Synchronisation methods for grid-connected voltage source converters

J.Svensson Abstract: Four differcnt synchronisation methods for a voltage source converter connected to a three-phase grid are investigated. The methods are adapted for use in a digital controller. The performance of the synchronisation methods is studied by response characteristics of phase-shift steps, frequency steps and low-frequency grid voltage harmonics. The low-pass filtering method can be used only if the frequency of the grid is constant and phase jumps do not occur. If phase jumps occur. the novel spacc vector filtering method is recommended. The extended space vector filtering method is adapted to handle frequency variations and is also preferred if fast frequency vanations occur. This method even has a higher performance than the extended Kalman fdter method, in spite of the large number of calculations that must be performed.

List of principal symbols

Introduction

rKl, eK3 = grid phase voltages e,*. = phase voltages in point of common connecel, 02, e3 tion (PCC) = grid voltage in a/koordinates in PCC e,, q? = grid voltage in dq-coordinates in PCC "d. eq = e, +& = grid voltage vector in PCC = grid current in dq-coordinates id, iq = grid current in ap-coordinates i,, $ k = sample index kiS VF = integral gain kPS"F = proportional gain = short-circuit inductance of the grid L6 = series inductance of the grid filter L S = short-circuit resistance of the grid 4 = series resistance of the grid fdter RS TS = sample time X = state vector V = process noise W = sensor noise A s V F BsvF CsvFDsvF = discrete state equation matrices NO) = rotation matrix = process noise covariance matrix = sensor noise covariance matrix W Y = forgetting factor H = transformation angle A0 = transformation angle deviation m n = grid angular frequency = grid angular frequency deviation A% = estimated value

0IEE, 2001
IEE Proreediigs online no. 20010101 DOL lO.l04Y/ipgtd:20010101 ~ n Paper first received 14th J a n w and i revised form 27th August 1998 The author is with the Department of Electric Power Enginering, chalmers Uiliversity of Technology, 412 96 Goteborg Sweden
IEE Proc.-Gmer.
Transm. Distrib.,

In almost all power electronic equipment connected to the grid, a phase-locked loop (PLL) is used to obtain an accurate synchronisation to the grid. The synchronisation algorithm must cope with the existing electrical environment [I], e.g. with harmonics, voltage sags and commutation notches. Another problem is grid frequency variations. In strong grids, the frequency variations are usually small, but larger frequency variations occur in autonomous grids. Furthermore, the synchronisation must handle measurement noise. An accurate zero voltage crossing detection is essential for the control of grid-commutated converters. Otherwise, the converter system can suffer from poor performance or even instability. The firing angle unit for thyristor valves uses the zero voltage crossing as a reference; the zero voltage crossing must be accurately detected each half period of the grid [2]. A more demanding application is the grid connection of a forced-commutated voltage source converter (VSC) when vector current control is used [3]. The synchronisation must be updated not just at zero voltage crossings, but continuously under the whole period. In this case, a flux-based transformation angle detector (TAD) is used to obtain the rotating dq-coordinate system [3]. The classical synchronisator is an analogue PLL [4], designed around the IC c h p 4046. Today, control systems of converters are often implemented by a computer. By using a software solution to the PLL, a flexible and cost effective solution is obtained. The analogue PLL function can be translated into a software PLL function [5], but the low update frequency from zero voltage detection is a disadvantage. Digital filters and controllers can, however, increase performance and make new methods feasible. If the input signal is noisy, a predictive digital filter can be used to find the zero voltage crossings for the PLL [6]. The calculation time is critical for a real-time control system, which results in a demand for fast algorithms. In this paper, four synchronisation methods will be analysed. The complexity of the methods varies from a simple method adapted for stiff-grid applications to the
229

Vol. 148, No. 3, Ma~v 2001

extended Kalman filter (EKF) and a novel space vector filter (SVF), which can handle both phase angle steps and frequency steps. The performance of the synchronisation methods will be analysed based on three criteria: phase shift response, frequency slep response and sensitivity to grid voltage harmonics. The maximum allowed frequency change is lHzis in accordance with the IEC 1800-3. However, a good understanding of the perfoimance is obtained by using frequency stcp responses.

