Beruflich Dokumente
Kultur Dokumente
Abstract— This paper presents the application of long-range knowledge of the rotor parameters is vital to the performance
predictive control to the problem of induction motor stator of the vector controller, as is accurate control of the stator
current regulation. The particular controller used is the general- current.
ized predictive controller in a synchronous reference frame. The
theory and implementation of this controller are discussed and There are two main types of CRPWM inverters, the hys-
simulation results are presented which compare favorably with teresis regulator and the sine-triangle comparison controller
those for proportional integral compensated controllers. [1]–[3]. A hysteresis current controller gives a near-sinusoidal
Index Terms— Generalized predictive control, current regu- motor current with small ripple but requires a high and
lated PWM, voltage source inverter, induction machine drive. nonconstant switching frequency in the inverter. A sine-
triangle controller uses a natural sampled PWM scheme, in-
corporated with a proportional integral (PI) current controller,
NOMENCLATURE where the parameters are adjusted to minimize the magnitude
Supply frequency rad/s. and the phase errors in the ac currents [2]–[4]. With fixed
V Stator voltage vector in the stationary reference switching frequency and linear characteristics, a PI controller
frame. generates less acoustic noise and inverter switching loss than
V Stator voltage vector in the synchronous reference its hysteresis counter-part. However, a sinusoidally varying
frame. steady-state error is always present in the stator current [4]–[6]
I Stator current vector in the stationary reference of magnitude dependent on the machine operating frequency.
frame. Various approaches to solving the current regulation prob-
I Stator current vector in the synchronous reference lem have been reported, most of which have attributed the
frame. problem to a loss of either gain or bandwidth [7]–[10]. Only
R Stator resistance. Schauder has recognized that the problem stems from the
L Stator leakage inductance. inherent structure of the controller [5]. Rowan and Kerkman
L Rotor leakage inductance referred to primary turns. suggested the design of a PI-sine-triangular comparison reg-
L Mutual magnetizing inductance. ulator in the synchronous rotating reference frame, where the
L Total leakage inductance. input and output are dc quantities [6] rather than the stationary
E Machine back EMF vector in the stationary refer- reference frame. A synchronous PI controller relies on the
ence frame. accurate tuning of the proportional gain and the integration
E Machine back EMF vector in the synchronous time, which is difficult because of changes in the operating
reference frame. point and variations in the machine parameters.
K Proportional gain factor of the PI controller. Among various forms of predictive controller, the gener-
K Integral factor of the PI controller. alized predictive controller (GPC) is particularly robust due
K Equivalent gain factor of a voltage source inverter. to its use of model structure [11]–[13]. This paper presents a
PWM current regulator using a generalized predictive control
I. INTRODUCTION scheme is developed in a synchronously rotating reference
frame. The controller predicts the stator current over several
(4)
II. THE INDUCTION MOTOR MODEL
An induction motor supplied through a voltage source
inverter has an input signal that is nonsinusoidal. This can be III. STATOR CURRENT REGULATOR USING A PI CONTROLLER
decomposed to a set of sine waves of different frequencies The objective of the stator current controller is to ensure
by Fourier analysis and for each there is a corresponding that the measured stator currents track the required values
equivalent circuit. Motor current response via superposition accurately and with as short a transient interval as possible.
is complex so a simplified approach is usually adapted, under This is hard to achieve because:
the following assumptions [1], [15]: 1) the stator circuit is nonlinear due to speed-dependent
1) The total leakage inductance per phase can be repre- back EMF’s in the stator windings;
sented as L , where 2) the controlled variables are three-phase ac quantities, re-
and are stator and rotor leakage inductances and quiring controller parameters suitable for both amplitude
is the mutual leakage inductance. and phase angle regulation;
2) The harmonic slip is unity, hence for harmonic currents 3) the PWM voltage source inverter introduces dead times
the motor behaves as a leakage inductance. into the system which result in waveform distortion in
3) The fundamental current loop under normal conditions, the power stage.
