Sie sind auf Seite 1von 6

CGE’07 EMP, ALGERIE, 12-13 Avril, 2011

Optimized Dual Randomized PWM Technique


for the Three Phase Inverter
N. Boudjerda†, A. Boudouda†, M. Melit†, B. Nekhoul†, K. El Khamlichi‡ and K. Kerroum‡
†LAMEL, univ. M. S. Ben Yahia, Jijel Algeria, ‡LASMEA, univ. B. Pascal, Clermont Ferrand France

(Randomized Pulse Position Modulation: RPPM) in both


Abstract — Randomized Pulse Width Modulation (RPWM) of DC-DC and DC-AC conversions [3] - [5]. It has been
has become a viable alternative to Deterministic PWM shown that RCFM scheme allows a better spreading of the
(DPWM). Indeed by spreading the power spectrum over a spectrum, than RPPM scheme [6]. However, to obtain a
continuous noise, this new technique complies better with
Electromagnetic Compatibility (EMC) standards for
maximum spread, the combination of the two schemes has
conducted Electro-Magnetic Interferences (EMI) and allows been proposed (RCFM-RPPM) that we call Dual RPWM
reducing the emitted acoustic noise in Variable Speed Drives scheme [5]-[7]: the switching functions are generated by
(VSD). The most popular RPWM schemes are Randomized comparing a carrier with two random parameters to three
Pulse Position Modulation (RPPM) and Randomized Carrier deterministic reference signals: one switching function for
Frequency Modulation (RCFM). A combination of these two each leg of the converter. We propose in this paper an
schemes (RCFM-RPPM) or Dual RPWM (DRPWM) has also
been proposed. In this paper, we propose an Optimized Dual optimized scheme in which the randomness levels are
Randomized Pulse Width Modulation (ODRPWM) scheme optimized so as to obtain the best spread shape of PSD
for three phase inverter. First, a modulating principle is “Optimized RCFM-RPPM”.
proposed. A mathematical model of power spectral density First, we present the modulating principle. Then an
(PSD) of output voltage is then developed. PSD analysis analytical model of the Power Spectral Density (PSD) of
shows that the proposed scheme is more effective on
the output voltage is developed. This model is expressed
spreading PSD. Moreover, this analysis reveals the existence
of optimal randomization parameters for a maximum spread directly in terms of the two random parameters of the
of the PSD. The optimization problem is then modeled and carrier. Note that the simple schemes (RCFM and RPPM)
solved using two powerful nonlinear methods: the trust are deduced directly, from the proposed general model
region method and the simplex algorithm. which has been validated in previous works [8]-[10].
Compared to the simple randomization schemes, the PSD
Keywords — EMC in power electronics, spread spectrum, analysis shows that the proposed dual RPWM scheme
random PWM, optimized RPWM scheme.
allows a better spread shape of PSD that is the desired
EMC advantage. In addition, a parametric study shows that
I. INTRODUCTION
there is an optimum shape of the PSD based on statistical

