a constant load torque of
and Fig. 3(g)) also has good performance. The phase-a stator 90% applied all the time. The speed control shows good
current has sinusoidal-waveform without distortion. FS-MPSC reference tracking, as it is shown in Fig. 6(e). Notable is
results, shown on the right, also achieve the speed control that MMPSC achieves the negative reference speed at the
 40

20

-20

-40
-40 -20 0 20 40

40
Fig. 5. Harmonic spectrum of stator current isa of: (a) Finite-set model
predictive speed control; (b) Modulated model predictive speed control.

20
with a total harmonic distortion of the 3.2 %. The most
important result is that the speed dynamic response achieves
0 the same performance previously reported FS-MPC strategies,
but without their well documented disadvantages in terms of
current switching ripple.
-20
ACKNOWLEDGMENT
The authors acknowledge the support provided by the
-40 Chilean National Fund of Scientific and Technological
-40 -20 0 20 40
Development (FONDECYT) under Grant 1170167.
R EFERENCES
Fig. 4. Current behavior in αβ-frame: (a) Finite-set model predictive speed [1] S. Vazquez, J. I. Leon, L. G. Franquelo, J. Rodriguez, H. A. Young,
control; (b) Modulated model predictive speed control. A. Marquez, and P. Zanchetta, “Model predictive control: A review
of its applications in power electronics,” IEEE Industrial Electronics
Magazine, vol. 8, no. 1, pp. 16–31, March 2014.
[2] J. Rodriguez, M. P. Kazmierkowski, J. R. Espinoza, P. Zanchetta,
same time than FS-MPSC (Fig. 6(a)), demonstrating its fast- H. Abu-Rub, H. A. Young, and C. A. Rojas, “State of the Art of
dynamic response. Respect to the current control variables, Finite Control Set Model Predictive Control in Power Electronics,”
the proposed method shows a clear performance advantage IEEE Transactions on Industrial Informatics, vol. 9, no. 2, pp. 1003–
1016, May 2013.
at all the time, with minimum ripple and no low frequency [3] S. Kouro, M. A. Perez, J. Rodriguez, A. M. Llor, and H. A. Young,
oscillations. “Model Predictive Control: MPC’s Role in the Evolution of Power
Electronics,” IEEE Industrial Electronics Magazine, vol. 9, no. 4, pp.
V. C ONCLUSIONS 8–21, dec 2015.
[4] Z. Zhang, C. M. Hackl, and R. Kennel, “Computationally efficient dmpc
This paper presented a modulated model predictive speed for three-level npc back-to-back converters in wind turbine systems with
pmsg,” IEEE Transactions on Power Electronics, vol. 32, no. 10, pp.
control for a permanent-magnet synchronous machine. The 8018–8034, Oct 2017.
method used a finite-set model predictive control principle to [5] R. Baidya, R. P. Aguilera, P. Acuña, S. Vazquez, and H. d. T. Mouton,
identified the optimal voltage vectors to be applied but uses “Multistep model predictive control for cascaded h-bridge inverters:
Formulation and analysis,” IEEE Transactions on Power Electronics,
PWM modulation to achieve zero average current error within vol. 33, no. 1, pp. 876–886, Jan 2018.
one modulation period. [6] C. F. Garcia, M. E. Rivera, J. R. Rodriguez, P. W. Wheeler, and
The control strategy proposed achieves fixed switching R. S. Pena, “Predictive Current Control With Instantaneous Reactive
Power Minimization for a Four-Leg Indirect Matrix Converter,” IEEE
frequency, no steady-state speed error thanks to the use of Transactions on Industrial Electronics, vol. 64, no. 2, pp. 922–929, feb
a load torque observer and fast dynamic response due to the 2017.
predictive model nature. [7] A. Dekka, B. Wu, V. Yaramasu, and N. R. Zargari, “Model predictive
control with common-mode voltage injection for modular multilevel
Simulation results have been presented to validate the converter,” IEEE Transactions on Power Electronics, vol. 32, no. 3, pp.
proposed strategy. The steady-state performance is excellent, 1767–1778, March 2017.
with minimum switching ripple, characteristic of PWM [8] B. Fan, K. Wang, P. Wheeler, C. Gu, and Y. Li, “An optimal full
frequency control strategy for the modular multilevel matrix converter
converter and without low frequency distortion or steady state based on predictive control,” IEEE Transactions on Power Electronics,
error, obtaining a sinusoidal-waveform of the stator current vol. 33, no. 8, pp. 6608–6621, Aug 2018.
 300 300
200 200
100 100
0 0
-100 -100
-200 -200
-300 -300
40 40
20 20
0 0
-20 -20
-40 -40
20 20
10 10
0 0
-10 -10
-20 -20
40 40
20 20
0 0
-20 -20
-40 -40
0 0.01 0.02 0.03 0.04 0.05 0.06 0 0.01 0.02 0.03 0.04 0.05 0.06

Fig. 6. Dynamic response during a step-change in reference speed: (a)-(e) Motor speed ω and reference speed ω ∗ ; (b)-(f) Quadrature stator current isq and
reference quadrature stator current i∗sq ; (c)-(g) Direct stator current isd and reference direct stator current i∗sd ; (c)-(h) Phase-a stator current isa .

