Sunteți pe pagina 1din 4

2015 International Siberian Conference on Control and Communications (SIBCON)

Modal Optimization of AVR for Synchronous


Generator Using the Finite Gradient
D.V. Armeeva
Skoltech Center for Energy Systems
Skoltech
Skolkovo, Moscow Region, Russia.
denarpost@yandex.ru

A.V. Chekhonadskikhb
Department of Engineering Mathematics
NSTU
Novosibirsk, Russia
alexander_cheh@mail.ru

A.A. Voevoda
Department of Automation
NSTU
Novosibirsk, Russia
voevoda@ucit.ru

I. INTRODUCTION
AbstractTraditional power systems consist of few
centralized power plants, passive grid and uncontrollable passive
loads. It is expected that deep penetration of modern trends like
distributed generation, SmartGrids and microgrids will lead to a
plenty highly automated power units. Each of them should be
equipped with effective controllers. Like many years before PID
controllers remain relevant and popular despite many others
technologies, because they are based on well proven, reliable and
clear practice. But PID controllers should be well tuned in order
to keep stability of the unit and the entire grid. Existing methods
usually assume that the control parameters are chosen at first,
then the controlled unit is tested on applicability and stability.
Classical methods (D-decomposition et al.) are quite effective for
only one or two tunable parameters.
The method proposed in this paper is based on the
optimization concept of modal design. We consider Hurwitz
function (control system relative stability), which is determined
by the rightmost pole location and depends on the parameters of
the controller. Its minimization using finite gradient method
represents a complicated optimization problem. Number of the
rightmost poles defines dimension of the manifold. When several
poles is almost on the same vertical, we are approaching the zerodimensional manifold: points of global or local extrema.
The described method is suitable to various problems of
parametric optimization. The authors previously tested it in the
control of various plants such as multilink mathematical and
inverted pendulums. In the article it is used to optimize PDD2
controller for the linear model of a synchronous generator;
similarly it can be used for optimization of PID controllers and
other structure ones.
KeywordsAVR;
distributed
generation;
smartgrid;
synchronous generator; control system; low order controller;
parameter optimization; relative stability; finite gradient method;
pole location

This publication is based on work funded by Skolkovo Institute of


Science and Technology (Skoltech), program name Center for Research,
Education and Innovation for Energy Systems under Research Program
Contract number 199-MRA.
b
Work was financially supported by the Ministry of Education and
Science of the Russian Federation on the state task 2014/138, project 1052.

The voltage control in power systems is one of major


problems, determined by customer requirements for power
quality of the electricity.
System requirements for the voltage levels are determined
by stability conditions of synchronism of power generators.
The stability is considered in two aspects: steady state
small-signal stability, i.e. the system's ability to maintain
synchronism after small disturbances and transient stability
as the ability to regain an acceptable state after a significant
perturbation, for example, in emergency conditions during
short circuits, etc. [1-3].
The formatter will need to create these components,
incorporating the applicable criteria that follow.
The modern synchronous generators are equipped with
automatic voltage control (AVR). The AVR controls the
voltage in steady state regime and provide excitation boost
under emergencies.
In the article below authors consider issues related to the
steady state conditions and influence AVR on small-signal
stability.
The main AVR feature is control voltage at the generators
bus by mean of excitation system. AVR input signal is the
deviation from a voltage setpoint Uset (Fig. 1).
To change the voltage, AVR affects the excitation voltage
of the exciter (EX in Fig. 1), and further in proportion, on
excitation voltage of the generator after a short transient. Then
the excitation current If changes as a result of electromagnetic
transients in the rotor winding, which defines the field of the
machine. Field specifies the generators emf Eq, and hence the
voltage Ug on its terminals.
The main controlled variable is the voltage deviation V at
the terminals.
However, it is critical to take into account AVR impact on
the stability, which is judged by the evolution of another
parameter . Delta is a mutual rotor angle between the rotor

978-1-4799-7103-9/15/$31.00 2015 IEEE

2015 International Siberian Conference on Control and Communications (SIBCON)


and the bulk bus voltage Us. Dynamics in the windings is
described by nonlinear equation of second order.