T he use of PWM technique for the control of power


converters allows adjusting the useful component of
the voltage and eliminating some unwanted harmonics [1].
parameters, The treatment of this problem with two
powerful nonlinear methods (Trust Region method and
Simplex algorithm), gives the same results.
Thus, it is required for power converters to provide the
desired electrical functionality and to meet international II. MODULATING PRICIPLE
standards of Electro-Magnetic Compatibility (EMC) by
The structure of the converter is given in Fig. 1; it
reducing conducted and radiated emissions [2].
requires three switching functions qa, qb and qc.
In order to better meet the EMC standards for conducted
disturbances at low frequencies, we can use the filtering qa qb qc
technique (passive and/or active power filters).
Udc /2
Furthermore, RPWM technique is one of the most
effective and least-cost solutions: it allows spreading the a
power spectrum over a wide frequency range while 0 b
reducing its amplitude and this constitutes a significant
EMC advantage without any additional hardware. c
Several works regarding this new control technique have Udc /2 qa qb qc
been published recently, two conventional RPWM
schemes are proposed: the scheme in which the switching
period is randomized (Randomized Carrier Frequency Fig.1. Three phase inverter
modulation: RCFM) and the scheme in which the period is
kept constant and the pulse position is randomized The modulating principle is illustrated in Fig.2; the three
switching functions qa, qb and qc are obtained by
comparing three deterministic reference signals va*, vb*,
and vc* to a randomized triangular carrier c. The switching Where va*(t) depend on the modulation strategy: Sine
functions are characterized by three parameters: switching Modulation (SM), Third Harmonic Injection (THI) and
period T, delay report δ and duty cycle d. Hybrid Space Vector Modulation (HSVM).
For sinusoidal modulation:
βmTm 
reference signal : v* t 
1 v a* (t ) = M sin  2π (5)
 T 
 1
For third harmonic injection:
randomized
  t  1  t  
M  sin  2π
2
+ sin  6π
carrier : c
v a** (t ) = (6)
t 3   T  6
 1
 T 
 1 
Tm
q
T1 is the period of the reference signal (i.e. period of
δmTm dmTm
output voltage) and M the modulation index.
1 For Hybrid SVM, the general expression of the three
switching
function : q switching functions can be condensed as follows [8], [9]:
***
vabc = vabc
*
+ v *zs (7)
tm tm+1 t
Where:
Fig.2. Modulating principle
In practice, (Fc >> F1) (Fc: carrier frequency and F1:
[
v *zs = − (1 − 2k 0 ) + k 0 v max
*
+ (1 − k 0 )v min
*
(8) ]
k0 may take any value between 0 and 1. For the classical
reference frequency), the reference is assumed to be
SVM, k0 = 0.5:
constant during the switching period Tm (generally, the
reference is updated (sampled) once or twice per period), v *zs = −0.5 v max
*
(
+ v min
*
= 0.5v mid
*
) (9)
the duty cycle dm is then equal to the reference at the * * *
v min, v mid and v max are respectively minimum, middle and
sampling instant tm. Subsequently, only the switching maximum values of reference signals v*abc. Fig. 4 and Fig.5
period Tm, (i.e. period of the carrier) and/or the delay ratio show the reference signals of the three strategies [8], [10].
δm can be really randomized. This can be achieved by
randomization of Tm (randomization of carrier frequency: v*max, v*mid, v*min [pu]
RCF) and/or the fall time report β (randomization of pulse
v*max
v*mid
position: RPP), (Fig.2). Note that the delay report δm can
v*min
1
be expressed as follows:
vb*
δ m = β m (1 − d m ) (1) 0.5
vc* va*

For a randomized βm between 0 and 1, δm varies


randomly between 0 and (1-dm). The use of parameter β
0

rather than δ is paramount important in DC-AC, it allows -0.5


defining the pulse position independently from the duty
cycle d which is varying in this case. Related to the -1
switching function qa (phase a), per unit phase to ground 0 0.005 0.01 0.015 0.02
voltage va0 can be expressed as follows, (Fig.3): t [s]
 1 Fig.4. Derivation of v*max, v*mid and v*min from v*abc
va 0 (t ) = 2 qa (t ) −  (2)
 2 magnitude [pu]
1
ua(t) HSVM: k0 = 0.5
third harmonic injection
δamTm damTm sine reference
0.5
1

tm tm+1 tm+k t
Tm γk Tm+k 0

va0(t)
1

t
-0.5
-1
-1
0 0.005 0.01 0.015 0.02
Fig.3. Output voltage and switching function
t [s]
And per unit phase to phase voltage is deduced easily: Fig.5. Comparison of different references
u ab (t ) = va 0 (t ) − vb 0 (t ) = 2(qa (t ) − qb (t )) (3)
The duty cycle d is time dependent, thus for phase a: In practice, reference signals v*abc are sampled at the

