You are on page 1of 10

An Improved Finite Control Set -Model Predictive Control (FCS-MPC)

algorithm with Imposed Optimized Weighting Factor


S. Alireza Davari1,2, Davood A. Khaburi2, Peter Stolze1 and Ralph Kennel1
1
Institute for Electronics Drive Systems & Power Electronics, Technische Universitaet
Muenchen
Arcisstr. 21, Munich, Germany
2
Center of Excellence for Power Systems Automation and Operation, Elec. Eng. Department,
Iran University of Science and Technology
Narmak, Tehran 16846-13114, Iran

Keywords: induction motor control, predictive torque control, weighting factor

Abstract- In this paper an improved FCS-MPC algorithm with optimized weighting factor is
presented. The main goal is reducing the torque ripples for the two-level inverter utilization. The
weighting factor is optimized based on dividing the control interval in two parts. The method is
validated by simulations and experiments.

1- Introduction
The Direct Torque Control (DTC) method for AC machines is a suitable method for variable speed
drives especially in the case that torque control is more desired than speed control. In spite of the
advantages of the DTC method (fast response and straightforward algorithm [1]), conventional DTC
has two considerable drawbacks; variable switching frequency and high torque ripples [2]. In recent
years, several investigations have been performed in order to tackle the disadvantages of DTC method.
Among them, the predictive control has drawn more tendencies from researchers. The Model
Predictive Control (MPC) determines the optimum voltage by using the explicit model of the motor
and the inverter by minimizing a cost function [3]. The Finite Control Set- Model Predictive Control
(FCS-MPC) method uses a simple approach to minimize the cost function. In this method the discrete
nature of the power converters is contemplated. Therefore, the feasible voltage vectors (including zero
voltage) are examined in the cost function. The voltage vector (VV) that minimizes the cost function is
the most appropriate one [4].
Despite all advantages which FCS-MPC brings about, similar to all of the cost function based
predictive control methods, adjusting the weighting factor is an important challenge. Since the
weighting factor allocates the importance of each controlled state, it has a drastic effect on the
response of the method. [5] has introduced some guidelines for weighting factor adjustment which are
based on algorithmic empirics.
This paper presents an improved FCS-MPC in order to achieve low torque ripples. The weighting
factor of the cost function is calculated via an optimization method. The torque ripple is calculated as a
function of weighting factor in each control interval. The control interval is divided in two parts:
active time for application of the selected VV and zero time for application of the zero voltage. The
derivative of the torque ripple to the weighting factor is set to zero. This procedure lets the method be
applicable for using two-level VSI. In this method the feasible VVs are restricted to a couple of active
VVs.
2- Improved Predictive Torque Control Scheme

The cost function of the PTC method is the criterion to select the best voltage vector among the
feasible ones. Since the torque and flux are controlled directly in DTC branched methods, the
following equation is used as cost function.
1  
2
2 2 2
Jj = T j , n+1 − T * + Q λ sj , n+1 − λ s *  (1)
2  

j = 1,2,...., m

where, T j , n+1 and λsj , n +1 are the predicted torque and stator flux magnitude by considering the
application of jth voltage vector. T * and λs
*
are the torque and flux references. Q is a weighting
factor that determines the importance of flux control compared to torque control. m is the number of
feasible voltage vectors.
The weighting factor is a very important parameter in cost function based predictive control
because the merit of the response varies when the weighting factor is changed. The weighting factor is
usually tuned by means of empirical estimating the importance of the controlled state variables [5],
[6]. This paper presents a novel method for calculation of the appropriate weighting factor in each
control interval
The weighting factor is set to the value that minimizes the torque ripples. For this purpose the
torque ripple is attained as a function of weighting factor and the weighting factor that zeros the
derivative of the torque ripple is calculated.
In this paper the control interval is divided in two parts, i.e. active time (for active voltage exertion)
and zero time (zero voltage exertion). Therefore, the the effective value of the torque ripple can be
calculated as below:

