Electrical Power and Energy Systems 125 (2021) 106453

Contents lists available at ScienceDirect

Electrical Power and Energy Systems

journal homepage: www.elsevier.com/locate/ijepes

Parallel model predictive direct power control of DFIG for wind energy T
conversion
⁎
Ahmed Benzouaoui, Houari Khouidmi , Boubaker Bessedik
Hassiba Benbouali University of Chlef, Electronics Department, Faculty of Technology, P.O Box 78C, Ouled Fares, 0218 Chlef, Algeria

A R T I C LE I N FO A B S T R A C T

Keywords: This paper presents a parallel model predictive power control (PMPPC) method of doubly-fed induction gen-
Variable-speed wind turbines erator (DFIG) for variable-speed wind turbine applications (VSWT). The use of a hysteresis controller and a
Wind energy conversion predefined switching table will inevitably lead to select a non-optimal voltage vector in terms of minimizing
Direct power control power errors, power ripple and THD currents. In this research paper, a new approach of direct power control
Predictive power control
(DPC) of DFIG based model predictive direct power control is developed by minimizing two cost functions to
select the voltage vectors for stator active and reactive power in synchronously manner and consequently the
optimal switching vector state of two-level voltage source inverter (2L-VSI). The proposed PMPPC control
scheme takes full advantage and the power regulation objective of DFIG is confirmed by the simulation results
compared to the DPC and PPC methods.

1. Introduction 1.1. Literature Review

In 1986 I. Takahashi proposed a new quick-response and high-ef- To improve the power regulation of DFIG, different control strate-
ficiency control strategy of an induction machine. This control strategy gies have been presented in the literature. These include vector control
is commonly referred to as direct torque control (DTC) and has been (VC) [19], direct torque control (DTC) [1–3], direct power control
continuously developed and improved by many researchers [1–3]. The (DPC) [4–9], and recently numerous predictive power control (PPC)
purpose of DTC is to select an optimal voltage vector or combination of methods employing predicted stator active and reactive power such as
voltage vectors generated by a VSI in order to keep torque and stator the work proposed by J. Hu et all in [14] and Y. Zhang et all in [15]. In
flux as close as possible to its reference value using hysteresis con- [14,15], the stator active and reactive power is realized through the
trollers and switching table. conventional predictive direct power control, but the disadvantages of
The control strategy of direct power control (DPC) [4–9] is the same these methods are that they depend on very complex formulas and
with DTC that based on hysteresis controllers and switching table to require extensive knowledge of the system to obtain the duty cycle
keep stator active and reactive power as close as possible its reference control, and to adjust the duty ratio of the selected active voltage
value. However, the presence of hysteresis controllers leads to a vari- vectors.
able switching frequency operation. Furthermore, the DPC-SVM [5,6] Recently, a sequential model predictive torque and stator flux
strategies operating at a constant switching frequency can be im- control (SMPTC) was proposed [16,17], which calculates the voltage
plemented by a space-vector modulator instead of the switching table. vectors of the system in a sequential way by using two cost functions for
Predictive direct power control (PDPC) methods are one of the most torque and stator flux control, separately. Although, this SMPTC
important recent achievements in power control of DFIG [10–18]. The method is not necessary needed to design the weighting factor. How-
conventional PDPC method [10–13] uses the power error based cost ever, this method uses a torque cost function to select only two voltage
function and voltage vector selection to replace the hysteresis con- vectors first, and sequentially combines a stator flux cost function to
trollers and the switching table of direct power control. One of the confirm the optimal voltage vector.
major disadvantages of conventional PDPC is that the cost function As an exception, the work presented recently by F. Wang et al. in
need weighting factor design. [18] is the only work that deals with this type of predictive method on
an induction motor for torque and stator flux control, where the authors

⁎
Corresponding author.
E-mail addresses: [email protected] (A. Benzouaoui), [email protected] (H. Khouidmi), [email protected] (B. Bessedik).

https://doi.org/10.1016/j.ijepes.2020.106453
Received 13 March 2020; Received in revised form 12 August 2020; Accepted 13 August 2020
Available online 21 August 2020
0142-0615/ © 2020 Elsevier Ltd. All rights reserved.
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

presented the basics and principles of this new method. P*S Q*S

PMPPC control of DFIG

1.2. Novelty of the Study
S1 S2 S3
Comparing with conventional model predictive direct power control
(MPPC); in which a weighting factor needs to be designed; sequential
model predictive power control (SMPPC) overcomes the drawbacks of
conventional MPPC. The drawback of SMPPC is that only two voltage S1 S2 S3
vectors are exactly selected first for the active power control; which is
not reasonable and exhaustive; and the priority of active power and Vdc a
b DFIG
reactive power is not considered. A sequential consideration of the c
variables instead of synchronously consideration is not an optimal so-
lution.
S1 S2 S3
This newly proposed parallel model predictive power control Wind Turbine
PMPPC method of DFIG overcomes both drawbacks of traditional DPC
and the drawback of conventional PPC. The main objectives of this
study are theoretical evaluation, design, simulation validation and (a)
comparison of the proposed parallel MPPC method with conventional
DPC and PPC control methods for 1.5 MW DFIG.
Im
The rest of the paper is organized as follows: First the WT’s and TSR Sector III Sector II
based MPPT control method are presented in Section 2. Section 3 V3010
V0000 V2110
presents the DFIG system description and power converter model. The
V7111
Direct Power Control of DFIG control design is presented in Section 4.
The Section 5 presents the proposed Parallel Model Predictive Power
Control method. The simulation results for stator active and reactive
Sector I
power are presented in Section 6. Finally, the conclusions of the paper
V4011 Re
are summarized in Section 7.
Sector IV V1100