with a slip close to zero, is presented with a back EMF In a PI type current control loop in a stationary reference
approximated as where frame, the three-phase sinusoidal stator currents are compared
rotor flux and is the angular frequency of with the reference currents generated by the speed/torque
the ac supply. The simplified motor dynamic model is control loop and the errors are used by three identical PI
shown in Fig. 1. controllers. The controller outputs are then compared with a
4) Machine speed is assumed to be constant because the fixed-frequency triangular carrier wave. The resulting PWM
circuit time response constant is very much shorter than signals, whose duty cycle is proportional to the control output,
that of the mechanical system. control the inverter switching. If the coordinate is used,
The axis voltage matrix in a stationary reference frame the three-phase variables are converted into axis forms
may be written and two identical controllers are used. The control
outputs are converted back to their three-phase equivalent
(1) before proceeding to the PWM waveform generation process,
as shown in Fig. 2. If the controller parameters are chosen
where the superscript denotes that the axes are adequately, this control scheme provides good quality control
stationary and denotes the differential operator . Terms during transient periods. The main problem is that, in the
represent supply voltage vectors and I and I are stator steady state, there is always a phase error between the stator
current vectors in the stationary reference frame. Its equivalent phase currents and the reference values. Moreover, the size of
in a synchronous reference frame is this error varies with the machine operating frequency. The
cause of this lies in the structure of the controller and is
(2) analyzed below.
Consider the transfer function of a PI controller
In (2) the superscript is used for the synchronous reference
frame. Note that (2) represents an interactive two-input/two- (5)
output system, whereas the input and output in (1) are com-
pletely decoupled and the system can therefore be represented where K is the proportional gain factor and K the integration
by two independent processes. Furthermore, the variables in gain factor.
IEEE TRANSCATIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 5, NO. 1, JANUARY 1997 121
Fig. 2. Closed-loop stator current control system in the stationary reference frame.
As back EMF is assumed constant, the stator circuit transfer form as that given by (5) but operates on the stator circuit
function for either I or I may be written model in a synchronous reference frame given by (2). The
input and output variables are all dc quantities in the steady
(6) state. The diagonal elements of the model (2) are identical
first order processes but the plant should be viewed as a
where represents the net circuit gain and T is . The multiple-input/multiple-output system. For a fixed value of
corresponding closed-loop transfer function is given by the eigenvalues of this second-order system lie at /L
j and at high values of the inherent damping becomes
very low, with possible stability problems. For low and
medium values of , the synchronous PI controller is likely
(7) to give stable closed-loop responses for currents I and I . It
also eliminates completely the steady-state error. Nevertheless,
the two nonzero off-diagonal elements, reactances and
will inevitably affect the performance of the transient
In (7), K represents the gain factor caused by the PWM response and hence the parameters of the controller must be
inverter, K equals the product of K and A, and the current carefully chosen to minimize the effect of interaction.
outputs can be expressed as The synchronous controller can also be implemented in
a stationary reference frame. This involves transforming the
(8)
controller (5) to a stationary reference frame, yielding the
It is known that the objective of the current controller is controller outputs V expressed by
to drive I (S) as close as possible to I (S) until they are
equal in the steady state at all operating frequencies, when the (10)
magnitude and phase-angle of the closed-loop transfer function
are The two-input/two-output control loop is shown in Fig. 3.
The controller consists of two parts, the conventional PI
(9) regulator and the cross-coupled control states .
In transient periods the PI part dominates the control action in
In the steady state, the modulating (drive) frequency needs order to force the current error to zero. When the steady state
to be in the range 0 2 100. It can be seen from (7) that is reached, the current errors are zero and the cross-coupling
the conditions of (9) can only be achieved when equals states provide sinusoidal voltage output to the machine at the
zero. For any other values, both the amplitude and phase operating frequency. Hence this controller effectively solves
angle of are different. The parameters of the controller the problem of nonzero steady-state error under ac operating
should, therefore, be tuned to meet the conditions defined conditions. Nevertheless, the transient performance is gener-
in (9) as closely as possible, but they cannot completely ally worse than that of the stationary regulator, indicated by
satisfy them. This, in turn, means that the controller has to be the cross-coupling states.