TAD based on low-pass filtering of grid voltage

VSC system connected to weak grid

When a VSC is connected to a weak grid, the voltage at the point of common connection (PCC) will vary and become a function of several variables: the grid voltagc behind the impedance describing the grid, the grid current and the short-circuit parameters of the gnd. A step in the power from the VSC (or another user of the grid at the PCC) results in a step change of the voltage vector at the PCC. The VSC connected to a weak grid and the vector currcnt controller are shown in Fig. 1. The transformation angle e(t) i s used to transform grid currents and grid voltages needed in the vector current controller into the dq-coordinate system. The d-axis of the rotating dqcoordinate system is synchronised with the grid flux vector. The reference voltages from the current controller are transformed back to three-phase values. The synchronisation must have a high perfoimance due to the structure of the vector current controllcr. A phase shift in the synchronisation will result in steady-state errors, and the cross-coupling current gain will increase. In addition, poor performance or instability can occur if the synchronisation i s sensitive to noise or reacts too distinctly to phase steps.
w w k grid
PCC

A simple mcthod is obtained by low-pass filtering (LP filtering) the grid voltage vector, and the method is dcnoted by LP-TAD. A first-order Buttenvorth filter is utilised. The cut-off frequency is tuned to betwecn 0.1 and 25Hz. The phase lag of a first-ordcr filter depends on the grid frequency and the cut-off frequency. The desired phase lag of 90" occurs at an infinite frequency. The phase lag will, thus, be less than Yo" due to the grid frequency of 50 or 60Hz. The phase lag increases when the cut-off frequency decreases. By using a coordinate transformator, the LPfdtered grid voltage vector can be rotated by an angle A 0 so that thc phase lag becomes 90". Fig. 2 shows a diagram of the LP-TAD. Due to the extensive calculation time of trigonometric functions of the DSP, the transformation angle itself is not used. The sine and cosine of the transformation angle are obtained by dividing the LP-filtered a- and /3-voltage components by the length of the LP-filtered voltage vector. The correct phase lag is obtained by using the rotation inatrix R. Two calculated step responses to a lo" phase-shift in the grid voltage vcctor are prcsented in Fig. 3. Thc grid frequency is 50Hz and the cut-off frequency is 0.5Hz or 5.OHz. Coordinate transformation is not used to correct the phase-shift. Different stcady-state angle errors, due to different cut-off frequencies, can be observed. The phaseshift errors are approximately 1" and 6" for cut-off frequencies 0.5 and 5Hz: respectively. Unfortunately, the detector characteristics arc undamped and oscillate. Thc damping is increased by a higher cut-off frequency. This behaviour makes the detector unsuitable for applications where phase jumps can occur. In practice, every grid-connected VSC is exposed to grid phase jumps due to short circuits in other nodes of the grid [7].The LP-TAD is also sensitive to grid lrequency deviations.

vsc

Fig.2
230
IEE Pro<:.-Gmrr. Transin. /lislr;b., Vol. 148, h'o. 3, May 2001

301 I

LP-TAD phase shift

where the matrices are given by

Ol 0

I'
0.2
time,s a

0.L

0.6

LP-TAD phase shift

J
P 20U

.L

If thc forgetting factor y = 1, the model has no connection to the inputs of the filter, and if the forgetting factor is 0, the output of the filter will be thc grid voltage. A scheme of the SVF-TAD is shown in Fig. 4. The SVF does not introduce any phasc shift in the steady state. Therefore, the scaled outputs from the SVF are rotated by 90". To determine TAD performance, two tests have been made. The first one is a step response to a phase-shift of thc grid voltage vector. The second test evaluates the influence of grid voltage harmonics on the filter.

