Sie sind auf Seite 1von 8

Frequency Domain Analysis of Three Phase Linear Current Regulators

D.N. Zmood, D. G. Holmes and G. Bode Department of Electrical and Computer Systems Engineering Monash University Wellington Road, Clayton, 3 168 Australia Abstract - Stationary frame linear PI current regulators are conventionally regarded as unsatisfactory for AC systems because they cannot eliminate steady state errors. Consequently, synchronous frame regulators are perceived to be superior, since they achieve zero steady state error by acting on DC signals in a rotating frame of reference. However, a synchronous frame regulator is more complex, and requires in particular a way of transforming a measured stationary frame o rotating frame DC quantities, and AC current (or error) t transforming the resultant control action back to the stationary frame for implementation. This paper presents a technique for interpreting the stationary/rotating frame transformations as modulation processes in the Laplace domain which move the control function from one part of the frequency spectrum to another. The technique is used to compare stationary and synchronous frame PI regulators on a common basis to better understand the advantages of a synchronous frame regulator, and then to develop a new form of stationary frame resonant regulator which achieves zero steady state error without requiring the complex transformations of a synchronous frame regulator. The performance of this new regulator is evaluated and found to be equivalent to that of the synchronous frame PI regulator.
1. INTRODUC~ON

the relevant control information from one part of the frequency domain to another, using techniques based on Carrier Servo Control [ 5 ] , [6]. This approach allows transformations between stationary and synchronous frame regulators to be done using Laplace transforms in the frequency domain, so that regulators developed in one domain can be quickly and easily transformed into the other domain for comparison and performance assessment. The technique is illustrated using the stationary and synchronous frame PI regulators first proposed by [2], and is then applied to develop a new stationary frame resonant regulator which achieves a performance very close to that of an optimum synchronous frame DQ regulator.
11. DEVELDPMENT OF FREQUENCY DOMAIN TRANSFORMATIONS.

A. Frequency Domain Stationary-SynchronousTransformation of a DQ Axis Current Regulator. Figure 1 shows the structure of a conventional three phase linear stationary frame PI current regulator, where the ABC three phase to stationary DQ transformations are defined by:

The regulation of load current is an important issue for power electronic converters, particularly for applications such as high performance motor drives and boost type PWM rectifiers. Over the last few decades, three major classes of regulator have evolved, being hysteresis regulators, linear PI regulators and predictive deadbeat regulators [ 11. These classes can be further divided into stationary and synchronous d-q reference frame implementations by applying AC machine rotating field theory [23 131 [4]. Stationary frame PI regulators are conventionally regarded as unsatisfactory for AC systems because of the supposedly unavoidable steady state amplitude and phase errors that they create. In contrast, synchronous frame regulators act on DC signals and can therefore be designed to achieve zero steady state error. However, a synchronous frame regulator is more complex, and requires in particular a means of transforming a measured stationary frame AC current (or error) to rotating frame DC quantities, and transforming the resultant control action back to the stationary frame for implementation. In this paper, the stationary/rotating frame transformations are viewed as modulation processes which move

[x,, (I)] = ABC frame time varying quantities.

Eqd~)] = stationary DQ frame time varying quantities.


(t

(note that x, (t) = -xu (i)- x, (t) and is therefore redundant). This stationary frame regulator becomes the well known synchronous frame regulator [2], [4]if the stationary DQ frame variables are transformed in the time domain to the synchronous frame before the PI control action occurs, using:

Figure 1 : DQ frame linear PI current regulation control system.

0-7803-5589-X/99/$10.00 0 1999 IEEE

818

A full derivation of this transformation is provided in Appendix A for reference.

Lqde(t)] = synchronous DQ frame time varying quantities.


Note also the reverse transformation of
(4)

In control system terms, Figure 1 can be described as a two input variable control system with a frequency domain transfer function of v, (SI= HI I (s)x E , (SI+ HI2 ( S ) X E d (5) Vd = H2 I E, (SI+ H22 ( 4 X E d where E,(s) and E d ( s ) are the current errors in the

(4

( 4 .