[9] F. Wang, X. Mei, J. Rodriguez, and R. Kennel, “Model predictive no. 3, pp. 1059–1067, sep 2016.
control for electrical drive systems-an overview,” CES Transactions on [17] L. Tarisciotti, A. Formentini, A. Gaeta, M. Degano, P. Zanchetta,
Electrical Machines and Systems, vol. 1, no. 3, pp. 219–230, September R. Rabbeni, and M. Pucci, “Model predictive control for shunt active
2017. filters with fixed switching frequency,” IEEE Transactions on Industry
[10] T. Geyer, “Model Predictive Direct Current Control: Formulation of Applications, vol. 53, no. 1, pp. 296–304, Jan 2017.
the Stator Current Bounds and the Concept of the Switching Horizon,” [18] L. Tarisciotti, P. Zanchetta, A. Watson, J. C. Clare, M. Degano, and
IEEE Industry Applications Magazine, vol. 18, no. 2, pp. 47–59, Mar. S. Bifaretti, “Modulated model predictive control for a three-phase
2012. active rectifier,” IEEE Transactions on Industry Applications, vol. 51,
[11] M. Amiri, J. Milimonfared, and D. A. Khaburi, “Predictive torque no. 2, pp. 1610–1620, March 2015.
control implementation for induction motors based on discrete space [19] S. Vazquez, A. Marquez, R. Aguilera, D. Quevedo, J. I. Leon, and L. G.
vector modulation,” IEEE Transactions on Industrial Electronics, Franquelo, “Predictive optimal switching sequence direct power control
vol. 65, no. 9, pp. 6881–6889, Sept 2018. for grid-connected power converters,” IEEE Transactions on Industrial
[12] E. Fuentes and R. M. Kennel, “A finite-set model predictive position Electronics, vol. 62, no. 4, pp. 2010–2020, April 2015.
controller for the permanent magnet synchronous motor,” in 2013 IEEE [20] E. Fuentes, C. A. Silva, and R. M. Kennel, “Mpc implementation
International Symposium on Sensorless Control for Electrical Drives of a quasi-time-optimal speed control for a pmsm drive, with inner
and Predictive Control of Electrical Drives and Power Electronics modulated-fs-mpc torque control,” IEEE Transactions on Industrial
(SLED/PRECEDE). IEEE, oct 2013, pp. 1–7. Electronics, vol. 63, no. 6, pp. 3897–3905, June 2016.
[21] C. F. Garcia, C. A. Silva, J. R. Rodriguez, P. Zanchetta, and
[13] A. Formentini, A. Trentin, M. Marchesoni, P. Zanchetta, and P. Wheeler, S. A. Odhano, “Modulated model predictive control with optimized
“Speed Finite Control Set Model Predictive Control of a PMSM Fed by overmodulation,” IEEE Journal of Emerging and Selected Topics in
Matrix Converter,” IEEE Transactions on Industrial Electronics, vol. 62, Power Electronics, pp. 1–1, 2018.
no. 11, pp. 6786–6796, nov 2015.
[14] E. Fuentes, C. A. Silva, and R. M. Kennel, “Mpc implementation
of a quasi-time-optimal speed control for a pmsm drive, with inner
modulated-fs-mpc torque control,” IEEE Transactions on Industrial
Electronics, vol. 63, no. 6, pp. 3897–3905, June 2016.
[15] A. Darba, F. De Belie, P. D’haese, and J. A. Melkebeek, “Improved
Dynamic Behavior in BLDC Drives Using Model Predictive Speed and
Current Control,” IEEE Transactions on Industrial Electronics, vol. 63,
no. 2, pp. 728–740, feb 2016.
[16] C. Garcia, J. Rodriguez, C. Silva, C. Rojas, P. Zanchetta, and H. Abu-
Rub, “Full Predictive Cascaded Speed and Current Control of an
Induction Machine,” IEEE Transactions on Energy Conversion, vol. 31,