.c I n

91

4.7
10grid phase shift

Phase shift step response

'L

Fig. 3
U

0.4 0.6 time,s b LI'-TAD phose shift respoiise 10 (I gridphuse shjli s k p

0.2

The step response to a 10" phasc shift of the grid voltage vector is shown in Fig. 5. The forgetting factor has values between 0.9 and 0.9995. The sample time is 200p and the grid frequency is 50Hz. The SVF-TAD does not introducc any phase shift, and thc step response is well damped. A high value of the forgetting factor results in a long settling time, which is preferable becausc the time constant should bc largcr than the other time constants in the system.

h Cut-off fi-cquencyof 5 z H

Cot-oflfrequcncy of 0.5Hz

4 TAD based on space vector filtering for constant grid frequency

151 grid phase shift

The novel space vector filter (SVF) is an LP-filter for space vectors; it is based on the fact that the fundamental a/$ components of the grid voltage depend on each other. These voltages can bc cxpressed as a rotating voltage vector with a constant amplitude and constant frequency. Thc filter has two inputs, e,(t) and e&t). The filter uses a model of thc grid voltage vector esvFu + je,,Fo, which is updated with every sample. By assuming a constant sample time and a constant grid frequency, the grid voltage vector of the model can be estimated for the next sample. The value of the forgetting factor y determines how much influence the grid voltage vector has on the model of the grid voltThe age. The outputs are efilia(t) and efiIlfi(t). filter equations in stale variable fom'become

Y =0.9995

0.4

0.8

1.2

4.2 Influence of grid voltage harmonics The gain of the SVF-TAD for grid voltage hannonics is dctcrmined by two siniulations performed by injecting a 10% voltage harmonic component of the 5th or 1lth order. The grid frequency is 50Hz and the sample time is 2 0 0 ~ . The gain from thc input to the output or the SVF for the

Fig.4
231

harmonic voltage components is displayed in Fig. 6a. The forgetting factor varies between 0.9 and 0.999. The gain of the voltage harmonics decreases as the forgetting factor is reduced.

5 TAD based on space vector filtering for varying grid frequency

C .-

cn 0

0.001

0.001

Y
4\\

0.01

1-v

0 .I

8
0)

QI

10'-

O .L

F
c

10-1 -

o u t p u t SVF (5th harmonic)

.s 'c

w
c n

(11th harmonic) 5 -1 0 0.001 0.01

0.I

1-Y b
a Gain from harmonic grid vollage component to harmonic voltage component after SVF h Variances of grid phase shift and SVP-TAD phase shift

Fig.6 Voltage component undphare shfi

The variance of the grid angle deviation displays how much voltage harmonics affect the SVF. Fig. 66 shows this variance when the forgetting factor varies between 0.9 and 0.999. The variance of the SVF output angle deviation is lower than O.0lo2when the forgetting factor is higher than 0.99. The SVF handles the distorted grid successfully. The 5th and 7th voltage harmonics have the same gain factor (and the same is valid for the 11th and 13th harmonics) due to the use of the rotating coordinate system.
( vector I

The SVF-TAD can be adapted to a varying grid frequency and is then called extended SVF-TAD. A phase shift occurs when the grid frequency varies and the forgetting factor of the SVF is high. It is complicated to determine thc phase shift directly. However, the phase shift will be proportional to the vector product of the grid Voltage vector and the output vol@e vector of the SVF for small deviations [8]. If the frequency of the SVF is highcr than the grid frequency, the vector product becomes ncgativc. The vector product is called the 'Q-vector' referring to the reactive power. To reduce the amplitude dependcnce of the Q-vector, the involved voltage vectors are normalised. The Q-vector is fed to a discrete PI-regulator, which is connected to the SVF, as shown in Fig. 7. The PI-regulator output is the estimated frequency deviation. The estimated frequency is the sum of the rated frequency and the estimated frequency deviation. Compared with the basic SVF-TAD, the proposed nonlinear filter for fluctuating frequency is difficult to analyse. Several parameters must be determined to get a good filter performance: the forgetting factor A, as well as the proportional gain lcpsvF and the intcgrating gain kIsrfF of the discrete PI-regulator. To minimise the influence of the voltage haimonics of the grid, a 150Hz first-order LPfilter is placed between the Q-vector block and the discrete PI-regulator. The problem of tuning the discrete PI-regulator is the conflict between frequency tracking and sensitivity to voltage harmonics. Another problem is the parallel operation of the frequency regulator and the SVF. If the phase displacenient is large, the frequency regulator responds to a frequency error. To reduce this problem, the frequency regulator is tuned to respond slowly. The time constant of the PI-regulator is set to one grid period. Two simulations are prescnted in the following: the phase-shift step response and the frequency step response. The simulation parameters, which have been selected by an iterative process to decouple the SVF and the frequency regulator, are listed in Table 1.
Table 1: Parametersof extended SVF-TAD
Ts = 2oops
y = 0.99