The characteristic polynomial has the fifth degree,


therefore P, PD and PDD2 control is the low order one. Stable
plants allow a variety of approaches and methods of optimal
controller settings [8]. The authors take modal method of low
order controller design and optimization [9, 10]. Hurwitz

Fig. 1.
Functional scheme of the control system: setpoint Uset subtracted
from the voltage Ug of the generator; AVR converts the error U in the control
action emf Eqe which transforms it into no-load emf Eq acting on the output
voltage Ug.

II.

EXCITATION CONTROLLER AND THE CHARACTERISTIC


POLYNOMIAL

In Russian UPS there are two types of AVR. It is mainly


depends on the type of excitation system due to features of the
actuator. When the controller limitations are strong then the
AVR uses P and PD controller. When the excitation current is
being produced by power electronics, the AVR can carry out
PDD2 control to response on deviation and its derivatives
[1,2].
In the latter case, there is an important problem of the
gain tuning in order to effectively control the stability of the
angle [1, 2].
A linearization of generator dynamic differential equations
as well as its transformation to the operator form is based on
the fact that system pole location in the left semi-plane
provides stability of the original nonlinear system [4]. The
stability area in the parameter space for the linear model may
be found in a variety of ways from D-decomposition method
[5] to the methods of the modern robust system theory [6].
But the D-decomposition technique is effective only for
controllers with one or two parameters, so the PDD2
parameters are tuned on the basis of models with quadratic
functional quality or empirically by real electromechanical
dynamics tests.
For this reason, the development of alternative approaches
for setting up the AVR is a significant problem. This is very
important for distributed generation and SmartGrids which
include synchronous generation.

function H(a, b, c) = max(Re z1,, Re z5) has been chosen as


the objective one (here z1,, z5 are the poles of the system);
its minimization corresponds exactly to maximization of
system relative stability.
So, to go from PDD2 controller to the widespread in
practice the P and PD controllers it is enough to consider the
cases when b = c = 0 and c = 0 respectively.
III.
CONTROLLERS OF PROPORTIONAL AND
PROPORTIONAL-DIFFERENTIAL ACTION
The authors have carried out Hurwitz
minimization by means of finite gradient method.

function

Choosing initial values for the gradient descent in the area


where necessary conditions of stability hold (i.e. the
coefficients of the characteristic polynomial are positive) we
obtain the stabilization in the set of points belonging to onedimensional manifolds in the space of the parameters a and b,
for example:
i) H (1.513; 4.258; 0) 0.3326; the characteristic root
ensemble is equal to {12.52; 0.333 6.990i; 0.333
1.093i};
ii)
H ( 0.420; 0.0001; 0) 0.1445; the characteristic
root ensemble is equal to {9.97; 3.452; 0.145; 0.145
6.888i};
Both of these cases correspond to the location of the two
complex pairs (i) or the complex pair of and the real root (ii) on
the same vertical (Fig. 2).
Note, that point (ii) corresponds to the P-control case
(b 0), but it doesnt yield Hurwitz function minimum for the
P-control, see (v) below.

The developing of the generator model in relative units and


linearization of its equations has been done in [7]; its transfer
function takes the form:
G(s) = nobj(s)/dobj(s) (2.77 s2+ 169.3)
(19.32 s3 + 10 s2 + 917.6 s + 207.8) 1(0.3s + 1)1
To express the characteristic polynomial for P, PD and
PDD2 controllers, lets consider first the general case, when
AVR structure is defined as a PDD2 controller. Thus, device
has the transfer function with a unit denominator: ncontr ~ a +
+ bs + cs2. The characteristic polynomial of the system is equal
f(s) = dobj(s) dcontr(s) + nobj(s) ncontr(s) 0.579s5 + (2.77c +
4

+ 8.028) s + (2.7b + 50.848) s + (2.77a + 169.3c +


+ 383.27)s2 + (169.3b + 1000.7)s + 169.3a + 207.8 .