2. Wind turbine model and MPPT technique

The block diagram of Fig. 1 corresponds to the aerodynamic and

V5001 V6101
mechanical modelling of the wind energy conversion system based
wind turbine. This blocks diagram shows that the generator speed Ωg, Sector V Sector VI
(b)
and therefore of the turbine Ωt, can be controlled by acting either on the
blades pitch angle β, or on the generator electromagnetic torque Tg. The
Fig. 2. (a) 2L-VSI scheme (b) 2L space vector representation.
wind speed Vv is considered a disturbing input to the system.
To determine the speed of the wind turbine, a specific quantity is
used, referred to as the relative speed (tip speed ratio TSR) λ, which is Cp (λ, β ) = (0, 5 − 0, 00167(β − 2))sin ⎡
π (λ + 0, 1) ⎤
the ratio of the linear speed at the end of the blades of the turbine and ⎢
⎣ 10 − 0, 3(β − 2) ⎥
⎦
the wind speed [20]: − 0, 00184(λ − 3)(β − 2) (2)
R·Ωt
λ= Where Cp is aerodynamic power coefficient and β is the blades wedging.
Vv (1)
Knowing the turbine speed, the mechanical torque Tt available on
Where λ is the tip speed ratio (TSR), R is the rotor diameter (blade the slow shaft of the turbine can therefore be expressed by [20]:
radius), Vv and Ωt are the wind and rotor speed (turbine speed), re-
Paer ρ ·S·Vv3
spectively. Tt = Cp·
We use the formula in (2) for the power coefficient as a function of Ωt 2·Ωt (3)
the relative speed λ and the blades wedging β [20].
Where Tt is turbine torque, Paer is aerodynamic power, ρ is air density
and S is the rotor surface (πR2).
Turbine Multiplier Mechanical
The multiplier adapts the rotational speed of the turbine (slow shaft)
Shaft
Eq.2 Ȝ Eq.1 ȍt ȍg (Generator) to the rotational speed of the generator (fast shaft).
Ǻ=0° Eq.5
This multiplier is modeled mathematically by the following two
equations (4) and (5):
Cpmax
Vv Tg =
Tt
Tt Tg G (4)
Vv Eq.3 Eq.4 +
-
Eq.6 ȍg

Ωg
T*g Ωt =
ȍ*t ȍ*g G (5)
Eq.7 Eq.8 + PI
-
ȍg Where Ωg is generator speed; Tg is generator torque, and G is gearbox
closed loop speed control
ratio.
The mechanical shaft generator is described by a simple two-mass
Fig. 1. Block diagram of controlled wind energy conversion system. model, which is given by dynamical equation (6):

2
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

Grid

Vdc

Is
DC Link
V*r
Vs
P*S HP

Wind Turbine Model

ǻP

MPPT Technique
Sa
- Switching Sb

&
Q*S ǻQ HQ Table DFIG
Sc
- ȍg

į
QS Ir
Ir
Ps & Qs
Calculation Is
PS ȥr & ȥs
Estimation Vs

Fig. 3. DPC Control scheme of DFIG.

Table 1 3.1. Power Converter model

. Switching Table of DPC for 2L-VSI Driven DFIG.
Ps Qs Rotor flux sector A two level- voltage source inverter (2L-VSI) is used to feed the rotor
of DFIG. The 2L-VSI has eight possible switching states, which produce
HP ΔPs HQ ΔQs 1 2 3 4 5 6 seven different voltage vectors. The topology of the 2L-VSI and its
voltage vectors are presented in Fig. 2.
+1 ↑ +1 ↑ V5 V6 V1 V2 V3 V4
0 → V7 V0 V7 V0 V7 V0
The switching state of a 2L-VSI can be summarized in a single
−1 ↓ V3 V4 V5 V6 V1 V2 complex switching function given by [15]:
+1 ↑ −1 ↓ V6 V1 V2 V3 V4 V5
2
0 → V0 V7 V0 V7 V0 V7 S= (S1 + aS2 + a2S3)
−1 ↓ V2 V3 V4 V5 V6 V1 3 (9)
2π
Where a = Si = 1,2,3 = {1, 0} and Si = 1,2,3 = {0, 1} means the “ON”
ej 3 ,
“OFF” switching state of the inverter.
J dΩt
⎛ t + GJg ⎞ = Tt − GTg − Kf G. Ωt The inverter voltage vectors Vr is related to the switching state S by:
⎝G ⎠ dt (6)
Vr = SVdc (10)
Where Jt and Jg are the rotor and generator inertia, respectively, and Kf
is viscous friction coefficient.
To extract maximum wind energy at VSWT, the turbine should 3.2. DFIG Model
necessary operate at optimal TSR ≈ λopt where the power coefficient is
equal to its maximum value Cpmax, which confirm the equation (2) for The DFIG model can be described in terms of space vector quantities
β = 0° due to the assumption of fixed pitch wind turbine [20]. using a rotor reference frame as [14–19]:
So if the wind speed changes, the referential rotor and generator
speed are given by: • Voltage equations
dψs
λ opt Vv Vs = Rs Is + + jωr ψs
Ω∗t = dt (11)
R (7)
dψr
Vr = Rr Ir +
Ω∗g dt (12)
Ω∗t =
G (8)
Where:

Vs, Vr: Stator and rotor voltage vectors

3. DFIG and power converter modeling Is, Ir: Stator and rotor current vectors
ψs, ψr: Stator and rotor flux vectors
In this section, models of the power converter and the doubly-fed Rs, Rr: Stator and rotor resistance
induction generator used for the implementation of the parallel model ωr: Rotor angular frequency
predictive control method are introduced.