kpsvF = -40 k,svF = -0.04

5.I

Phase shift step response

At the simulation time 0.2s, the grid voltage vector makes a positive phase shift step of 10". Fig. Sa shows the phase

tt

t t'

frequency

Fig.7 Diagram o extenrlerlSVF-TAD f

232

shift of the extended SVF-TAD. The estimated grid frequency deviation is shown in Fig. 8h. The phase shift out from the extended SVF-TAD has an overshoot when the estimated grid frequcncy deviation increases. This overshoot is caused by the integrating part of the PI-regulator; the integrated part cannot decrease to zero immediately. The estimated grid frcqucncy deviation responds by a step of 0.1 1Hz, after which it decreases monotonically to zero.

+I

'U
2

1 time,s a

1 time.s

- 0.5 I I I
0

1 time,s
b

bFrequeiicy de6ation response Grid frequency steps from 50 OHz to 52 5Hz at 0.2s

-00 5.0
1

TAD based on extended Kalman filter

time,s
b

Fi ,8 Extetukd SVF-TAD plum .shfi u??d.f?equenq deviution response lo grjphure .shift

Frequency deviation rcsponsc 11 Phase shirt rcsponse Phasz shiIt stcp is 10" at 0.2s
n

5.2 Frequency step response

The simulation results of ii frequency step from 50.OHz to 52.5Hz at 0.2s are examined. The step and the estimated gnd frequency deviation are shown in Fig. 9a, and the phase shift of the extended SVF-TAD due to the frequency step is shown in Fig. 9h. Due to the slow frequency regulator, the phase shift becomes -14" before it is decreased by the frequency regulator. The phase shift of the extended SVF-TAD tracks the real grid phase shift snioothly without any overshoots. By choosing a faster frcqucncy regulator, the maximum phase shift will decrease, although the frequency regulator will also be more sensitive to the harmonics of the grid voltage.

The EKF used as a TAD (EKF-TAD) tracks the amplitude, frequency, and the grid voltage angle inslantaneously. The EKF is extensively used in power clectronic applications to estimate parameters and variables in drive systems. In grid applications, the EKF has been used, for instance, to detect grid voltage harmonics 19, 101 and to fault relaying [ 1 I]. The use of the EKF as a grid synchroniser is not a common application.

6. I State equation model of grid voltage

The state vector contains three elements: the amplitude, the angle and the frequency of the grid voltage vector, i.e. x(k) = [eq@)q k ) ru;,(k)lT. voltage model of the grid in Cwp The coordinate system becomes
I O 0

z ( k )+ v ( k )

(3)

5.3 The influence of grid voltage harmonics

The influence of grid voltage hannonics on the extcnded SVF-TAD is minor when the frequency regulator parameters are chosen so that the regulator responds slowly. The variance of the grid voltage angle fluctuation is 16.46"* when a 10% 5th voltage harmonic component is injected to the grid. Thc filtered variance is 0.012"', i.e. the influence of voltage harmonics has been practically eliminated.
I E E Proc.-Gerier. Transin. Disfrih., Voi. I48, Nu. 3, May 7UOl

where v and w are the process noise and sensor noise, respectively. The state equation is nonlinear. The EKF [I21 is used to estimate the states of the grid in the model above. To change the performance of the filter, the sample time as well as the process and sensor noise covariance matrices are altered.
233

6.2 Simulations of EKF- TAD

In the following, the results of two simulations will be presented: the phase shift step response and the frequency step response. The grid voltage amplitudc e&t) has a constant value of 400V and the grid angular frequency w&t) is 100n radls. The sample time is 1 0 0 ~The process and measure . covariance matrices are set to

V=

0.0400 0 0

0 0 0.0006 0 0 0.0314

analysed by means of a positive phase shift step of 10" at 0.2 s. The EKF-TAD phase shift response to a grid phase shift step, and the step, are shown in Fig. lla. The EKFTAD frequency deviation is shown in Fig. 1IO. The phase shift slep response of the EKF-TAD has an overshoot of 20Y0but is well damped. The frequency deviation increases to a maximmi of 0.23Hz before it decreases to 0Hz.