Fig. 2. Root portrets for gradient descent stabilization points in 1-2 paramer
spaces. One-dimentional manifolds (i, ii) are defined by one equality or real
components; zero-dimentional manifold (iii) is defined by two equalities or
real and imaginary components of the roots.

2015 International Siberian Conference on Control and Communications (SIBCON)


An interesting case is the gradient descent stabilization on
root ensemble containing two complex root pairs close to the
double one, Fig.2 (iii):
iii)
H (0.050; 4.402; 0) 0.3296; the characteristic
root ensemble is equal to { 2.53; 0.330 7.001i; 0.330
0.705i} here the value of the proportional parameter a 0
almost corresponds to a D-regulation, i.e. derivative control.
Indeed, for a = 0 one-parameter minimization of the
parameter b leads to a point of global D-regulation minimum,
which coincides with the above with high precision:
H (0; 4.402; 0) 0.3296.
However, in terms of type (i, ii) belongs to the onedimensional manifold, where a further minimization is
possible, though it is carried out very slowly, because the
descent is along the very gentle bottom of the ravine (Fig. 3).
Global PD minimum is detected significantly further at the
point corresponding to the type (iii) in Fig.2 and zerodimensional sets in parameter space, defined by the conditions
of duplicity of complex root pairs:
iv) H (44.71; 0.451; 0) 0. 4612; the characteristic roots
are equal to {12.00; 0. 4612 5.76i}.
v) Optimization of P-regulation, that is a one-dimensional
minimization of the parameter a for b = c = 0, yields the value
H (41.0; 0; 0) 0.451, that almost coincides with the results
of PD-optimization.

This leads to the following important conclusion about suboptimality of P-regulation in the class of PD controllers:
Proposition. The proportional AVR at an optimal
parameter value allows achieving practically the same relative
stability of the system as the optimal PD controller; thus we
have almost the same rate of decay of transients.
Remark. Optimal P and PD parameters values was used in
nonlinear AVR system model, simulated in matlab (simulink)
[7]; by this way we have controlled modal optimization results
using the more explicit representation of the transients and
verified the effectiveness of the modal optimization. This
reason eliminates the need to achieve greater precision of
calculations in finding minima coordinates.
IV. PDD2 CONTROLLERS
Since the parameter space here is three-dimensional, the
minima of Hurwitz function will correspond to the root
locations defined by the three equalities of real and imaginary
root components: for example, quadruple real root or locations
of lower series in Fig. 4 (in fact, the number of extreme pole
locations for systems with three parameters is equal to eight,
[10]). The most typical case is the stabilization of the gradient
descent in the points with equal real root and real complex pair
components (Fig. 4, vi), for example:
vi) H (1.002; 3.570; 1.461) 1.3741; the characteristic
roots are equal to {1.374; 1.374 6.569i; 1.374 2.938i}.
However this manifold is a one-dimensional one and allows
continuing minimization.
A global extreme point is achieved at the point with a
double complex pair is placed at the same vertical line with the
real root close to the point (Fig. 4, vii):
vii) H (3.632; 1.726; 1.330) 1.498; the characteristic
roots are equal to {1.501; 1.499 5.321i; 1.498 5.358i}.
Lower left location in Fig. 4 was achieved in other
examples; but in this case such points were not detected

Fig.3. The level lines of Hurwitz function for PD controller from 0.40 on the
outside to 0.45 inside. The horizontal axis is the D control parameter b, the
vertical axis is the P control parameter a.

Fig. 4. Root portraits for gradient descent stabilization points in 3-parameter


space. One-dimensional manifold (vi) is defined by two equalities or real
components; zero-dimensional manifold (vii) is defined by three equalities or
real and imaginary root; the latter yields global minimum of Hurwitz function.