3
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

Grid

Is
DC Link
Vdc
Vs

Wind Turbine Model

MPPT Technique
P*S
Vkr

&
Q*S Cost Function
DFIG
J(Vkr)
PS(k+1)
ȍg
QS(k+1)

Ir
PS(k)
Ps & Qs
Ps & Qs Calculation
Is
Prediction QS(k) ȥr & ȥs
Estimation Vs

V*r ȥs ȥr

Vdc

Fig. 4. Conventional PPC Control scheme of DFIG.

• Flux equations controllers for the stator active and reactive powers, respectively. The
ψs = Ls Is + Lm Ir block diagram is shown in Fig. 3. The active power hysteresis controller
(13)
is a 3-level comparator with a band amplitude BP, whereas the reactive
ψr = Lr Ir + Lm Is (14) hysteresis controller is a 2-level comparator with a band amplitude BQ.
The meaning of output signals of these two hysteresis controllers are
Where: defined in Table 1.
The location of the estimated rotor flux can be calculated by:
Ls, Lr, Lm: Stator, rotor and magnetizing inductance
• Electromagnetic torque equation δ = arc 2 tan(ψrα ψsβ ) (18)
3 It is divided into six sectors in the αβ plane through the following
Te = pσLm Im{ψr∗ ψs}
2 (15) equation:
Where: π π
(2k − 3) < δ < (2k − 1) , k = {1, 2, 3, 4, 5, 6}
6 6 (19)
p: Number of pole pairs
σ: Leakage coefficient, σ = 1 (Ls Lr − Lm2 ) Combining the output signals of hysteresis controllers with the
*: Complex conjugate operator sector of the rotor flux, the inverter switching state can be selected from

• Stator active and reactive powers equations a predefined switching table (Table 1).
In DPC the stator active and reactive powers can be calculated by
3 [7,8]:
Ps = ωs σLm IM {ψr∗ ψs}
2 (16)
3
3 Ps = ωs σLm |ψr ||ψs| sin δ
Qs = ωs σ [Lr |ψs |2 − Lm Re{ψr∗ ψs}] 2 (20)
2 (17)
3
Where: Qs = ωs σ [Lr |ψs |2 − Lm |ψr ||ψs| cos δ ]
2 (21)

Ps, Qs: Stator active and reactive powers

ωs: Synchronous angular frequency 5. Predictive Power Control (PPC) Principle of DFIG

4. Direct power control principle of DFIG In this section, we present two predictive methods for power control
of DFIG, the first is the conventional PPC method with a weighting
The main idea of conventional DPC [4–9] is to use two hysteresis factor, and the second is the proposed Parallel PPC method without

4
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

Grid

CP constraint Is
P *
S
DC Link
Vdc
Vs
Ps Cost

Wind Turbine Model

MPPT Technique
Function r
VP
Vkr

&
CQ constraint Intersection DFIG
Q *S
Qs Cost VrQ ȍg
Function
Ir
PS(k)
PS(k+1) Ps & Qs
Ps & Qs Calculation Is
Prediction QS(k) ȥr & ȥ s
QS(k+1) Estimation Vs

V*r ȥs ȥr

Vdc

Fig. 5. Proposed PMPPC Control scheme of DFIG.

Table 2
. DPC and PMPPC comparison.
Criterion Traditional DPC Conventional PPC Proposed Parallel MPPC

Inner loop 2 Powers Single cost function 2 Powers cost functions

Hysteresis controllers with a weighting factor without a weighting factor
Switching Table Vectors selection Vectors intersection
Switching frequency Affected by amplitudes Affected by amplitudes Unaffected by constraints
Adjustment parameters 2 parameters HP & HQ weighting factor KQs 2 parameters CP & CQ
Flux Angle Necessary Unnecessary Unnecessary
Calculation demand Lower Medium Higher

weighting factors. The main objective of the PPC [10–18] methods here dPs 3
= ωs σLm [Im{Vr∗ ψs} + ωg Re{ψs ψr∗}]
is to select the optimal rotor voltage vector that previews the stator dt 2 (23)
powers according to its current references. Before introducing the
dQs 3
principle of PPC methods, predictions of the stator powers, at the k + 1 = − ωs σLm [Re{Vr∗ ψs } − ωg Lm Im{ψr∗ ψs }]
dt 2 (24)
sampling step, are required.
After the stator active and reactive power derivative have been
5.1. Stator active and reactive power prediction obtained, predictions of the stator powers are necessary to be calcu-
lated. According to equations (23), (24) and using Euler formula (22),
Based on the powers equations (16) and (17), all possible future the predictions can be derived as:
stator powers values can be predicted using the following Euler forward 3
Ps k + 1 = Ps k + Ts ωs σLm [Im{Vr∗ ψs} + ωg Re{ψs ψr∗}]
formula: 2 (25)
dX X (k + 1) − X (k ) 3
= Qs k + 1 = Qs k − Ts ωs σLm [Re{Vr∗ ψs } − ωg Lm Im{ψr∗ ψs}]
dt Ts (22) 2 (26)

From (16) and (17), we remark that the stator powers varies on the
stator and rotor flux. Since the stator is connected with a constant grid 5.2. Conventional predictive power control of DFIG
frequency, the stator powers depends only on the rotor flux [10–15],
which is controlled by the selected rotor voltage Vr∗. In order to find the similarities and differences between PPC
Neglecting the rotor resistance and according to (16) and (17), the methods, the principles of conventional PPC method is presented first,
stator powers derivative can be expressed as [15]: then the parallel MPPC method is discussed. The PPC methods have the

5
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

CP_min CP CP_max
Parameters initialization:
Vi_7 Ts, CP, CQ, Vdc, Vrk V0 , V1 ,...,V7
Vi_0 Vi_1 Vi_2 Vi_3 Vi_4 Vi_5 Vi_6
0 max
Estimation of stator and rotor flux
JP_V0 JP_V1 JP_V2 JP_V3 JP_V4 JP_V5 JP_V6
Eqns. (13 & 14).
JP_V7
(a)

CQ_min CQ CQ_max Calculation of stator powers and their derivatives

Eqns. (16, 17) and Eqns. (23, 24).
Vi_7

Vj_0 Vj_1 Vj_2 Vj_3 Vj_4 Vj_5 Vj_6

Select first voltage vector. // Vrk Vi 1
0 max

JQ_V0 JQ_V1 JQ_V2 JQ_V3 JQ_V4 JQ_V5 JQ_V6

JQ_V7 Prediction of stator powers. // Eqns. (25 & 26)
(b)

Select next voltage vector

Fig. 6. Selected voltage vectors and sorted values of cost functions applied
different voltage vectors. (a) Active power cost function. (b) Reactive power

Next control period

cost function.