W=

[4ooo
0

4000

by using the trial and error method. The initial state vector is i ( 0 ) = [l.le, 0 l . l d g ] T

6.3 Frequency step response

The frequency deviation and the phase-shift responses of the EKF-TAD to a grid frequency step are analysed by means of a frequcncy step from 50Hz to 52.5Hz at 0.2s. The grid frequency step and the estimated frcquency deviation are shown in Fig. 10cr. The estimated frequency devialion response has an overshoot of 0.2Hz but is well damped. The phase shift of thc EKF-TAD due to the frequency step is shown in Fig. lob. The phase shift decreases to -22" before it starts lo increase. The phase shift settles to 0" after a small overshoot of 1".

0.L time,s a

0.8

1.2

-0.11 0

,
1.2

0.L

0.8

.step
time,s

Fig.11

time,s b EKF-TAD phaw ,dfi u d fiequmcy deviation lo g d phure shifi

3r

&ha% shift xsponse IiErequency deviation response Phase shift step i 10" at 0.2s s

6.5 Influence of grid voltage harmonics

The influence of grid voltage harmonics on the EKF-TAD is small when the frequency regulator parameters are chosen so that the regulator responds slowly to variations. For a 10% 5th voltage harmonic coniponent injected into the grid voltage the variance of the grid voltagc angle fluctuation is 16.46"2.The variance of the phase shift of the EKFTAD is 0.013"*,which is negligible.

2
c

grid frequency deviation

e -

U 3

E K F - TA D frequency deviation

6.6 Simulation experience from EKF-TAD

-1

0.4
time,s b

0.8

1.2

quenc-v deviation step Fi-cqucncy dcviatiuii responsc h Phase shift response Gild fi-equcncysteps from 50.OHz to 52.5Hz at 0.2s

Fig. 10 EKF-TAD phase ,shqi DildfkqmCJ~ rlevhtiuti rapotare io pd.fu,-

6.4 Phase shift step response

The phase shift step response of the estimated grid frequency and of the phase shift of the EKF-TAD are
234

The EKF cstimates the frequency, the angle and the amplitude successfully. Depending on what kind of disturbances occur in the grid, the weighting matrices must be changed so that the EKF operates adequately. The bandwidth of the estimated states decreases if the grid voltage vector is noisy. Low-frequency harmonics have frequencies near or in the active frequency band of the system, which requires larger weighting matrice elements to reduce the noise. The EKF-TAD tracks the amplitude and frequency steps with an overshoot response. One problem of using the EKF is finding the optimal weighting matrices V and W. The method used was bascd on trial and error; other weighting
IEE Proc.-Genw Tumsm. Dlwib., Vol 148, h'o. 3, May 2001

matrices V and W might have resulted in a slightly better performance of the system. Another drawback is the calculation time of the EKF algorithm. If a high sample frequency is required, problems will be encountered due to the extensive time consumption of thc EKF-algorithm.
7

TAD algorithm implementation in DSP

The TAD algorithms have been implemented in an experimental system to detcrminc the exaution times. The synchronisation algorithms are implemented in a TMS320C30 DSP using 40MHz clock frequency. The algorithms are written in C. At every sample, the main interrupt function calls thc synchronisation function. In Table 2, the execution tinic is shown for the four synchronisation functions. Three of them were implemented in the DSP. A thrce-phase systcm was creatcd internally in the DSP to simulate the phase jumps and frequency steps, due to difficulties in making frequency and phase steps in the grid. The implemented TADs operate as expected from the analysis. The calculation tinie of the EK F-TAD algorithm was estimated due to extensive calculation tinie.
Table 2: Execution time of TAD algorithms
TAD type LP-TAD SVF-TAD
Extended SVF-TAD