2015 International Siberian Conference on Control and Communications (SIBCON)


V. CONCLUSION

ACKNOWLEDGMENT

As a result, we can state that the method of modal


optimization of PDD2 controller have yielded optimal
parameter values under which the relative stability is maximal
and the decay rate ~ H(a, b, c) < 0 became minimal, so the
transient decreases as U (t) ~ U0 exp (t). This reduces the
typical suppression time of the disturbance caused by the
power deviation.

The authors thank A.V. Miheev, which carried out much


preliminary work. The authors are also grateful to prof. A.S.
Vostrikov for supporting the development of this theme and
high appreciation of the results.

P-control gives suboptimal result 0.45 in the class PD


controllers (in the global PD minimum 0.46), which was
empirically known from years of practice operating the P
controller.
The use of PDD2 regulators allows tripling the speed of
convergence: 1.5. Calculation of transients in the matlab
(simulink) shows that the optimal coefficients of the linear
model provide a relatively high quality of transients for the
original nonlinear system [7].
Hurwitz function minimization using finite gradient method
proved effective for this task; before authors have used it in
optimization of control parameters in the multilink
mathematical pendulum [9]. Thus, an important advantage of
this approach is its independence of plant type and controller
structure. Of course, the computational complexity
significantly depends on the number of parameters.
The authors hope that the method will be useful for
optimization of systems such as SmartGrids and microgrids.
Another advantage of the method is the fact that if Hurwitz
stability is achieved, one need not seek excessive precision.
Indeed, in the transition from the linearized model of the
generator or mathematical pendulum to the initial non-linear
construction, the excessive precision becomes meaningless.

REFERENCES
[1]

Venikov, V.A.
Perehodnye jelektromehanicheskie processy v
jelektricheskih sistemah, [Electromechanical transients in electric
systems]. Moscow: Vysshaja shkola [Higher School], 1985.
[2] Kundur, P., Balu, N. J., & Lauby, M. G. Power system stability and
control, New York: McGraw-Hill, 1994.
[3] Machowski, J., Bialek, J., Bumby, J. Power System Dynamics:
Stability and Control, New Dehli, Chichester: Wiley, 2008.
[4] Godunov, S.K. Obyknovennye differencial'nye uravnenija s
postojannymi kojefficientami, [Ordinary differential equations with
constant coefficients]. Novosibirsk: NGU, 1994.
[5] Polyak, B.T., . Gryazina, E.N, Stability domain in the parameter space:
D-decomposition revisited, Automatica, vol. 42, 2006, pp. 1326.
[6] Polyak B.T., Shcherbakov P.S. Hard Problems in Linear Control
Theory: Possible Approaches to Solution, Automation and Remote
Control, vol. 66, iss. 5, 2005, pp. 681718.
[7] Armeev D.V., Miheev A.V., Chehonadskih A.V. Raschet parametrov
ARV sinhronnogo generatora metodom modal'noj optimizacii, [AVR
parameter calculation for synchronous generator using modal
optimization method]. Sb. nauch. tr. NGTU [Proceedings of NSTU], no.
2(64), 2011, pp.105116.
[8] Zhmud' V.A. Modelirovanie, issledovanie i optimizacija zamknutyh
sistem avtomaticheskogo upravlenija, [Simulation study and
optimization of closed-loop control systems]. Novosibirsk: ZAO
KANT, 2012.
[9] Voevoda, A.A., Chekhonadskikh, A.V. Optimization of the pole
location of automatic control systems with a reduced-order controller,
Optoelectronics, Instrumentation and Data Processing, vol. 45, no. 5,
2009, pp. 472480.
[10] Chekhonadskikh A.V. Extremal pole placement in control systems with
a low order controller, Automation and Remote Control, vol. 75, iss.
10, 2014, pp. 17171731.

S-ar putea să vă placă și