(4 (4

B. Frequency Domain ABC-Stationary DQ Transfonnation of a Current Regulator To complete the formal theoretical developmenf a frequency domain transformation from the ABC to the stationary DQ frame can also be derived as follows: Equations (1). (2) are linear time invariant transformations between the ABC and stationary DQ frames, under the assumption that the ABC frame is balanced so that line currents and phase voltages sum to zero. Hence by the linearity property of Laplace transforms,
LqdS

(41 = [Cl[I,(Sll
=h,dS G ~ l b q d S ( ~ ~ l h q d S G)1[cl[Eub

(8)

frequency domain. Note that this general form of controller allows cross coupling between the control variables through the offdiagonal terms H,,(s) and H,,(s), as shown in Figure 2. Equation (5) can be transformed into time varying terms by recognising that multiplication in the frequency domain is equivalent to convolution in the time domain, viz:
~ , ( ~ ) = ~ l l ( ~ ) ~ ~ , ( t ) + ~ I 2 ( t ) ~ ~ d ( ~ )
Vd

Using ( 5 ) for the stationary DQ frame and substituting from (8) gives:
kqdSGI1

+ [Cl[V,b( S I ] =

(41

(9)

Left multiplying (9) by [Cr' gives:


Pub

(41 = [cl-'[H,dS (dl [CI[Eob(4 (41= [cl-'b , d S (dl [cl

(10)

(4= h2, ( t ) @e, (t>+h22 ( t b ed (4

(6)

where 8 = convolution operation. More usefully, substituting from (3) and (4) into (3,and after some manipulation, a general frequency domain form of regulator transformation from the stationary frame to the synchronous frame, and vice versa, can be derived as:

Hence the frequency domain transformations between the ABC and stationary DQ frames become:
[Hob

(1 1)
(12)

111. A NEWRESONANT STATIONARY FRAME CURRENT REGULATOR WITH ZERO STEADY STATE ERROR

A. Single Phase Implementation For a DC control system, the standard control network is a PI compensator, which has an s-plane transfer function of

for controllers which have no cross coupling and equal direct path coefficients, ie. H,,(s) = H2(s), HI2(s) = H,,(s) = 0.

This transfer function would also typically be used in a synchronous frame current regulation system. For an equivalent single phase stationary frame AC current regulator, (13) can be transformed using the first term of (7b) to create a control network which achieves the same DC control response centred around the target AC control frequency, ie.
L

=K,

+-

2Kp
s2 + W O 2

Figure 2: Frequency domain representation of general cross coupled DQ axis current regulator.

Equation (14) can immediately be seen to be a resonant controller which achieves infinite gain (and hence zero error as a zero order system is assumed) at the resonant frequency. Such a system cannot of course be realised in a physical filter network, but a practical alternative is given by:

819

HAC(S)=

K, +

2K I%lt

s
2

s2 + 2 w m t s + o o

(15)

where a , = integrator low frequency cutoff. H A C ( s )has a frequency response characteristic as shown in Figure 3. The realisation and performance implications of this practical network have been discussed in an earlier paper [7]. The advantage of this approach is that the resultant controller requires much less signal processing and so is less sensitive to noise. Furthermore, the application of this technique to single phase systems is straightforward and undifferentiated from its application to three phase systems. But the approach does have the complication for variable frequency applications that the resonant frequency of the AC compensator must be shifted in accordance with the required output fundamental. However, this is not a major issue for a digital implementation, particularly if the magnitude response is broadened by reducing the resonant circuit Q as discussed in [7].

di,(t) t., ( t ) R - Ldi f-= ( tO ) dt dt Rearranging (16) and (17) gives: -vb(t)+vc(t)-2iC(t)R-~~--

(17)
,

.-

If C v i ( i ) =o then,

B. Extension to Three Phase Current Regulation The new resonant regulator can be directly applied to a three phase system with no neutral connection by recognising that for a linear system in the stationary frame, two independent regulators in either the ABC or the DQ reference frame are all that are required (ie. just like a conventional stationary frame PI current control system). For such systems the single phase regulator can be used for three phase systems by regulating two of the three line currents of the converter, as illustrated in Figure 4. The validity of this can be verified by considering a balanced three phase system in state variable form. By loop analysis and Kirchoff s current law: di ( t ) di ( t ) - vb(t)+v, (t)- 2i, (t)R- 2 ~ - - - i , ( t ) R -L ' = O (16) dt dt