∫0 (s1t + Tn − T ) dt
2 1 tvv * 2
Tripp =
ts
(2)
∫ (s t − s t )
1 ts 2
+ 2 2 vv + s1t vv + Tn − T * dt
ts tvv

where Tn is the present time torque, tvv is the active time, t s is the sampling and control interval and s1
and s2 are the slopes of torque during active and zero time respectively. These slopes are calculated by
the following equations.
 R R  Vsq λrd − Vsd λrq 
s1 = − s + r T (t n ) + K ′  (3-a)
 σLs σLr  − ω r (λsd λrd + λsq λrq )
3 p Lm
K′ = (3-b)
2 2 σLs Lr
 R R 
s 2 = − s + r  (
T (t n ) − K ′ω r λds λdr + λqs λqs ) (4)
 σLs σLr 
Where λs and λ r are the stator and rotor fluxes, σ = 1− L2m Ls Lr , τ r is the rotor time constant, Ls , Lr
and Lm are the stator, rotor and mutual inductances, respectively. Rs and Rr are stator and rotor
resistances, respectively and ωr is the rotor speed.
If the upper limit of the torque is set to T * + ∆T , by defining parameter M as below
M = T * + ∆T − Tn (5)
the torque ripple can be calculated by the following equation.
 1 M 2 1  M 2 
− + s 2   
=  
2 1 2 s1 2  s1 
Tripp  (6)
ts
− s M t + 1 s t 2 + Mt + t T − T *
 2 s1 s 2 2 s s s n ( )


In order to find the weighting factor that minimizes the torque ripple, the derivative of torque ripple
to the weighting factor has to be set to zero. It can be easily found out that the only parameters in (6)
that may relate to weighting factor, are s1 and s2. By considering (3) and (4), the only parameter that is
related to weighting factor is s1 because that is the only parameter related to  and  and the
selection of them is based on cost function. Therefore,
2
Tripp 2
= Tripp (s1 ), s1 = s1 (Vs ), Vs = Vs (Q ) → Tripp
2 2
= Tripp (Q ) (7)
In order to find the optimized weighting factor, the following equation is used.
∂ 2
Tripp = 0 (8)
∂Q
Solving (8) leads to the optimized weighting factor which minimizes the torque ripples as below
(refer to Appendix).
β 2 λ rd − β1λ rq
Q= (9-a)
( ) (
G + λ rq α 1 + Vsd , n −1 − λ rd α 2 + Vsq , n −1 )
s2 M
G=
1 
K ′ M + s 2Ts 
 2  (9-b)
1  Rs R 
+  + r Tn + ω r (λsd λrd + λsq λrq )
K ′  σLs σLr 

Symbols of (9) are elaborated in Appendix.
The weighting factor which is calculated via (9) is used in (1) in order to calculate cost function for
each feasible voltage vector.
In order to predict the torque and flux for the next sampling interval, the discrete motor and inverter
model is used as below:
λs j , n +1 = λs , n + ts Rs ⋅ is + ts V j (10)
3 p
T j , n +1 =
2 2
{
λ sj , n +1 × isλ , n } (11)
where i s is the stator current. V j is the jth voltage vector. ts is the control interval and isλ (tn ) is
calculated as below [2]:
 t  λ  λ 
isλ , n = 1 − s is , n − s , n + jω r ⋅ t s  i s , n − s , n  (12)
 στ r  σL s  σLs 
 
The most appropriate VV is the one which is predicted to minimize the cost function. In
conventional method the appropriate VV is applied for the whole control interval. In this paper the
selected VV is applied for a calculated portion of the control interval. The remainder of the control
interval is allocated to zero voltage.
Fig. 1 shows the torque rise and fall in a control interval. It is depicted that if the selected voltage
vector is held for the whole interval, the torque will reach to T j − selected , n+1 , but because the active time is
controlled the torque will reach to the desired value and then started to descend by zero voltage
insertion.
In order to calculate the interval that the active VV has to be applied (active time), the slope of the
torque for selected VV is predicted.
T j − selected , n+1 − Tn
T& = (13)
ts
Therefore, if the acceptable ceiling of the torque is T * + ∆T , the active time can be calculated as
below if the torque rise is assumed linear.
t vv =
T&
[
1 *
T + ∆T − Tn ] (14)
where tvv is the active time and ∆T is the allowed upper band for torque rising.
The rest of the control interval is allocated to zero voltage.
t zv = t s − t vv (15)
T j − selected (t n+1 )

T&

∆T Tref

T (t n )

tn t n+1

tVV t ZV

ts

Fig. 1: Typical rise and fall of the torque

Fig. 2: The variation of the absolute value of the (a) ascending and (b) descending slope of the torque
due to speed