Vkr = Vi+1
Evaluation of cost functions JP and JQ.
Eqns. (28 & 29)
same principle, but each method has its own structure. Fig. 4 presents
the block diagram of conventional predictive power control method for
stator active and reactive power regulation of DFIG. As shown in Fig. 4,
conventional PPC method uses a sequence of single cost function with a
weighting factor for the stator active and reactive power to select the
optimal rotor voltage vector. No
Vkr= V7 ?
The conventional PPC method minimizes a future error between the
predicted and reference active power and a weighted future error be-
tween the predicted and reference reactive power. Its designed as Yes
[10–13]: Sort JP and JQ output in descending values
J(Vrk ) = |Ps∗ − Ps k + 1 (Vrk )| + KQS |Qs∗ − Qs k + 1 (Vrk )|Vrk = {V0, V1,..., V7} (27)

Where KQs denotes the weighting factor and Vrk

= {V0, V1, ...,V7} are the Compare sorted values with constraints:
inverter voltage vectors previously determined by equations (9) and JP < CP , JQ < CQ.
(10).
According to cost function (27), the weighting factor KQs handles
the relation between the stator active and reactive power conditions
and must be fixed. However, the choice of this parameter turns out to Select VrQ Select VrP
be difficult because of the non-existence of relationships linking this
parameter to conventional or advanced indicators in automatic control.
If the same weight were assigned for the active and reactive power, KQs
would correspond to the ratio between the nominal values of active and V*r =Intersection (VrP, VrQ)
reactive power, namely K Qs = λ Pn Qn . However, this coefficient is
usually determined empirically, because the optimal value can be dif-
ferent depending on the defined criteria for optimal operation. Note Select the optimal voltage vector:
that the tuning of the term λ is not necessary for cost function com-
posed of variables with the same nature (same units) [16].
Vrk Vr* .
Finally, seven different values of J (Vrk ) can be obtained according to
seven different active power predictions Ps k + 1 (Vrk ) . The voltage vector,
which produces the smallest value of J (Vrk ) , is selected as the optimal Fig. 7. Flowchart of the proposed PMPPC method of DFIG.
voltage vector. One of the major advantages of conventional PPC is that
several control targets can be included in a cost function, but it also
5.3. Parallel model predictive power control principle of DFIG
brings new problems. The drawbacks of conventional PPC are that the
weighting factor is difficult to be designed and the calculation time is
The block diagram representation of the parallel model predictive
not considered.
power control (PMPPC) method of DFIG for stator active and reactive
Remark 1. It is observed that there are other cost functions to control power regulation is shown in Fig. 5. There are many similarities and
Ps and Qs, except (27) such as: differences between the traditional DPC, conventional PPC and parallel
MPPC control methods. Regarding the inner control, DPC uses two
J(Vrk ) = (Ps∗ − Ps k + 1 (Vrk ))2 + K Qs (Qs∗ − Qs k + 1 (Vrk ))2Vrk = {V0, V1,..., V7} hysteresis controllers with a switching table, whereas PMPPC uses two
cost functions with a switching vectors intersection instead the single
According to the work presented in [21], there are no clear differ-
cost function with a weighting factor used in conventional PPC method
ence in system performance using these cost functions as long as the
to select the voltage vectors. However, DPC needs to calculate the
errors between P*s (and Q*s) and Ps (and Qs) can be minimized.

6
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

Table 3 JP = |Ps∗ − Ps k + 1 (Vrk )|Vrk = {V0, V1,..., V7} (28)

Wind conversion system parameters.
WTs Parameters Value IS-Unit JQ = |Qs∗ − Qs k + 1 (Vrk )|Vrk = {V0, V1,..., V7} (29)