(
d a (t ) = 1 + va* (t )
1
2
) (4)
beginning of each switching period. Thus for the period
Tm = tm+1 – tm (Fig. 2), the resulting duty cycle dam depend Approximates closed forms of phase to ground and
on the random sampling instant tm that gives: phase to phase per unit voltage PSDs are set based on:
d am = d a (t m ) = (1 + ra (t m ))
1 1. Relations (2) and (3) between the switching functions
(10)
2 and the output voltages.
For b and c phases, reference signals are respectively 2. General expression (12) of the PSD of a wide sense
(2π / 3) shifted from phase a. stationary pulse signal.
3. Bech’s approximation [4] which assumes the random
III. POWER SPECTRAL DENSITY (PSD) value of the lag time γk (Fig.2) to be equal to its
statistical expectation:
The PSD of a τ long random signal uτ (t) is, [11]:
γ k = t m + k − t m +1 ≈ (k − 1)T
[ ]
(15)
W ( f ) = lim E F {uτ (t )}
1 2
(11)
τ →∞ τ
A. PSD of phase to ground voltage
Where:
F{uτ (t)}: Fourier transform of the signal uτ(t). In expression (16) below, the random aspects of T and β
E[.] : Statistical expectation. are clearly separated, which allows distinguishing RPPM
Fig.6. shows a randomized switching function in the and RCFM schemes:
Eθ ,Tm [sin 2 (πfd a (0,θ )Tm )]
ε
general case of (DRPWM): T and δ are randomized: Wa ( f ) ≈
T (πf )2
uτ(t)
  
  sin (πfd a (0, θ )Tm )e jπf (2 − d a (0,θ ))Tm    
   ETm   
[ ]

δmTm dmTm    × Eβ m e − j 2πfβ m (1− d a (0,θ ))Tm   
  Eθ   
jπfd a (k T ,θ )Tm + k   
  
 N  × E
sin πfd k T , θ T
 a ( ( m+k e) ) 
 

+ Re ∑  T +   j 2πfβ m + k (1− d a (k T ,θ ))Tm + k    

T (πf )2 × Eβ m + k e
m k
tm Tm tm+1 γk tm+k Tm+k t  k =1      

  
 × ET e j 2πfT

[ k −1
]  

Fig.6. Randomized switching function
[ ]
 −1 
 j 2πfT N 
A general expression of the PSD for a Wide Sense  × 1 − ET e  
   
Stationary (WSS) pulse signal u(t) (Fig. 3) can be derived (16)
as follows [8]: ε = 1 for f = 0 and ε = 2 otherwise

1 m [
E U ( f ) 2 ] 

θ : Random variable between 0 and 2π
W(f )=  (12) T 


[ 
T + 2 Re  ∑ E U m ( f )U m* + k ( f )  

] N = round  1 
T 
 k =1 
Where:
U m ( f ) : Fourier transform of the signal um(t) during the B. PSD of phase to phase voltage
switching period Tm given by: sin 2 (πfd a (0, θ )Tm ) + sin 2 (πfd b (0, θ )Tm ) 
 
Eθ ,Tm − 2 sin (πfd a (0,θ )Tm )sin (πfd b (0,θ )Tm )
2
Wab ( f ) ≈ 
1 − jπf (2δ m + d m )Tm T (πf )2
Um(f )= e sin (πfd m Tm )e − j 2πft m (13) × E [cos(πf (1 − 2β ) (d (0, θ ) − d (0,θ ))T )]
 βm m 
πf
m a b

δ m = β m (1 − d m )   
  sin (πfd a (0, θ )Tm )e jπf (2 − d a (0,θ ))Tm    
U m* + k ( f ) : The complex conjugate of U m+ k ( f ) .     
[ ]
 
   × E β m e − j 2πfβ m (1− d a (0,θ ))Tm   
   ETm    
T : Statistical expectation of switching period T. − sin (πfd b (0, θ )Tm )e π ( − ( θ ) )  
  
j f 2 d b 0 , T m 

 
Re (.): Real part of the expression between brackets.   
  