0 i,(t) 1 0 v,(t) (I9) z [ i , ( t d = y [ 0 R][i,(t)l+[O - J [ v , ( t d Hence, each of the two line currents are independently determined by their phase voltage, and the same transfer function gains as used in a single phase system can be used to achieve the same transient and steady state response. One limitation with this scheme that was observed in simulation is that when the VSI saturates, exponentially decaying transients occur, which typically last a cycle or two. Anti-integral-wind-up logic eliminates these transients, but more research is required to fully understand this effect.
d io(t)

1 -R

Iv. COMPARISON OF VARIOUS CURRENT REGULATORS


A.

10'

1o?

1o4

Frequency (rad/sec)

IO'

lo2
Frequency(radlsc)

loJ

1oJ

Figure 3: Bode Plot of AC Compensator from Equation (15). K,=10, K, = 1.f. = 50 Hz, O, = 10 Hz

Comparison of Stationary and Synchronous Frame Current Regulators in the Frequency Domain The transformations developed in Section 2 can now be used to compare a stationary regulator, a synchronous regulator and the new resonant regulator in the frequency domain, to see how their transfer functions differ in a common reference frame. This comparison provides useful insights as to why a synchronous reference frame regulator is superior to a stationary frame regulator, and provides a basis for determining how a stationary frame regulator with zero steady state error can be achieved. Using transformations (7a) and (7b), the frequency domain transfer functions of a stationary frame PI, synchronous frame PI and the new resonant stationary frame regulator can be expressed in both the stationary and synchronous frames as shown in Table 1. These expressions show why a stationary frame PI regulator can never achieve zero error, since its transformed control function in the synchronous frame does not have an integral term. In contrast, the new resonant regulator does transform to include a synchronous

820

frame integrator, and can therefore achieve zero error in the stationary frame. In fact, the only significant difference between the synchronous frame PI and the resonant stationary frame regulator is the cross-coupling terms that exist in the stationary frame for the PI regulator, and in the synchronous frame for the resonant regulator. These terms add to the complexity of regulator implementation when they are present, which suggests that the resonant regulator may have implementation advantages in the stationary frame. Note also that the cross-coupling terms for the resonant regulator are at double the reference frequency because of the frequency modulation process. However, whilst it might be anticipated that transients could excite these double frequency terms, extensive simulation has shown no evidence of any double frequency oscillation or instability. The cross-coupling terms can also affect the dynamic performance of the regulator, but evaluation to date for the new resonant regulator suggests that the influence seems minimal. However, a thorough analysis of the nature of this cross-coupling is left for future work.

[Ypd' (s)]'

[He

)] Lqde

',de

b)]

(21)

the system will have a closed loop response of


[Iqde(S)]= [Ge(s)+ He(s)r' ["(s)]kqd'*(s)]= A(s$qd'*(s)] (22) The steady state response of (22) can be found using the final value theorem for each regulator, and is as follows:
+ KP@o

A=

for the stationary frame PI regulator (derived using MAPLE). ,, and:

B. Verijication of Zero Steady State Error To verify that regulators with a synchronous frame integrator achieve zero steady state error, their steady state response when operating in a closed loop system must be considered. For a power electronic converter, this can be confirmed by investigating the response of the regulator when driving an R-L load, as follows. The synchronous domain frequency response of an R-L load driven by a sinusoidal voltage is given by:

Now; if the voltage is derived from a current regulator with the synchronous frame transfer function of Current Regulator Stationary frame PI

A=[:, for the synchronousframe PI and resonant stationary frame regulators. Clearly from (24), the synchronous frame PI and resonant regulators achieve zero steady error, in contrast to the stationary frame PI regulator. Furthermore, as a point of interest, (23) shows that the steady state errors created by a stationary frame PI regulator derive from the finite gain in the direct regulator path and not from the cross coupling between the phases as is sometimes assumed. In general, any current regulator that is required to achieve zero steady state error must have infinite DC gain in the synchronous reference frame, and this can be readily assessed using the techniques described above.