Rated power 1500 kW Where Vrk = {V0, V1, ...,V7} are the inverter voltage vectors.
Rotor diameter R 3 m
Rotor moment of inertia Jt 494 × 104 Kgm2 B. The PMPPC Implementation
Maximal power coefficient 0.510 –
Optimal TSR λopt 4.975 –
Gearbox ratio G 6.8 – Similar to hysteresis controllers with band amplitudes used in
conventional DPC method, PMPPC uses the active power cost function
DFIG Nominal values Value IS-Unit
Rated power 1500 kW
with a constraint CP and the reactive power cost function with a con-
Stator frequency 50 Hz straint CQ (see Fig. 5 & Table 2). The outputs values of JP and JQ are
Nominal stator voltage (Δ/Y) 398/690 V sorted in descending order and compared with the value of constraint
Nominal stator current (Δ/Y) 3.2 kA through evaluation effects of all possible rotor voltage vectors. The
Nominal rotor voltage (Δ/Y) 350/590 V
available voltage vectors for JP and JQ are selected (VrP, VrQ) and the
Nominal rotor current (Δ/Y) 6.2 kA
Rotor speed 1485 Rpm intersection of these vectors gives the one (the optimal Vkr = V*r) which
respect the constraints CP and CQ.
DFIG Constant Value Unit
Poles of pair p 2 In order to select the optimal voltage vector, the voltage vectors for
Stator resistance Rs 0.455 Ω active and reactive power need to be selected and intersected. As shown
Rotor resistance Rr 0.19 Ω in Fig. 6, CP and CQ need to be set properly to guarantee the best three
Stator inductance Ls 0.007 H voltage vectors for active power and reactive power, respectively. The
Rotor inductance Lr 0.213 H
outputs of cost functions are the constrained voltage vectors with green
Mutual inductance Lm 0.034 H
Moment of inertia J 21.9388 Kgm2 color area for active and reactive power, separately (VrP = V(i,0–7) and
Viscous friction Kf 673 × 10−3 Nm(rad/s)−1 VrQ = V(j,0–7) are a notation of voltage vectors corresponding to the
sorted values of JP and JQ, respectively). The number of the constrained
voltage vectors depends on values of the constraints, If the value of CP is
Table 4 set up to choose the best two voltage vectors and the value of CQ is big
DPC, PPC and PMPPC controller parameters. enough that leads to all the voltage vectors are selected (because values
Ts Average Switching HP/CP HQ/CQ Weighting of CP and CQ can neither be too big nor too small), this situation is
Frequency factor KQs exactly the same with sequential model predictive power control
SMPPC. Indeed, if the intersection is empty, this means unsuitable
DPC&PMPPC 10−5s 23.75 kHz 17.5 kW 17.5kVar –
PPC 10−5s 23.75 kHz – – 1.1
constraints are selected, it will results in poor performance.
The implementation of the proposed PMPPC method is detailed in
flowchart of Fig. 7; it includes thirteen (13) basic steps as shown in
location of the stator flux for using of the switching table. The ad- Algorithm 1.
vantage of using cost functions with the switching vectors intersection
Algorithm 1: Implementation of the PMPPC algorithm for DFIG
in PPPC is that all available voltage vectors are evaluated in different
sectors for one control period. Besides, both DPC and PMPPC algo- Step 0: Parameters initialization:
rithms are based on a stationary reference frame. The comparison Ts, CP, CQ, Vdc,Vrk = {V0, V1, ...,V7}
among traditional DPC, conventional PPC and PMPPC is given in Step 1: Estimation of stator and rotor flux
Table 2. // Eqns. (13 & 14).
Step 2: Calculation of stator powers and their derivatives
// Eqns. (16, 17) and Eqns. (23, 24).
A. Active and Reactive Power Cost Function Minimization
Step 3: Select first voltage vector. //Vrk = Vi = 1
Step 4: Prediction of stator powers. // Eqns. (25 & 26).
The proposed parallel MPPC method [18,16,17] uses a sequence of Step 5: Evaluation of cost functions JP and JQ.
two cost functions for the stator active and reactive power to control // Eqns. (28 & 29)
each control objective instead of using a single cost function with Step 6: Select next voltage vector // Vrk = Vi + 1.
several control objectives related by a weighting factor [10–13]. In this Step 7: Go to step 4 Until i = 6 // Vkr= V7 ?
regard, we remind here that the advantage of PMPPC method is that the // Evaluation of all voltage vectors.
Step 8: Sort JP and JQ output in descending values.
design of weighting factor is avoided. Step 9: Compare sorted values of JP and JQ with the value of constraints CP and CQ.
Based on the predicted stator active power (25) and stator reactive Step 10: Select the available voltage vectors for JP and JQ.
power (26), the two cost functions are designed as [17,18]: // VrP and VrQ
Step 11: Intersect the selected voltage vectors of JP and JQ.

15
180
14 160
Rotor and DFIG speed (rad/sec)

13 140
Wind speed (m/s)

12 120

100
11
80 Rotor speed
10 DFIG speed
60
9
40
8 (a) 20
(b)
7 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time (sec) Time(sec)

Fig. 8. (a) Wind speed profiles; (b) rotor and generator speed.

7
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

5
x 10
2 5
x 10 P
s(PPC)
0 P
-5 s(DPC)

-2 P*
s
-10
P
s(PMPPC)
Stator active power (W)
-4
-15
-6 5 0.76 0.78 0.8 0.82 0.84 0.86
x 10

-8
-5
-10
-5.05
-12
-5.1
-14
0.4 0.45 0.5
-16
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time(sec)
5
x 10
5
x 10
Q*
0 s
6
Q
-1 s(DPC)

4 Q
-2 s(PMPPC)
Stator reactive power (Var)

Q
-3 s(PPC)
2
-4
5
0 -5 x 10
3
-2 0.3 0.35 0.4 0.45
2.5
-4
2
-6

-8 0.5 0.55 0.6 0.65 0.7 0.75 0.8

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

Time (sec)

Fig. 9. Performances evaluation of PMPPC, PPC and DPC control for DFIG with variable rotor speed; step change of active and reactive powers.