× E β m e [
− j 2πfβ m (1− d b (0,θ ))Tm
] 
  
 
  
In DC-DC, the duty cycle d is constant, then the
switching signal is wide sense stationary and a closed form
  Eθ 
N   a ( (
sin πfd k T , θ T
m+k) ) e jπfd a ( k T ,θ )Tm + k   
  


+
4
Re ∑  
   π β ( − ( θ )) m+ k 
  
× E β m + k e
j 2 f m+ k 1 d k T , T
of the PSD expression can be set easily [1], [7], [8], [10]. a
  
T (πf )2  k =1
× E 

 
( ( ) )
In DC-AC, for each of the three phases, the duty cycle   T m+ k 
   − sin πfd b k T , θ Tm + k e jπfd b ( k T ,θ )Tm + k    
dm depends on the sampling instant tm (i.e. the beginning of       
   × E  j 2πfβ m + k (1− d b (k T ,θ ))Tm + k    
the mth switching period) and the switching function can’t     

β m + k e
 
    
  
be considered as a Wide Sense Stationary signal (WSS),
[4], [5], [8], [9]. To obtain a WSS signal, the duty cycle da  × ET e [
j 2πfT k −1
]  
 
[ ]
−1
needs to be stationarized as follows:  × 1 − E e j 2πfT  N 
  T  
   
d am = d a (tm ) = (1 + ra (t 'm ,θ ))
1
(14) f>0 (17)
2
Where: t 'm = (m − 1)T and θ: random variable between 0 and Although the expression (17) of the PSD is more
complicated, the two expectations on T and β are still well
2π, it represents the total uncertainty about the time
separated.
reference of the modulating signal ra(t).
C. Particular cases: RCFM and RPPM schemes PSD [dB/Hz]
20
RPPM
For the two simple schemes, the expressions of PSDs can RCFM
RCFM-RPPM
be deduced directly from expressions (16) and (17)
respectively as follows: 0

• RPPM scheme: fixed period T and randomized fall time


report β.
• RCFM scheme: randomized period T and fixed fall time -20
report (β = 0.5).

IV. ANALYSIS OF THE PSD -40


0 50 100 f/f1 150
The PSD analysis is performed using the following
modulation parameters: a. phase to ground voltage
PSD [dB/Hz]
• Random triangular carrier: 20
RPPM
a. For RPPM and RCFM-RPPM schemes, the parameter RCFM
β is randomized in the interval:
RCFM-RPPM

  Rβ   Rβ  0
 β 1 − , β 1 + 
  2   2 

β max − β min 1
Where Rβ = and β = . -20

β 2
In practice, the max value of Rβ is: (R β )max = 2 ,
-40
which gives βmin = 0 and βmax = 1. 0 50 100 f/f1 150