:]

Stationary frame DQ transfer function

Rotating frame DQ transfer function

S2 +WO

Synchronous frame PI KP + -

K1s
s +WO

- KIWO
s2 +WO

K s + A Kl +-+
s

Resonant stationary frame


Kp

+-

2Kp
s2 + W O
2

0
2Kp KP 2
S2

Kp

KP
s2 +4W0
2

- 2KPo
s2 +4W0
2

0
I

2Kl WO
s2 +4W02
I -

K , +I+ K K1s
s
S2+4W,2

+WO

Table 1 : DQ Stationary and Synchronous Frame Transformations for PI Current Regulators

82 1

C. Validation o f Frequency Domain Transformation against Time Domain Synchronous Frame Regulator From [2] and [4] the time domain form of a synchronous frame DQ current regulator referred to the stationary frame can be expressed as:
Vqds

= XqdS + K p kqds* - i q d s )

(25)

pxqds = K I k q d S * -iqdS)+Re x q d s where:

(26)

Substituting eqdS = iqd -iqds into (25) and (26) and taking the Laplace Transform gives:
qd

S*

xqds

KPEqds

(s)

sxqS (s)= K , E , ~ (s)+o,xds (s) sxdS (s)= K

~ (s)E o,xqS ~ (~ s)
x d s in

Solving for X q s and terms gives:

terms of their respective error

D. Transfonnation to ABC Frame Current Regulators The end point of any current regulator development is to create a regulator structure which achieves zero steady state error and is simple and effective to implement. Since all current regulators have as their measured inputs the ABC phase currents, it make sense to consider whether a regulator can be directly implemented in the stationary ABC frame of reference, to avoid both DQ and synchronous reference transformations. This can be done in either the frequency or time domain, but there is some advantage in using the frequency domain for comparison purposes. Hence using the stationary frame ABC-DQ transformation given in (1 1). the three linear regulators considered in Table 1 can be transformed to the stationary ABC frame as shown in Table 2. Figure 5 shows the ABC frame representation of the synchronous frame PI regulator, which is functionally equivalent to the frequency domain form of the synchronous frame regulator given in Table 2. Note that similar to [2] and [4] this implementation in the stationary frame feeds back the output variables and is a more efficient implementation than the direct forward transfer form shown in Table 2 since the cross coupling terms do not have to be created separately but are a direct consequence of the summations. Even so, the complexity and cross coupling requirements for this implementation compared to the simplicity of the new resonant regulator in the ABC frame are clearly evident.

Substituting (29) back into (27) and putting into matrix form achieves the same result (30) as in Table 1 for the synchronous frame regulator in the stationary frame and hence verifies the validity of the frequency domain transformation.

v. SIMULATION AND EXPERIMENTAL RESULTS


The three current regulators discussed in this paper have been investigated and compared in simulation, using a MATLAB based discrete time step representation of the physical switched inverter system. This simulation gives a very realistic output without the usual simplifying assumptions that are present in many linear simulations, and in particular allows the digital regulator delays of a real system to be incorporated without difficulty. Figure 6 shows the simulated steady state and transient response of the stationary frame, synchronous frame and K p +2K,s
s2 + W O 2

K p +-

[vobl(.)I=

0
Kp + 2

2KIs

s +WO

Staiionary PI regulator in ABCfrnme

New Resonant Regulator in ABC frame

KIs Kp+----

&

b o 6 1(41

KIw,

S2+W02

3 S2+W0 2& K , o -o 3 S2+W0 2


Synchronous PI regulator in ABC frame Table 2: ABC Frame PI Current Regulators

822

Figure 5: Stationary ABC frame current regulator referred f r o m the synchronous DQ rotating frame.