Note that there is no need to constrain the descending of the torque during application of zero
voltage if the control interval is adjusted properly (less than 200 µs in this work). The reason can be
elaborated by considering the ascending and descending slopes of the torque in high speed and low
speed regions. Fig. 2 shows the variation of the absolute values of ascending and descending slopes
due to the speed variation.
The proposed method can be summarized to the following main steps:
1- Current and voltage measurement.
2- Estimation of the present fluxes and torques by means of voltage model observer.
3- Calculation of the optimized weighting factor.
4- Prediction of the torque and stator flux for the feasible VVs (if the torque reference is changed
the number of feasible VVs is 6 otherwise that is 2).
5- Examination of the predicted torques and stator fluxes in the cost function which the
weighting factor of it is calculated in first step.
6- Calculation of the active time for selected VV.
7- Application of the selected VV during calculated active time.
8- Application of the zero voltage for the rest of the control interval.
The whole process flow chart is shown in Fig. 3.
start

measure is and Vs

predict λ s j , n +1 and T j , n +1
use Voltage Model Observer
and calculate λ s , n , λr , n calculate the cost
, Tn and section function ( J j ) by using
the optimized Q
calculate the optimized weighting
factor (Q ) by means of λ s , n , λrs , n , Tn ++ j

no
is the Torque j > m
in the band? yes

yes
no
min {J j }j =1,.., m
m=2 m=6
VV selection
j =1
calculate
active time (t vv )
V j = VVsec + j
end

Fig. 3: The flowchart of the improved predictive torque control with weighting factor optimization

3- Results
In order to verify the the proposed method, simulation and experiment results are proposed. In order to
compare the features of the proposed method with the conventional predictive torque control [4], by
considering two-level VSI, both methods have been implemented. The control interval duration is set
to 100µs in all simulations and experiments. The specification of the induction motor is shown in
Table I.
A. Simulation Results

First, in order to demonstrate the importance of the value of the weighting factor, Fig. 4 shows the
comparison among the detailed torque responses of the proposed improved predictive torque control
method with various values of weighting factor and the optimized one. The results prove the drastic
effect of the weighting factor on the torque ripple.
In order to show the effect of the weighting factor on the stator flux and current Fig. 5 shows the
comparison among the detailed stator flux and torque responses when different weighting factor is
used. It proves that when the weighting factor is extremely small all of the controlled states will have
high ripple. When the weighting factor is calculated via the proposed method, the flux and current
don’t possess the best responses but their treatment is acceptable.
Fig. 6 shows the torque, speed, stator flux and stator current responses of the conventional FCS-
MPC and the proposed PTC method with weighting factor optimization, when a torque step is
imposed. The torque ripple for conventional FCS-MPC is 21% but it is reduced to 3.2% by using the
proposed method. The THD for the stator current is 79% when the conventional FCS-MPC is applied
but it is reduced to 20% by using the proposed method.
TABLE I

INDUCTION MACHINE SPECIFICATIONS

Pn=1.5 [kW] Rs=5.63 [Ω]

Vn=220 [V] Rr=4.35 [Ω]

In=3 [A] Ls=489 [mH]

ωn=2820 [rpm] Lr=489 [mH]

p=1 Lm=460 [mH]

Tn=5.1 [Nm] fsn=50 [Hz]

Fig. 4: Detailed torque responses of improved predictive torque control- simulation result

Fig. 5: Detailed comparison of (a), (b) stator flux and (c), (d) stator current for different weighting

factors - simulation result


(a) (b)
Fig. 6 : Nominal torque step response of the (a) conventional FCS-MPC (b) improved predictive
control- simulation result.
B. Experimental Results

Fig. 7 shows the experimental setup. Utilizing eZdsp TM F2812 card renders excellent digital signal
processing ability to implement the proposed algorithm without problems. In order to have a short
time delay and also avoid of the noise, the switching is done in the middle of the control interval.
Fig. 8 shows the responses of the conventional FCS-MPC method and proposed method. The
torque ripple for conventional FCS-MPC is 19% but it is reduced to 4.4% by using the proposed
method. The THD for the stator current is 83% when the conventional FCS-MPC is applied but it is
reduced to 28% by using the proposed method.

Fig. 7: Experimental setup (a) Power and signal boards (b) Induction motor (c) DC power supply
(a) (b)
Fig. 8 : Nominal torque step response of the (a) conventional FCS-MPC (b) improved predictive
control- experimental result.
Fig. 9 shows the low speed performance of the conventional FCS-MPC method and the proposed
method. Experimental results prove that the proposed method keeps its privileges at low speed region.
Fig. 10 shows the detailed variation of calculated active time and weighting factor at low speed
region experiment. The coordination among active time, weighting factor and stator flux is
perceivable. Similar to the simulation results, the experimental results imply that when the stator flux
deviates from its reference value (1 Wb), the weighting factor increases to amend flux response. Also
when the flux is getting close to the reference value and the active time decreases, which means
increase of the torque ripple, the weighting factor descends to keep the torque ripples low.

