You are on page 1of 7

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 17, NO.

2, MARCH 2002

279

Vector Current Controlled Voltage Source ConverterDeadbeat Control and Saturation Strategies
Rolf Ottersten, Student Member, IEEE, and Jan Svensson, Member, IEEE

AbstractThis paper presents a voltage source converter connected to a grid with software specially designed for limited control voltage. The voltage source converter uses a deadbeat vector current controller. The paper deals with limiting reference voltage, integrator windup and delay time compensation. Simulations and experimental verifications of the proposed controller are included. Index TermsPulse width modulation, vector current control, voltage source converter.

I. INTRODUCTION OR HIGH performance applications, such as active power filters and adjustable speed drives with servo performance, it is important to maximize the current bandwidth of the voltage source converter (VSC). The current bandwidth depends mainly on the dc-link voltage, the control program, the load impedance and the sampling and switching frequencies. A cost-effective solution to increase the bandwidth is to improve the control program of the VSC. The vector current controller using deadbeat gain offers a high bandwidth and good steady-state performance. However, during large transients the demanded reference voltage of the deadbeat controller is likely to exceed the VSC control voltage capability, subsequently referred to as saturation. Traditional deadbeat controllers are not designed for saturation, resulting in reduced performance such as integrator windup, sluggish response to current steps and cross-coupling between active and reactive currents. A reference voltage that exceeds the VSC control voltage capability must be appropriately limited. A new high performance method [1] to limit the reference voltage considers the dynamics of the plant. In [2] the current cross-coupling is used to obtain a faster response to steps in active current. However, neither [1] nor [2] deal with integrator windup, which occurs when the VSC saturates. There are several ways to avoid integrator windup. One method is to stop updating the integrator during saturation. Unfortunately, this method gives a reduced performance since there will be no integration during the VSC saturation. Another possibility is to use a conditional integrator, as proposed in [3], which deactivates the integrator when the current control error exceeds a certain value. The method requires an accurate tuning in order to perform well. Back-calculation [4] yields a good antiwindup performance without tuning. To overcome oscillations due to the computational delay of one sample when implementing the vector deadbeat current controller digitally, a delay time compensator must be used. The delay time compensator in [5] is a high performance and rather complex neural net, which predicts the currents one sample ahead. The classical Smith predictor was proposed in [3]. However, [3] used a direct calculation of the state variables. This paper presents a vector deadbeat current controlled VSC connected to a three-phase grid. The current controller uses algorithms specially designed for a limited control voltage. Various methods of limiting the reference voltage are presented

NOMENCLATURE Superscripts denoting reference and estimated quantities. Subscript denoting stationary two-axes -frame. Subscript denoting rotating two-axes -frame, which has its -axis aligned with grid the flux vector. Subscript denoting two-axes -frame (Fig. 3). Instantaneous and nominal value of grid voltage. Grid, sample and switching frequencies. VSC current, VSC current control error. Sample, observer gain. Integration and proportional gain. Inductance and resistance of inductor filter. Modulation index. Sector in VSC voltage hexagon. Function indicating VSC saturation. Sampling period. Voltage source converter control voltage. Instantaneous and nominal dc-link voltage. Angular grid frequency. Space vectors. Feedback voltage. Integration and feedback voltage. Proportional gain voltage. The eight realizable space vectors of the VSC. Transformation angles.

sat

Manuscript received August 18, 1998; revised October 10, 2001. This work was supported by the Swedish National Board for Industrial and Technical Development. Recommended by Associate Editor L. Morn. The authors are with the Department of Electric Power Engineering, Chalmers University of Technology, Gothenburg SE-412 96, Sweden (e-mail: rolf.ottersten@elteknik.chalmers.se; jan.svensson@elteknik.chalmers.se). Publisher Item Identifier S 0885-8993(02)02251-2.

08858993/02$17.00 2002 IEEE

280

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 17, NO. 2, MARCH 2002

Fig. 2. Fig. 1. System overview.

Proposed vector current controller.

A. Delay Time Compensation and reviewed in combination with an antiwindup integrator using back-calculation. The presented limiting methods include both traditional and recently presented methods. The Smith predictor [6], using a full-order state observer, handles the computational delay of one sample. The performance of the proposed system is studied and compared with that of a conventional deadbeat vector current controlled system, both theoretically and experimentally. II. SYSTEM CONFIGURATION The scheme of the VSC connected to the grid and its vector current controller is shown in Fig. 1. The grid voltages and currents are sampled synchronously in the block s&h and transformed to the rotating two-axes -coordinate system, in which the -axis is aligned with the grid flux vector. Pulse width modulation (PWM) is achieved using the sub-oscillation method. By adding a zero order sequence to the three-phase reference voltages [7], linear modulation is possible up to the modulation , where corresponds to six-step operindex ation. The VSC uses IGBT-valves, which are connected to the grid over an inductor filter. The sampling frequency is equal to the switching frequency. III. VECTOR CURRENT CONTROLLER This section will present a deadbeat vector current controller specially designed for a limited VSC control voltage. The proposed vector current controller, shown in Fig. 2, is based on the vector current controller presented in [8], having the following control law in space vector notation: (3) When implementing the deadbeat controller digitally, the computational delay time of one sample will be approximately equal to the current rise time, which gives rise to undesired dynamics. To overcome this, the delay time must be compensated. This paper proposes the Smith predictor using a state observer. When compared with direct calculation of the state variables, the observer is less sensitive to disturbances and modeling errors. Using the Euler feed-forward discretization of the plant, where is the observer gain, the observer is given by