retuned when changes to avoid steady-state performance
degradation. Consequently, an unlimited number of regulators
IV. LONG-RANGE PREDICTIVE
with different characteristics are required. Clearly, this is
CONTROL OF STATOR CURRENT
unrealistic to achieve and hence the PI control scheme in the
stationary reference frame is not ideal for high-performance
machine drives. A. System Equations
To overcome the problem of controlling ac quantities, a Predictive control is a type of control scheme which, given
synchronous PI regulator is defined which takes the same a model of the system and a knowledge of past inputs
122 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 5, NO. 1, JANUARY 1997
Fig. 3. Closed-loop stator current control system using the synchronous reference frame controller.
Fig. 4. Closed-loop stator current control system using the GPC controller.
..
.
(11) (13)
IEEE TRANSCATIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 5, NO. 1, JANUARY 1997 123
where
.. (19)
.
Note that both g and are not just the step response
sequences of the model. The last two terms on the right-hand
side of (18) and (19) are due to the cross-coupling terms of
the model. This is different from the situation when the model
(14) consists of two independent processes with no interaction
where between them. In such cases the terms are zero and the
terms are samples of the model step response. The first
element of equals zero because cross-coupling does
not occur until the second sample instant.
Rewriting (16) by separating and terms and using the
matrix form, gives the final form of the predictors as
Note that terms in
(14) are all zero, as the future controls and back EMF are ..
defined to be constant. Rearranging the above equations so that . ..
.
the predicted terms and
..
are expressed by two parts; the past states of the .
system and the part related to the future controls and ..
, the currents at time and at sample steps in the .. .
future are predicted using .
.. ..
. .
(20)
..
.
(15)
Terms and in (15) account for the past states The GPC defines a quadratic cost function of the form
of the system. They are both initially zero because the system
starts from the steady state and the incremental form is used
so that
(21)
(16) where is the finite number of output prediction sample steps,
term represents the future reference currents, is an
(17) identity matrix, and represents a constant weighting factor.
Note that is equivalent to when the back EMF’s
In (16) and (17) C equals . In (15), terms are constant. Minimizing this cost function of the future current
form a parameter vector whose elements are derived errors and control voltage increments gives a relationship for
from the step response of the model and the cross-coupling calculating the next voltage inputs
term . Hence
where .. ..
. .
.. ..
Similarly, terms in (15) gives vector . .
of order whose parameters are obtained from combining
124 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 5, NO. 1, JANUARY 1997
(a)
(b)
(c)
Fig. 5. Responses of phase A stator current when command current changes. (a) Stationary reference frame PI controller. (b) Synchronous reference
frame PI controller. (c) GPC controller.
IEEE TRANSCATIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 5, NO. 1, JANUARY 1997 125
[9] D. M. Brod and D. W. Novotny, “Current control of VSI-PWM invert- [12] D. W. Clarke, C. Mohtadi, and P. S. Tuffs, “Generalized predictive
ers,” IEEE Trans. Ind. Applicat., vol. IA-21, pp. 562–570, May/June control—Part 1: The basic algorithm,” Automatica, vol. 23, no. 2, pp
1985. 137–148, 1987.
[10] H. Nagase, Y. Matsua, K. Ohmishi, H. Ninomiya, and T. Koike, “High- [13] C. R. Cutler and B. L. Ramaker, “Dynamic matrix control—A computer
performance induction motor drive system using a PWM inverter,” in control algorithm,” in Proc. Int. Symp. Advanced Process Supervi-
IEEE Ind. Applicat. Soc. Annu. Mtg. Conf. Rec., 1983, pp 596–603. sion and Real-Time Knowledge-Based Contr., Newcastle, U.K., Nov.
[11] D. W. Clarke and L. Zhang, “Long-range predictive control using 1988.
weighting-sequence models,” IEE Proc. Pt. D, vol. 136, no. 3, 1987, [14] D. M. Brod, “Current control of VSI-PWM inverters,” M.S. thesis, Univ.
pp. 187–185. Wisconsin—Madison, 1984.