resonant regulators driving a 4 pole permanent magnet motor. The elimination of steady state error for the resonant regulator can be clearly seen, and its transient response is almost identical to that of the synchronous frame regulator. The new resonant PI Regulator has also been experimentally verified in the ABC frame using a TMS320F240 DSP based VSI regulator driving a RL load with parameters as listed in Table 3. The algorithm was implemented by discretising the continuous time transfer function of (15), which is essentially a bandpass filter, into difference equation form. The discretisation method used was the Pre-Warped Tustin Transformation, ie. U z-1 c,(z)=G(S) s =-* (31) Z+I tan(a~/2) where: T = sampling frequency = twice switching frequency, a = frequency at which the distortion is to be minimised. After substituting (31) into (15), the discrete transfer function becomes, b, - b3z-2 HA, (4= 1 + q z - ' + a2z-2 where: b, = b, = 0.0062, a, = 1.9865, a*=0.9875. Finally, this can be implemented in the DSP using fixed point arithmetic as:

-a"

0.01

0.02

sec

0.03

0.04

0.05

-a"

0 20

0.01

0.02

sec

0.03

0.04
I

0.05

Y (n)= b 1 kerror (n)-

'error

(n - 211

+ a,y(n - 1)+ a2y(n - 2)


where y(n) is the output of the transfer function and I,,. the current error.

(33)

is

PERPHASE LOAD Supply Volts Fund. Freq. Regulator Gains

LL I R (simulation)

I7mH

I( I 580Vdc
I 50Hz I 0.125

I
I

I I

I I

-20 0

0.01

0.02

sec

0.03

0.04

0.05

I Resonant

Proportional

Figure 6: Steady state and step response for stationary PI, synchronous PI and resonant + P current regulators. (NOTE: Initial transient is caused by simulation startup initialisation for all regulators and has no significance)

Table 3: Simulation & Experimental System Parameters

823

Figure 7 shows the experimental response for the new resonant plus P regulator, compared against a conventional stationary frame PI regulator. The improvement in steady state performance and elimination of steady state error is clear. Figure 8 shows the transient response for the new regulator, with excellent risetime and minimal overshoot very similar to that predicted by simulation. VI. SUMMARY This paper has presented a method of analysing current regulators in the frequency domain, so-that regulators can be quickly and easily transformed between the stationary and the rotating frame of reference for comparison and performance assessment. The approach has been used to identify the primary reason why a stationary frame regulator cannot eliminate steady state AC error, and is then used to

develop a new resonant stationary frame current regulator which overcomes this limitation. The new regulator offers significant advantages over synchronous frame regulators in terms of complexity and ease of implementation, whilst achieving equivalent transient and steady state performance. VII. REFERENCES
M. P. Kazmierkowski and M. A. Dzieniakowski, Review of Current Regulation Techniques For Three-phase PWM Inverters, Conference Record.lECON 94, pp. 567 - 575, 1994. T. M. Rowan and R. J. Kerkman, A New Synchronous Current Regulator and an Analysis of Current Regulated PWM Inverters, IEEE Transactions on Industry Applications, Vol. IA-22, No. 4, pp. 678 - 690, 1986. Y. Sato, T. Ishizuka. K. Nezu, and T. Kataoka, A New Control Strategy for Voltage-Type PWM Rectifien to Realise Zero SteadyState Control Error in Input Current, IEEE Transactions on Industry Applications, Vol. 34, No. 3, pp. 480 - 486, 1998. D.W. Novomy & T.A. Lipo, Vector Control and Dynamics of AC Drives, Oxford Science Publications, 1996. J.J. DAzm and C.H. Houpis, Feedback and Control System Analysis and Synthe~is,2~ Edn.,McGraw-Hill, 1966. Gordon J. Murphy, Control Engineering, D. Van Nostrand, 1959. D. N. Zmood and D. G. Holmes, Stationary Frame Current Regulation of PWM Inverters with Zero Stteady State Error, Conference Proceedings PESC99, (In Print), 1999.