B. Integrator Windup To avoid integrator windup, the integration of the current control error must be stopped or modified when the VSC saturates. This paper adapts back-calculation [4] to deal with integrator windup. During saturation, the equivalent control error is back-calculated, by using the limited voltage source converter control voltage. The back-calculation of the current control error is given by (4)

C. Limiting Reference Voltage Vector Limiting the reference voltage vector that exceeds the voltage source converter control voltage capability is performed in the block limit reference voltage vector. The modified reference is used as input to the back-calculation voltage vector and the state observer blocks. IV. OVERMODULATION AND SATURATION

(1) Fig. 3 shows the eight realizable voltage vectors for a threephase VSC, using amplitude-invariant transformation between -coordithe three-phase system and the two-axes stationary nate system.

where the feedback voltage

is equal to (2)

OTTERSTEN AND SVENSSON: VECTOR CURRENT CONTROLLED VOLTAGE SOURCE CONVERTER

281

Fig. 4. Principle of the circular limit method. Fig. 3. Eight realizable voltage vector of three-phase VSC. Amplitude invariant transformation has been used.

where fined by

is the angle between the -axis and the -axis, de(6)

The VSC is capable of delivering voltages within the hexagon that is spanned by the six active voltage vectors. Linear modulation is possible up to the radius of the maximum circle that can be inscribed within the hexagon. Control in the range of , i.e., the region outside the circle but within the hexagon, is possible but results in low-frequency current harmonics. However, control in this region, often referred to as the overmodulation region, may be preferable to enhance the transient behavior of the VSC. Control in the overmodulation region when using suboscillation PWM with an injected zero sequence signal is possible by simply limiting the three-phase reference voltages . Some additional computations are required to within obtain control in the overmodulation region when using space vector PWM [9]. Maximum VSC control voltage is obtained at the hexagon boundary, which results in six-step operation. During transients, the vector current controller may demand a reference voltage vector outside the hexagon, thus exceeding the VSC control voltage capability, i.e., saturation. VSC saturation is particularly likely to occur in a flux orientated vector current controller during a positive step in the -axis current, since the back-emf of an electric machine or the grid voltage vector will be aligned with the -axis.

is located. A reference voltage and is the sector where vector outside the hexagon will be equivalent to (7) indicating that the reference voltage vector B. Circular Limit Method (CL) The circular limit method (CL) chooses the largest voltage vector on the maximum circle within the hexagon and orientated in the same direction as the reference voltage vector, as shown in Fig. 4. Thus, continuous modulation and average sinusoidal voltages are ensured at all times. A reference voltage vector is located in the overmodulation region or outside the hexagon if the following condition holds true: (8) The modified voltage vector is given by (9) Unfortunately, the transient behavior of the CL method will be rather sluggish, since it does not make use of the available voltage in the overmodulation region. C. Minimum Amplitude Error Method (MAE) The minimum amplitude error limit method (MAE) chooses the voltage vector on the hexagon boundary that is located nearest the reference voltage vector, so that the voltage amplitude error will be maintained at a minimum, as shown in Fig. 5. In the -coordinate system, the components of the modified reference voltage vector become (10) (11) must be limited.

V. LIMITING REFERENCE VOLTAGE VECTOR To maintain a proper current control, it is necessary to detect and appropriately limit reference voltage vectors outside the hexagon. The transient behavior of the VSC depends on how the reference voltage vector is modified. This section will present a review of methods used to limit a reference voltage vector outside the hexagon. A. Detecting Reference Voltage Vector Outside the Voltage Hexagon A reference voltage vector outside the hexagon can be deinto tected by transforming the reference voltage vector the -coordinate system, presented in Fig. 3. The transforma-coordinate system and the -coordinate tion between the system becomes (5)

. In the stationary voltage vector

-coordinate system, the modified reference (12)

282

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 17, NO. 2, MARCH 2002

Fig. 5.

Principle of the minimum amplitude error limit method.