b. phase to phase voltage PSD


b. For RCFM and RCFM-RPPM schemes, the period T
Fig. 7. PSD of output voltage
is randomized in the interval:
  RT   R  A. Effect of the randomness level RT
T 1 −  , T 1 + T 
  2   2  Fig. 8. shows a very important effect of RT on spreading
Tmax − Tmin PSD : as RT increases as the PSD flattens and the peaks
Where: RT = and T = 1 pu. decreases. In practice RT doesn’t exceed generally 0.5.
T
PSD [dB/Hz]
• Reference signal: sine signal with amplitude r = 0.5 and 20
Rβ = 0
period T1 = N .T for N = 45. Rβ = 1
• The uniform distribution is used, thus the probabilities of Rβ = 2
0
random variables T and β are:
1 1
PT = and Pβ =
T RT β Rβ -20
Fig.7.a and Fig.7.b show respectively computed PSDs of
phase to ground and phase to phase voltages for the three
schemes (RPPM, RCFM and RCFM-RPPM). Note that in -40
0 50 100 150
f/f1
order to show the peaks, the PSDs are voluntarily shifted.
For RPPM scheme, the PSD contains two parts: a discrete Fig. 8. Effect of RT, (Rβ = 2, N = 45 and r = 0.5)
part with infinite PSD (but finite power harmonics in
“volt2”) and a continuous part in “volt2/Hz”. In the other B. Effect of the randomness level Rβ
side, for RCFM scheme, the PSD is completely spread in For fixed RT, a randomization of parameter β (RCFM-
“volt2/Hz”, which gives less amplitude of peaks than RPPM) adds a significant spreading to the PSD compared
RPPM scheme. to the simple scheme (RCFM). This spreading is also
However, it appears clearly that the proposed scheme is accompanied by a decrease in peaks. Nevertheless, a
more effective on spreading PSD, the only significant peak significant increase in Rβ (Fig. 9) causes an increase in the
is around the switching frequency while RCFM scheme peak of the PSD around Fs = NF1 (Fs = Fc is the average
gives a wavy shape and RPPM scheme gives discrete switching frequency or average carrier frequency, F1 is the
power harmonics. reference frequency) and a decrease in the peak around
2Fs. Knowing that our purpose is to spread best the PSD
and to reduce its peaks (in order to meet EMC standards),
a compromise between the two peaks (at Fs and 2Fs) can
be achieved with an optimal value of Rβ. The optimization
problem is treated in the following paragraph.
20
PSD [dB/Hz] introducing the statistical mean β and the randomness
Rβ = 0 level Rβ, we obtain:
Rβ = 1
Rβ = 2
 Rβ Rβ 
0 β ∈ β 1 − , 1+ 
 2 2 
Where: β = 0.5 et 0 ≤ R β ≤ 2
-20
In expression 18, F (x ) is a vector of dimension m = 2
representing the PSD at frequencies Fs and 2Fs and x
-40 represents the searched value of Rβ. We can then consider
0 50 100 150
f/f1
the problem as an optimization problem in the sense of
Fig. 9. Effect of Rβ, (RT = 0.3, N = 45 and r = 0.5) nonlinear least squares with constraints of the search area
limits as follows:
V. OPTIMIZATION OF PARAMETER Rβ
 Unweighted problem :
Generally, an optimization problem subject to the
constraints of domain boundaries can be written as  Min F ( x ) = W 2
(R β , FS ) + W 2 (R β ,2 FS ) (21)

follows: 0 ≤ R β ≤ 2
 Min F (x ) ∈ ℜ n
 (18) • Weighted problem : Usually the low frequencies are
l ≤ x k ≤ u k = 1....... n more harmful for inductive loads, then we can introduce
weighting coefficients in the objective function so as to
Where:
give more important to the frequency Fs compared to 2Fs
• x : n-dimensional vector representing the parameters of
as follows:
the problem to optimize (xk, k = 1,…,n).
• l and u : Lower and upper limits of the search area (
Min F (x ) = a *W Rβ , FS( ))2 + (b *W (Rβ ,2FS ))2
(domain boundaries).  (22)
0 ≤ Rβ ≤ 2
• ℜ n : Search domain bounded by the limits l and u.
• F (x ) : Objective function or optimization criterion. Where:
• R β : Optimization parameter of the objective function.
Our purpose is to find an optimal value of Rβ
(optimization parameter) in order to achieve the best • a and b: weighting coefficients, in practice the
impedance of an inductive load is proportional to the
spread PSD, while Rβ is subject to the following
frequency, thus its judicious to choose the ratio a/b
constraints of domain boundaries:
inverse of the ratio between the frequencies Fs and 2Fs,
(0 ≤ Rβ ≤ 2 ) (19)
which gives a = 2 and b = 1.
• W : PSD of the phase to phase output voltage evaluated
In our case, the criterion F is chosen with the least at frequencies Fs and 2Fs using the general expression
squares (quadratic). It is defined by the sum of m non- (17):
linear functions squares. Thus, the resolution of a least
Note: In the expression (17) of PSD, the frequency f takes
square optimization problem consists of searching the
only two values (Fs and 2Fs) and the variable Rβ is
vector x* ∈ ℜ n which minimizes the objective function as introduced while replacing the expectations related to the
follows, [12]: random variable β, (ET ,β [..]) by their expressions.
m
minn F (x ), F (x ) = f (x ) 2 = ∑ f i 2 (x ) m ≥ n
2
(20) B. Optimization methods
x∈ℜ i =1
To solve this optimization problem, we propose two
Where: algorithms: the trust region method, available in MATLAB
f ( x) = [ f1 (x ) f 2 (x )............... f m (x )]T : Vector containing as the utility lsqnonlin and to consolidate the results, we
m non-linear functions. use the simplex algorithm (Nelder and Mead) available in
f (x )
MATLAB as the utility fminsearch. These two algorithms
: Second order norm of vector f (x) .
2 are well suited to our problem, which is nonlinear with the
x ∈ ℜ n : Vector of optimization parameters. constraints of the domain limits.