6 4 2
0

VIII. APPENDIX A

FREQUENCY DOMAIN TRANSFORMATION BEIWEEN THE STATIONARY AND ROTATING DQ FRAMES

A -2
-4

For the purpose of this analysis it is assumed that the reference inputs Zq* and I,* are bandlimited to less than the modulation frequency. From a signal viewpoint if the references contain frequencies greater than the modulation frequency, aliasing will occur and it is not possible to exactly reconstruct the reference signals. The practical rational is that the integral term in a DQ PI regulator is mainly significant for the steady state response of the system, ie. the frequency
81
I

-6

8 msec 12

16

20

10

2o msec

30

40

50

Figure 7: Experimental steady state response for stationary PI current regulator and new resonant plus P current regulator

Figure 8: Experimental transient response for new resonant plus P current regulator

824

response of an integrator is a low pass filter characteristic with an infinite gain at DC, so it can be safely assumed that only a narrow band of frequencies are relevant. The general form of a linear current regulator in the time domain is given by, vq ( t )= hl 1 e, A12 (t )e e, (A.1) Vd (t)= h21 (de e, (t)+ h22 edk)

(de

(4

(4@

where: and:

e, (t ) = i,* (t ) - i, (t)

e, (t) = id* (t ) - id (t )
j ( t ) @ g ( t ) = j;f(z)g(t -~)62,(convolution product).

Equations (A.7) are shifted into the frequency domain by taking the Laplace transform of each term, i.e. (in generic form): LbS ( t ) e (t)cos(o,t)]. cos(o,t)}

E?.

= [HS(s)LP (t)cos(o,t)]e-

s2 +WO2

Taking the Laplace transform and assuming all initial conditions are zero, (A. 1) becomes:

= HS

( s t be

(s+ jw, )+ Ee(s - j o ,

The transformation from the stationary to the synchronous DQ frame for an arbitrary quantity is given by: cos(o,t) -sin(o,t) T(o,t) = (A.3) sin(o,t ) cos(o,t ) Hence the rotating frame voltages in terms of the stationary frame voltages can be expressed as: vqe(t) = vqs cos(w,t)- vds sin(o,t)

=-[

H S( s + jo,)E"(s + 2jO,)+ H S ( s - io,)Ee(s)+ 4 HS ( s+j o ,)Ee(s)+ H S ( s- j o ,)E'(s - 2jw,)


1

v,'(r)= vqssin(o,t)+vds cos(o,t)


where:
x' indicates a rotating frame DQ quantity. x' indicates a stationary frame DQ quantity.

Substituting (A.l) into (A.4) gives:


v,'(t>= [hl,'(t)ee,s(t)+h12s(t)eedl(t)]cos(o,t)

- EZls(t)ee,s (t)+ h 2 2(~ t ) e e,' (t)]sin(o,t)


vde(t)= [ ~ l I " ( t ) s e , ~ ( t ) + h l ~ ( r ) e e ~ ~ ( t ) ] s i n ( w , t )

E21~ ( t ) e: e

(t)+ hz2s (t)@edS(t)]cos(o,t)

Using the inverse of (A.3), the error in the rotating frame can be expressed as: e,' (t)= e,' cos(o,t)+ e,' sin(o,t) (A.6) eds(t) = -e,' sin(o,t)+ e," cos(o,t) Substituting (A.6) into (AS) gives:

For the cases considered in this paper, the general expansion of (A.7) using (A.9) is considerably simplified h, ( t ) [e,e.(t)cos(w,t)+e,'(t)sin(w,t)]+ ~ because the regulators under consideration are matched in the cos(wo t) eqe(r)sin(w~r)+e,e(r)cos(~~r)l (A.7a) d-q axis and have no cross coupling terms, so that H , , ( s ) = H , ( s ) and H , * ( S ) = H 2 1 ( s ) = 0 ' h2,' ( t )L,e(t)cos(o,t)+ ~ e,'(t)sin(q,t)]+ sin(w,t) Thus (A.7) simplifies to (7a) given in the main paper for e,e(t)sin(o,t)+ e,e(t)cos(q,t)] the transformation of a stationary DQ current regulator to its equivalent in the rotating reference frame and to (7b) for the reverse process.

1 1

825

Das könnte Ihnen auch gefallen