Fig. 7. Principle of the dynamic vector limit method. TABLE I PARAMETERS OF THE SYSTEM

Fig. 6. Principle of the space vector limit method.

is equal to (13) When using sub-oscillation PWM with zero sequence injection, MAE can also be performed by limiting the three-phase refer. ence voltages within D. Space Vector Limit Method (SVL) The space vector limit method (SVL) chooses the voltage vector on the hexagon boundary orientated in the same direction as the reference voltage vector, as shown in Fig. 6. The SVL method is very suitable when using space vector modulation PWM, but can also be applied to sub-oscillation PWM. The modified voltage vector components can be determined by using simple geometry as (14) (15) The modified reference voltage vector is then transformed to the -coordinate system as described in the previous stationary section. E. Dynamic Vector Limit Method (DVL) The dynamic vector limit method (DVL) was first presented in [1]. Fig. 7 shows the principle of the DVL method. When a reference voltage vector outside the hexagon has been detected, DVL considers the dynamics of the plant and the hexagon boundary, thus, forcing the current control error to move in the same direction as stated by the vector current controller. F. -Axis Cross-Coupling Method (DQC)

Fig. 8. Vector current controller and circular limit method. (a) Simulation. (b) Measured.

In Fig. 7, is the voltage demanded by the current controller excluding the proportional term, given by

(16) and is the proportional gain voltage (17) The DVL method chooses the components of the modified reference voltage vector by using simple geometry as (18) (19) (20) (21)

The -axis cross-coupling method (DQC) was presented in [2] as a simplified version of the minimum time current control method, presented in [10]. The DQC method is not a reference voltage vector limiting method, but a method to enhance the

OTTERSTEN AND SVENSSON: VECTOR CURRENT CONTROLLED VOLTAGE SOURCE CONVERTER

283

Fig. 9. Vector current controller and minimum amplitude error limit method. (a) Simulation. (b) Measured.

Fig. 11. Vector current controller and dynamic vector limit method. (a) Simulation. (b) Measured.

Fig. 10. Vector current controller and space vector limit method. (a) Simulation. (b) Measured.

Fig. 12. Vector current controller and dq -axis cross-coupling method. (a) Simulation. (b) Measured.

response to reference steps of the -axis current. In the rotating grid flux orientated -coordinate system the -axis current can be written as (22) Normally, a reference step of the -axis current is handled by the . According to (22), though, VSC -axis control voltage the response of the -axis current can be enhanced by utilizing the cross-coupling of the -axis current. It was proposed by [10] that the VSC -axis reference current be modified according to (23) Furthermore, [10] proposed that the transient current be limited as (24) is the rated peak current of the IGBT-valves. Given where the modified value of the -axis reference voltage must be updated . Finally, the updated reference voltage must be limited in an appropriate way, possibly by using the MAE method. VI. COMPUTER SIMULATIONS AND EXPERIMENTAL RESULTS This section will demonstrate computer simulations and experimental results of the proposed deadbeat vector current controller. The simulation parameters are shown in Table I. The

voltage and the current base values are V, A. The positive reference current steps are positive slope triggered when the phase voltage crosses zero voltage. Simulations and measurements have been performed for a step in the reference -current with constant reference -current and the reverse. ms and The positive step from 0.5 pu to 0.5 pu occurs at ms the reference changes from 0.5 pu to 0.5 pu. The at other reference current is constant at 0.2 pu. The function indicates VSC saturation. A. Steps in Reference -Current Computer simulations and experimental results from Figs. 812 show the reference -current step responses of the proposed vector current controller using various limiting methods. Fig. 8 shows the response of the vector current controller using the CL method. The transient behavior is rather sluggish, since the available voltage in the overmodulation region is not used. As shown in Fig. 8, the step down does not result in saturation due to the grid voltage vector direction. Fig. 9 shows the response of the vector current controller using the MAE method. In comparison with the CL method, the response is faster but there is a dip in the -current. Fig. 10 shows the response of the vector current controller using the SVL method. The response is approximately as fast as when using the MAE method, and the -current is almost constant during the step. Fig. 11 shows the response using the vector

284

IEEE TRANSACTIONS ON POWER ELECTRONICS, VOL. 17, NO. 2, MARCH 2002

Fig. 13. Traditional deadbeat vector current controller and minimum amplitude error limit method. (a) Simulation. (b) Measured.

Fig. 15. Vector current controller and space vector limit method. (a) Simulation. (b) Measured.

Fig. 14. Vector current controller and minimum amplitude error limit method. (a) Simulation. (b) Measured.