A. Mathematical formulation B1. Trust region method

The parameter β is randomized in the interval This method simplifies the resolution of the problem
posed in equation (18) by solving a series of simpler sub
[β min , β max ] , with βmin ≥ 0 and βmax ≤ 1 . By problems, where the optimization criterion F is replaced by
a simpler function m called model function, which reflects
the behavior of F in a neighborhood N around the point x,
this neighborhood called trust region, is generally spherical Using the proposed modulating principle, a general
or ellipsoidal [13]. Originally, this algorithm is used in analytical model of Power Spectral Density is established
nonlinear optimization problems for both cases: without for both phase to ground and phase to phase voltages.
constraints and with constraints of the domain limits [14]. PSD analysis shows that the dual randomized
modulation realizes a most effective spreading of output
B.2. Simplex Algorithm voltage PSD. Moreover, a parametric study reveals the
This algorithm is based on the concept of direct search: existence of optimal statistical parameters. This problem
it attempts to solve the problem by using directly the has been treated with two powerful nonlinear methods.
objective function without its derivatives. It is suitable for
strongly nonlinear optimization problems without REFERENCES
constraints [15]. In addition, it may take into account the [1] Tse K. K., Henry Shu - hung, Hui S. Y. R. and So H. C., “A
comparative investigation on the use of random modulation
domain boundaries. A simplex is a geometric figure of n schemes for DC/DC converters”, IEEE Trans. On Ind. Electron.
dimensions, represented by means of (n + 1) points; each vol. 47, N0. 2, pp 253-263, April 2000.
dimension corresponds to a parameter of the optimization [2] G.A. Covic and J.T.Boys, Noise quiting with random PWM AC
drives, IEE Proc. Electr. Power. Appl., vol. 145, N0. 1, Jan. 1998,
problem. The minimum is sought by changing the simplex
pp. 1-10.
using standard operations: reflection, expansion and [3] Andrzej M. Trzynadlowski, “Active Attenuation of Electro
contraction, [16]. magnetic Noise in an Inverter-Fed Automotive Electric Drive
System”, IEEE Trans. On Power Electronics, vol. 21, N0. 3, May
2006, pp. 693-700.
C. Numerical results of optimization [4] Lynn Kirlin R., Michael M. Bech and Andrzej M. Trzynadlowski,
“Analysis of power and power spectral density in PWM inverters
For fixed RT , we search the optimal value of R β in the
with randomized switching frequency”, IEEE Trans. On Ind.
interval [0, 2] with an arbitrary initial value (Rβ0 ∈ [0, 2]) . Electron. vol. 49, N0. 2, pp 486-499, April 2002.
[5] Liaw C. M., Lin Y. M., Wu C. H. and Hwu K. I., “Analysis, design,
The obtained results are given in tables 1 and 2. These and implementation of a random frequency PWM inverter”, IEEE
results remain unchanged for any value of R β 0 , Trans. On power Electron. vol. 15, N0. 5, pp 843-854, Sept. 2000.