(a) (b)
Fig. 9 : Low speed performance of (a) conventional FCS-MPC (b) improved predictive control -
experimental result.
Fig. 10: Detailed control parameters at low speed region for the improved predictive control. (a)
Active time (b) Stator flux (c)Weighting factor - experimental result
4- Conclusion
In this paper an improved cost function based predictive torque control with weighting factor
optimization has been proposed. The method of examining the cost function with feasible Voltage
Vectors (FCS-MPC) has been improved with the goal of achieving the possibility of utilizing two-
level VSI. For this purpose, the weighting factor of the cost function is optimized in every control
interval in which the active and zero voltage are applied.
The validity of the proposed method is verified by simulations and experiments. The results proved
that the proposed method possesses acceptable torque ripple besides acceptable flux response in wide
speed range. Because of active time calculation and weighting factor optimization the method keeps its
merit at low speed region which is the common disputed speed region for the DTC branched methods.

Appendix
Proof of the Calculated Optimized weighting factor
Equation (8) can be satisfied if one of the following equations is true.

s1 = 0 (16)
∂Q
s2 M
s1 = (17)
1
M + s2t s
2
If (16) is considered as a criterion for weighting factor calculation parameter Q will be vanished
from the equation.
Consequently, (17) is the criterion for optimized Q calculation.
Therefore,
V sq λ rd − V sd λ rq = G (18)
Equation (18) can elaborate the criterion of the optimized weighting factor.
By considering the linear model of induction motor the relationship between voltage and weighting
factor is:
β
∆Vsd = α 1 + 1
Q
(19)
β
∆Vsq = α 2 + 2
Q
where
− f 1 (e222 e11 − e12 e21e22 ) − f 2 (e122 e21 − e11e22 e12 )
α1 = , β1 =
(e11e22 + e12 e21 ) 2
(e11e22 + e12 e21 )2
(20-a)
− f (e 2 e − e e e ) − f 2 (e112 e22 − e11e12 e21 )
α 2 = 1 21 12 11 212 22 , β 2 =
(e11e22 + e12 e21 ) (e11e22 + e12 e21 )2
∆λsd 
∆λ 
 f1   sq 
f  = H ⋅ A ⋅ (20-b)
 2 ∆λrd 
 
∆λrq 
e11 e12  − K ′Ts λrq K ′Ts λrd 
0 0

e e   2T λ0=  (20-c)
 21 22   s sd 2Ts λ0sq 
 1 
a11 + t a12 a13 a14 
 s 
 a 1
a22 + a23 a24 
 21 ts 
A = ts   (20-d)
1
 a31 a32 a33 + a34 
 ts 
 1
 a41 a42 a43 a44 + 
 ts 
By substitution of (19) in (18) the optimized weighting factor is calculated as (9).
References
[1] I. Takahashi and T. Noguchi, “A New Quick-Response and High Efficiency Control Strategy of an
Induction Machine”, IEEE Trans. Ind. App. no. 22, 820–827, 1986.
[2] J. K. Kang, and S. K. Sul, “New Direct Torque Control of Induction Motor for Minimum Torque
Ripple and Constant Switching Frequency,” IEEE Trans. Ind. App., vol. 35,no 5, pp. 1076-1072,
Sept./Oct. 1999.
[3] A. Linder, R. Kanchan, R. Kennel and P. Stolze, “Model-Based Predictive Control of Electric
Drives, Cuvillier Verlag Goettingen, 2010.
[4] R.Vargas, M. Rivera, J.Rodrıguez, J. Espinoza and P. Wheeler, “Predictive Torque Control with
Input PF Correction Applied to an Induction Machine fed by a Matrix Converter,” IEEE PESC, 15-
19 June 2008.
[5] P. Cortes, S. Kouro, B. L. Rocca, R. Vargas, J. Rodriguez, J. I. Leon, S. Vazquez, L.G Franquelo,
“Guidelines for Weighting Factors Design in Model Predictive Control of Power Converters and
Drives,” IEEE International Conference on Industrial Technology, ICIT, 10-13 Feb. 2009 pp. 1 –
7.
[6] R. Vargas, J. Rodriguez, U. Ammann and P.W. Wheeler, “Predictive Current Control of an
Induction Machine Fed by a Matrix Converter With Reactive Power Control,” IEEE Trans. Ind.
Elec., vol. 55, no 12, pp. 4372-4380, Dec. 2008.

You might also like