Fig. 16. Vector current controller and dynamic vector limit method. (a) Simulation. (b) Measured.

current controller and the DVL method. The result is almost identical to the SVL method. Fig. 12 shows the response using the vector current controller and the DQC method. The response is slightly faster when compared with those of the MAE, the SVL and the DVL methods, but at the expense of a huge dip in the -current. As a comparison, Fig. 13 shows the response of the traditional deadbeat vector current controller using the MAE method. The response is rather sluggish, there is a dip in the -current and due to integrator windup there is an overshoot in the -current. B. Steps in Reference -Current The computer simulations and experimental results from Figs. 1416 show the reference -current step responses using the MAE, the SVL and the DVL methods. All methods have approximately the same settling time due to the grid voltage vector direction. However, the current cross-coupling differs between the different limiting methods. The DVL method has the lowest cross-coupling. C. Summary of Step Responses As shown, the method of limitation should be chosen according the application since the different methods result in different step responses. The positive step in the reference -current is a demanding step according to the rise time. In applications such as servo drives, the -current is allowed to fluctuate

Fig. 17. Positive step in: (a) q -current and (b) d-current.

during steps in the -current and the DQC method is preferable, especially if the speed is high. Fig. 17 shows an illustration of the various limiting transient behaviors. In Fig. 17(a) and (b) there are positive reference steps in the -current and -current, respectively. For simplicity, it has been assumed that (25) In Fig. 17(a), the MAE method yields a negative voltage trajectory on the -axis, which results in a -current dip. The DVL and SVL methods yield approximately the same voltage trajectory on the -axis as the minimum amplitude error method, but no negative voltage trajectory on the -axis. As a result, the dynamic vector limit and space vector limit methods yield a constant -current during the positive -current step. Fig. 17(b) shows that all limiting methods except the DVL method yield a

OTTERSTEN AND SVENSSON: VECTOR CURRENT CONTROLLED VOLTAGE SOURCE CONVERTER

285

reduced -axis voltage. Therefore, only the DVL method yields a constant -current during the reference step in the -current, which makes the DVL method suitable for an active power filter. VII. CONCLUSIONS This paper has proposed a deadbeat vector current controller, specially designed to handle VSC saturation. Various methods of limiting the reference voltage vector have been investigated. Simulation and experimental results have shown that the controller handles saturation successfully. The dynamic vector limit method gives the best balance between fast response to reference changes and independent current control. REFERENCES
[1] J.-K. Seouk and S.-K. Sul, A new overmodulation strategy for induction motor drive using space vector PWM, in Proc. 10th Ann. Applied Power Electron. Conf. Expo. (APEC95), vol. 1, Dallas, TX, Mar. 59, 1995, pp. 211216. [2] S.-K. Sul and J.-W. Choi, Fast current controller in three phase ac/dc boost converter using dq axis crosscoupling, IEEE Trans. Power Electron., vol. 13, pp. 179185, Jan. 1998. [3] J.-W. Lee, An intelligent current controller using delay time compensation for PWM converter, in Proc. 7th Eur. Conf. Power Electron. Applicat. (EPE97), vol. 1, Trondheim, Norway, Sep. 810, 1997, pp. 342347. [4] K. J. strm and T. Hgglund, PID controllers: Theory, design and tuning, Tech. Rep., ISA, Research Triangle Park, NC, 1995. [5] F. Kamran and T. G. Habetler, An improved deadbeat rectifier regulator using a neural net predictor, IEEE Trans. Power Electron., vol. 10, pp. 504510, July 1995. [6] Z. J. Palmor, The Control Handbook. Orlando, FL: CRC Press, 1996, pp. 224237.

[7] K. G. King, A three-phase transistor class-b inverter with sinewave output and high efficiency, in Proc. IEE Int. Conf. Power Electron.Power Semicond. Applicat., vol. 1, London, U.K., Dec. 35, 1974, pp. 204209. [8] J. Svensson, Inclusion of dead-time and parameter variations in VSC modeling for predicting responses of grid voltage harmonics, in Proc. 7th Eur. Conf. Power Electron. Applicat. (EPE97), vol. 3, Trondheim, Norway, Sep. 810, 1997, pp. 216221. [9] J. Holtz, Pulsewidth modulationA survey, IEEE Trans. Ind. Electron., vol. 39, pp. 410420, Dec. 1992. [10] S.-K. Sul and J.-W. Choi, New control conceptMinimum time current control in the three-phase PWM converter, IEEE Trans. Power Electron., vol. 12, pp. 124131, Jan. 1997.

Rolf Ottersten (S99) was born in Gothenburg, Sweden, in 1974. He received the M.Sc. degree in electrical engineering from Chalmers University of Technology, Gothenburg, Sweden, in 1997, where he is currently pursuing the Ph.D. degree in the Department of Electric Power Engineering. His research interests include power quality issues and modeling and control of electric drives.

Jan Svensson (S96M99) received the M.Sc. and Ph.D. degrees in electrical engineering from Chalmers University of Technology, Gothenburg, Sweden, in 1991 and 1998, respectively. He is currently an Assistant Professor in the Department of Electric Power Engineering, Chalmers University of Technology. His research interests include control strategies, power quality, applications of grid-connected converters, and wind power.

You might also like