[6] Bech M. M., Pedersen J. K., Blaabjerg F. and Trzynadlowski A.
demonstrating the robustness of the proposed algorithms. M., “A methodology for true comparison of analytical and
Finally it’s important to note that the two algorithms give measured frequency domain spectra in random PWM converters”,
IEEE Trans. On Power Electron. vol. 14, N0. 3, pp 578-586, May
almost the same values (difference less than 0.1%), 1999.
reinforcing the obtained results. [7] El Khamlichi Drissi K., Luck P. C. K., Wang B. and Fontaine J., “A
novel dual-randomization PWM scheme for power converters”, in
TABLE 1: OPTIMAL PARAMETER Rβ, (r = 0.8, a = b = 1) proc. IEEE PESC’03, 2003, pp 480-484.
[8] N. Boudjerda, Réduction des perturbations conduites dans les
RT Rβopt Opt. meth.
convertisseurs de l’électronique de puissance par une commande en
0.2 1.3888 MLI aléatoire, PhD Thesis, University of Setif Jul. 2007, Algeria.
Trust region [9] N. Boudjerda, M. Melit, B. Nekhoul, K. El Khamlichi Drissi and
0.3 1.2328 K. Kerroum, “Reduction of Conducted Perturbations in DC-AC
method
Converters by a Dual Randomization of Hybrid Space Vector
0.4 1.2436
Modulation”, International Review of Electrical Engineering, vol.
0.2 1.3888 1, N0. 1, pp.154-161, March-April 2006.
Simplex [10] N. Boudjerda, M. Melit, B. Nekhoul, K. El Khamlichi Drissi and K.
0.3 1.2327 Kerroum, “Reduction of Conducted Perturbations in DC-DC
algorithm
0.4 1.2436 Voltage Converters by a Dual Randomized PWM scheme”, journal
of communications software and systems (JCOMSS), Vol.5, N0. 1.
[11] D. Middleton, Introduction to Statistical Communica--tion
TABLE 1: OPTIMAL PARAMETER Rβ, (r = 0.8, a =2, b = 1) Theory, IEEE Press, 1996.
[12] Dennis, J.E., Jr., "Nonlinear Least-Squares", State of the Art in
RT Rβopt Opt. meth. Numerical Analysis, ed. D. Jacobs, Academic Press, 1977, pp.
269–312.
0.2 1.1584
[13] J. F. Rodriguez, J. E. Renaud, B. A. Wujek and R. V. Tappeta,
Trust region
0.3 1.0430 "trust region model management in multidisciplinary design
method optimization", Journal of Computational and Applied
0.4 1.0905 Mathematics, Vol.124, Issues 1-2, December 2000, pp.139-154.
[14] T. F. Coleman, T.F. and Y. Li, "An Interior, Trust Region Approach
0.2 1.1584
for Nonlinear Minimization Subject to Bounds", SIAM Journal on
Simplex
0.3 1.0430 Optimization, Vol. 6, 1996, pp. 418–445.
algorithm [15] R. M. Lewis, V. Torczon and M. W. Trosset, "Direct search
0.4 1.0906 methods: then and now", Journal of Computational and Applied
Mathematics, Vol 124, Issues 1-2, December 2000, pp.191-207.
[16] J.A. Nelder, R. Mead, "A simplex method for function
VI. CONCLUSION minimization", Computer Journal, vol.7, 1965, pp.308-312.
The purpose of this work is the reduction of conducted
EMI in DC-AC converters by an optimized dual RPWM
technique. A modulating principle based on a carrier with
two randomized parameters is proposed for both simple
and dual RPWM techniques. This principle is also
available for DC-DC converters.

Das könnte Ihnen auch gefallen