Algorithm 1: Implementation of the PMPPC algorithm for DFIG 17.5 kW/kVar, respectively. The DPC, PPC and PMPPC controllers
parameters are shown in Table 4.
// Vkr = Intersect (VrP, VrQ)= V*r
The applied wind profile is shown in Fig. (8.a), this wind speed
Step 12: Select the optimal voltage vector // Vrk = Vr∗.
Step 13: Go to step 1 // For next sampling period. varies between 8 m/s and 14.5 m/s with mean speed of 10.94 m/s. In
this test, we impose steps changes of active and reactive power while
6. Simulation results the rotor speed (turbine speed) is controlled through MPPT technique
and fixed at mean value of 85% synchronous speed, and consequently
The performances of the DPC, PPC and PMPPC control schemes the generator speed change from 94.2 rad/s to 158.7 rad/s (Fig. 8.b),
have been tested and verified using Simulink Matlab for a 1.5 MW DFIG this variation is a result of the wind speed variation.
as shown in Table 3. In order to record and compare the calculation In order to observe the dynamic and steady-state response of the
time during the simulations test, the sampling frequency is fixed at DFIG, various active and reactive references power steps changes with
100 kHz for the three control methods, which give the same sampling variable rotor speed were carried out (Fig. 9). Initially, the reference
period within Ts = 10 μs. The values of band amplitudes HP and HQ of reactive power Qs∗ is set to −0.5MVar for 0.3 s while the reference
hysteresis controllers in DPC are adjusted empirically to17.5 kW/kVar, active power Ps∗ is set to zero for 0.2 s from the beginning. To test the
respectively, with average switching frequency of 23.75 kHz (shorter dynamics of the control methods, a step changes of the reference active
than sampling frequency 100 kHz). In order to select the optimal vol- power Ps∗ are applied; the first one from 0 MW to −0.5 MW at t = 0.2 s,
tage vectors for the switching vectors intersection in PMPPC under the the second is from −0.5 MW to −1.5 MW at t = 0.5 s and the third is
same switching frequency with DPC and PPC (23.75 kHz), constraints from −1.5 MW to −0.5 MW at t = 0.8 s. (“−” refers to generating
CP and CQ of PMPPC are adjusted empirically at the same value active power and absorbing reactive power). Therefore, step changes in

8
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

5
x 10
P

Active power (W)

-4.8 s(PMPPC)
P
-4.9 s(DPC)
P*
-5 s
P
-5.1 s(PPC)

0.25 0.3 0.35 0.4 0.45

Time(sec)
5
x 10
Reactive power(W)

-4.8
Q*
-4.9 s

-5 Q
s(DPC)
-5.1 Q
s(PMPPC)
-5.2 Q
s(PPC)
0.05 0.1 0.15 0.2 0.25
Time(sec)

Fig. 10. Zoom of Fig. 9 (step change of reactive power at t = 0.3sec and active power at t = 0.2sec).

5
x 10
6
Estimated power error (DPC)
4 Predicted power error (PMPPC)
Predicted power error (PPC)
2

0
P error (W)

5
-2 x 10
5
-4 4
s

3
-6
2
-8
1

-10 0
0.28 0.29 0.3 0.31 0.32 0.33 0.34 0.35 0.36
-12
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time(sec)
6
x 10
1.5
5
x 10
0
1
-2

-4
0.5
-6
Q error (W)

0.19 0.2 0.21 0.22 0.23 0.24 0.25 0.26

0
s

-0.5

-1
Predicted power error (PMPPC)
Estimated power error (DPC)
Predicted power error (PPC)
-1.5
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time(sec)

Fig. 11. Variation of powers errors in step change of active and reactive power of DPC, PPC and PMPPC methods for DFIG.

9
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

5000
6000
4000
4000
2000
0
3000
-2000
-4000
2000
0.48 0.5 0.52 0.54
Stator current I (A)

1000
s

-1000

-2000

-3000

-4000

-5000
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Time (sec)

6000

4000

2000
Rotor current I (A)
r

-2000

-4000 2000
0

-6000 -2000
0.14 0.16 0.18 0.2 0.22 0.24 0.26
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Time (sec)
Fig. 12. Step change responses in stator and rotor currents of PMPPC control for DFIG with variable rotor speed.

Table 5 t = 0.8 s.
DPC, PPC and PMPPC performance evaluation. The steady-state and dynamic behaviors of traditional DPC, con-
Criterion Traditional DPC Conventional PPC Proposed
ventional PPC and Parallel MPPC is shown in Fig. 9. From these results,
PMPPC it can be noted for the PMPPC, PPC and DPC control methods, that the
active and reactive power references are well followed and reached by
%Powers ripple %Pripple 2.87% 2.42% 2.17% the generator in about 0.01 s for PMPPC method, whereas, in about
%Qripple 4.77% 3.87% 3.43%
0.03 s and 0.08 s for PPC and DPC methods, respectively. An important
% Currents THD IsTHD 3.89% 3.76% 3.52%
IrTHD 4.25% 3.85% 3.66% test in Fig. 10 at t = 2sec and t = 3sec, is that the reactive (active)
Calculation Time 3 μs 4 μs 6 μs power not much affected when the active (reactive) power changes by
PMPPC method compared to DPC and PPC methods.
In order to show the feasibility of the PMPPC method, we showed
the reference reactive power Qs∗ are also applied; the first one from the estimated power error for DPC and predicted power error for the
−0.5MVar to 0MVar at t = 0.3 s, the second is from 0MVar to PPC and PMPPC methods as shown in Fig. 11. As is apparent from these
0.25MVar at t = 0.5 s and the third is from 0.25MVar to 0.5MVar at results, the power error in the proposed method is the smallest

10
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

PERIODIC SIGNAL (Fundamental 50 Hz, RMS=1340.8582, THD=3.89%) PERIODIC SIGNAL (Fundamental 50 Hz, RMS=1341.4889, THD=3.76%)
5000 5000

sa
sa

Stator Current I
Stator Current I

0 0

-5000 -5000
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Time (sec) Time (sec)
FOURIER COMPONENTS
1500 FOURIER COMPONENTS
1500
AMPLITUDE

1000

AMPLITUDE
1000
500
500
0
0 20 40 60 80 100
HARMONIC ORDER 0
PERIODIC SIGNAL (Fundamental 10 Hz, RMS=2971.2766, THD=4.25%) 0 20 40 60 80 100
HARMONIC ORDER
ra

5000
Rotor Current I

PERIODIC SIGNAL (Fundamental 10 Hz, RMS=2972.8867 THD=3.85%)