The extended SVF-TAD also handles frequency variations and frequency steps. The performance of the method is almost equal to that of the basic SVF-TAD. The voltage distortion attenuation is slightly lower than that of the basic SVF-TAD, and the phase-shift step response has an overshoot of approxinyately 5%. The method is suited to weak grids with phase jumps and voltage distortion. Furthermore, the method can be used in autonomous grids, where frequency variations occur. The execution time i s longer in coinparison to the basic SVF-TAD, but is still acceptably low. The EKF-TAD can operate in a weak grid where the frequency and the phase shirt step and fluctuate. Grid voltage distortion is well attenuated. Compared with the extended SVF-TAD, the EKF-TAD has a lower performance, even if thc sample time is shorter. The EKF-algorithm rcsults in a long execution time, and thus, is not suited to high sample frequencies.
9
References

Time
10.0ps

9.6~s
25us

EKF-TAD

75us

Conclusion

Four synchronisation methods with different complexities have pbeen analysed. The synchronisation methods for a VSC connected to a weak grid are compared with using phasc and frcquency steps and fluctuations. Furthermore, the immunity to low-frequency voltage haiinonics in the grid is addressed. The TAD based on low-pass Butterworth filtering of grid voltage works adequately in a distorted grid; problems occur wlien phase shifts and frequency variations are introduced. The algorithm is highly time-efficicnt. The basic SVF-TAD for a fixed and known grid frequency can deal with phase-shirt steps and works adequately in distorted grids. The method has a short execution time and is well suited to thc synchronisation of VSCs to distorted grids where phase jumps can occur.

1 Voltage charactcrislics of electricity supplied by public distribution systems. European Standard EN50160, CENELEC. 1994 2 WHDENBRUG, K., DAWSON, F.P.; and BONERT, R.: New synchronisation method for thyristor powcr convcrtei-s to weak AC systems, IEEE Tn Ind E/ecrr.on., 1993, 40;(5), pp. 505-51 1 3 SVENSSON; J.: Inclusion of dead-timc and paramctcr variations in VSC modclling for predicting responses of mid voltage haiinonics. Proceedings of 7th European conference on Punw elrclrunics und rppliculiurw (EPE97), Truiidheim, Norway, 1997. Vol. 3. pp. 216221 4 HSIEH, C.C., and HUNG, J.C.: Phase-locked loop techniques A survey, IEEE 7?l.cm.r. hid rori., 1996, 43, (6), pp. 609-6 I5 5 EL-AMAWY, A.A., and MIKBOD, A.: An etficieiit software-controlled PLL for low-frcq tiency application, IEEE Truns. fnd E l m . fWJFZ., I%%, @), pp. 3 4 1 - 3 ~ 35, 6 VAINIO, O., and OVASKA. S.J.: Noise reduction in zero crossing detection by predictive digital filtering, IEEE Truns.I d Eleclrun., 1995, 42, (I), pp. 5 5 4 2 7 BOLLEN, M.H.J.: East assessment methods for voltage sags in distribution systems, IEEE Truns. b d Appl, 1996, 32, (6), pp. 14141423 8 JUNG,G.H., CHO,G.C., HONG,S.W.. and CH0,G.H.: DSP based control of high power static VAR compensator using novel vmtor product phase locked loop. Proceedings of 27th Annual IEEE Power clectroilics specialists conference (PESCY7), Baveno, Italy, 1996, pp. 238-243 9 MA, H., and GIRGIS; AA.: Identificationand tracking of harmonic sources in a puwcr system using a Kalinnan filter. Proceedings of IEEE /PES winter meeting, Baltimore. MD, 1996, Paper 96WM086YIWKD 10 CIRGTS. A.A., CHANG, W.B., and MAKRAM, E.B.: A digital recursive measurenient scheme Cor on-line tracking of powcr syslciu harmonics, IEEE Truti.?.POIVO. Deb, 1991, 6, (3), pp. 1153-1 160 11 MURTY, Y.V.V.S., and SMOWNSKI, W.J.: A Kalniaii filter bascd digital percentage ditkential and ground h l t relay for a 3-pbasc power transfonner. IEEE Trms Poower Deb., 1990, 5, (3). pp. 129% 1306 12 SODERSTROM. T.: Discrete-time stochastic systems cstimation and control (Prentice Hall Intcmational (UK), Cambridge: 1994)
~

235