4
0 1
x 10

-5000
0.5

ra
Rotor current I
0 0.2 0.4 0.6 0.8 1
Time (sec) 0
FOURIER COMPONENTS
3000 -0.5
AMPLITUDE

2000 -1
0 0.2 0.4 0.6 0.8 1
Time(sec)
1000 FOURIER COMPONENTS
2500

0 2000
0 20 40 60 80 100
AMPLITUDE

1500
HARMONIC ORDER
1000
Fig. 13a. Stator and rotor currents spectrum analysis for traditional DPC.
500

0
0 20 40 60 80 100
compared to the other two methods, especially when one of the two HARMONIC ORDER
powers changes. This means that the PMPPC dynamic and steady-state
response is much better compared to PPC and DPC methods. Fig. 13b. Stator and rotor currents spectrum analysis for conventional PPC.
For further verification of the proposed method, the rotor and stator
currents are presented in Fig. 12. Although there is a reverse in both In terms of active power ripple, it shows that the power ripple value
rotor and stator currents during step changes in the references power as of DPC is more than 0.7% as the value of PMPPC, namely, 2.87% for
is evident in the zoom of Fig. 12, the results show that the stator and DPC and 2.42% for PPC versus 2.17% for PMPPC and more than 1.34%
rotor phase currents have a sinusoidal shapes. In addition, a comparison for reactive power, namely, 4.77% for DPC and 3.87% for PPC versus
of the steady-state performances of DPC, PPC and PMPPC in terms of 3.52% for PMPPC. With regard to the stator and rotor currents spec-
stator active and reactive powers ripple, stator and rotor currents total trum of Figs. 13a, 13b and 13c for DPC, PPC and PMPPC methods,
harmonic distortion (THD) and calculation time is summarized in respectively, the DPC, PPC and PMPPC have very close THD values with
Table 5. 3.89% and 3.76% versus 3.52% for stator current, whereas, with 4.25%
The currents THD and Powers ripple can be expressed as follows: and 3.85% versus 3.66% for rotor current, respectively.
The calculation time of PMPPC is approximately 2 μs more than
- The THD of stator and rotor currents are calculated by: PPC, and 3 μs more than DPC (Table 5) which is shorter than the
sampling period Ts = 10 μs. This enables us to conduct practical ex-
I22 + I22 + ...+In2
%THD = . 100 periments of the proposed methods.
IF (30)
Where In is the root mean square (RMS) value of the harmonic n, IF is 7. Conclusion
the RMS value of the fundamental current.
This paper presents the PMPPC method for wind energy conversion
- For the ripples of active and reactive power are defined as: and power regulation of DFIG. The proposed predictive method uses
Pmax − Pmin two cost functions without weighting factor, which have been mini-
%Pripple = . 100 mized, introducing two adjustable parameters (active power constraint
Pavg (31a)
and reactive power constraint) to select the optimal voltage vector from
Qmax − Qmin the switching vectors intersection that takes advantage of the predicted
%Qripple = . 100
Qavg (31b) stator active and reactive power. This new method leads certainly to
more computational burden. However, it is simpler and more clearly to
Where Pmax, Qmax is the maximum active and reactive power; Pmin, Qmin
control stator active and reactive powers in concept and more com-
is the minimum active and reactive power; Pavg, Qavg is the average
parable with the concept of conventional DPC. In conclusion, this re-
active and reactive power, respectively.
search makes several contributions for DFIG power regulation. First, the

11
A. Benzouaoui, et al. Electrical Power and Energy Systems 125 (2021) 106453

PERIODIC SIGNAL (Fundamental 50 Hz, RMS=1341.5798 THD=3.52%)

4000
References
sa

2000
Stator current I

[1] Takahashi I, Noguchi T. A new quick-response and high-efficiency control strategy

0 of an induction motor. IEEE Trans Ind Appl 1986;IA-22(5):820–7.
[2] Arnalte S, Burgos J, Rodriguez-Amenedo J. Direct torque control of a doubly-fed
-2000 induction generator for variable speed wind turbines. Electr Power Compon Syst
2002;30(2):199–216.
-4000
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 [3] Madishetti S, Bhuvaneswari G, Singh B. Improved power quality converter for direct
Time(sec) torque control-based induction motor drives. IET Power Electron.
FOURIER COMPONENTS 2013;6(2):276–86.
1500 [4] Datta R, Ranganathan VT. Direct power control of grid-connected wound rotor in-
duction machine without rotor position sensors. IEEE Trans Power Electron
2001;16(3):390–9.
AMPLITUDE

1000
[5] Eloy-Garcia J, Arnaltes S, Rodriguez-Amenedo JL. Extended direct power control for
multilevel inverters including DC link middle voltage control. IET Electr Power Appl
500
2007;1(4):571–80.
[6] Eloy-Garcia J, Arnaltes S, Rodriguez-Amenedo JL. Direct power control of voltage
source inverters with unbalanced grid voltages. IET Power Electron
2008;1(3):395–407.
10 20 30 40 50 60 70 80 90 100 [7] Abad G, Rodriguez MA, Iwanski G, Poza J. Direct power control of doubly fed in-
HARMONIC ORDER duction generator based wind turbines under unbalanced grid voltage. IEEE Trans
Power Electron 2010;25(2):442–52.
[8] Santos-Martin D, Rodriguez-Amenedo JL, Arnalte S. Direct power control applied to
PERIODIC SIGNAL (Fundamental 10 Hz, RMS=2973.5836, THD=3.66%
ra

doubly fed induction generator under unbalanced grid voltage conditions. IEEE
Rotor Current I

5000 Trans Power Electron September 2008;23(5):2328–36.

[9] Xu L, Cartwright P. Direct active and reactive power control of DFIG for wind en-
0 ergy generation. IEEE Trans Energy Convers September 2006;21(3):750–8.
[10] Phan V-T, Lee H-H. Improved predictive current control for unbalanced stand-alone
-5000 doubly-fed induction generator-based wind power systems. IET Electr Power Appl
0 0.2 0.4 0.6 0.8 1 2011;5(3):275–87.
[11] Filho AJS, Filho ER. Model-based predictive control applied to doubly-fed induction
Time (sec) generator direct power control. IEEE Trans Sust Energy 2012;3(3):398–406.
FOURIER COMPONENTS [12] Zhi D, Xu L, Williams BW. Model-based predictive direct power control of doubly
4000 fed induction generators. IEEE Trans Power Electron February 2010;25(2):341–51.
[13] Hu J, Zhu J, Zhang Y, Platt G, Dorrell DG, Ma Q. Predictive direct control of doubly
AMPLITUDE

2000 fed induction generator for grid synchronization in wind power generation. 2011
IEEE energy conversion congress and exposition, Phoenix, AZ; 2011. p. 2381–8.
[14] Hu J, Zhu J, Zhang Y, Platt G, Ma Q, Dorrell D. Predictive direct virtual torque and
0 power control of doubly-fed induction generators for fast and smooth grid syn-
chronization and flexible power regulation. IEEE Trans Power Electron Jul.
-2000 2013;28(7):3182–94.
0 20 40 60 80 100 [15] Zhang Y, Hu J, Zhu J. Three-vectors-based predictive direct power control of the
doubly fed induction generator for wind energy applications. IEEE Trans Power
HARMONIC ORDER
Electron July 2014;29(7):3485–500.
[16] Rojas CA, Rodriguez J, Villarroel F, Espinoza JR, Silva CA, Trincado M. Predictive
Fig. 13c. Stator and rotor currents spectrum analysis for Parallel MPPC.
torque and flux control without weighting factors. IEEE Trans Ind Electron Feb
2013;60(2):681–90.
[17] Zhang Y, Zhang B, Yang H, Norambuena M, Rodriguez J. Generalized sequential
new parallel MPPC method was developed and compared with con- model predictive control of IM drives with field-weakening ability. IEEE Trans
ventional DPC and PPC methods. Second, high-quality power genera- Power Electron Sept. 2019;34(9):8944–55.
[18] Wang F, Xie H, Chen Q, Davari SA, Rodríguez J, Kennel R. Parallel predictive torque
tion was investigated and achieved. Finally, the proposed predictive
control for induction machines without weighting factors. IEEE Trans Power
method has been proven by simulation validation to be high efficient in Electron Feb. 2020;35(2):1779–88.
wind energy conversion and DFIG power regulation. However, this [19] Ebrahim Zarei M, Asaei B. Combined vector control and direct power control
methods for DFIG under normal and unbalanced and distorted grid voltage condi-
newly developed method needs experimental validation. As a per- tions. 4th Annual International Power Electronics, Drive Systems and Technologies
spective, the proposed method will be integrated into the real appli- Conference, Tehran. 2013. p. 107–12.
cation together with LIDAR devices. [20] Taibi F, Benzineb O, Tadjine M, Boucherit MS, Benbouzid MEH. Hybrid Sliding
mode control of DFIG with MPPT using three multicellular converters. 19th IFAC
world congress. Cape Town, South Africa, 2014; 2014. p. 24–9.
[21] Hu J, Zhu J, Dorrell DG. Model-predictive direct power control of doubly-fed in-
duction generators under unbalanced grid voltage conditions in wind energy ap-
CRediT authorship contribution statement plications. IET Renew Power Gener August 2014;8(6):687–95. https://doi.org/10.
1049/iet-rpg.2013.0312.
Ahmed Benzouaoui: Methodology, Software, Validation, Formal
analysis, Investigation, Writing - original draft, Visualization. Houari Ahmed Benzouaoui (Ing, Assis Prof, Asso Prof) was born in Chlef, Algeria. He received
his BS degree in electrical engineering from the University of Abd El Hamid Ibn Badis of
Khouidmi: Supervision, Conceptualization, Writing - review & editing. Mostaganem, Algeria until 2007 and the MS degree from USTO-Oran-Algeria 2010. He
Boubaker Bessedik: Resources, Writing - review & editing, joined the Department of Electrical Engineering, USTO-Oran –Algeria as a D.sc. student in
Visualization. 2012. He is currently an Associate Professor in the Faculty of Technology at the Hassiba
Benbouali University of Chlef-Algeria. His research interests include Automatic control
and robotics, Nonlinear control and Power systems.

Declaration of Competing Interest Houari Khouidmi is currently an Associate Professor in the Department of Electronics at
the Hassiba Benbouali University of Chlef. His research interests include Power electronic
converters, Predictive control, Intelligence control and the advanced control of electrical
The authors are not aware of any affiliations, memberships, funding, drives. He is also interested in energy management systems in microgrids and renewable
or financial holdings that might be perceived as affecting the objectivity energy applications such as PV, wind turbines, and hybrid generation and storage sys-
of this paper. Furthermore, there is no conflict of interest exists. tems.

Boubakar Bessedik (Ing, Assis Prof) was born in S.M.Benali, Algeria. He received his BS
degree in electrical engineering from the University of Sciences and Technology of Oran,
Acknowledgements (USTO), Algeria until 1996 and the MS degree from USTO-Oran-Algeria 2003. He joined
the Department of Electrical Engineering, USTO-Oran –Algeria as a D.sc. student. He is
currently an Assistant Professor in the Faculty of Technology at the Hassiba Benbouali
The author(s) would like to thank the anonymous reviewers for University of Chlef-Algeria. His research interests include Energy systems, Power elec-
several helpful suggestions on the manuscript revision. tronic converters and advanced control.

12