Você está na página 1de 12

Discrete Event Dynamic Systems: Theory and Applications, 9, 319–330 (1999)

°
c 1999 Kluwer Academic Publishers, Boston. Manufactured in The Netherlands.

Discrete Events in Power Systems


LESTER H. FINK
Life Fellow, IEEE

Abstract. The relevance of Discrete Event System theory (whether synthetic or analytic) to power system analysis
and control is not clear. However, based on several characterizations of the domain of discrete systems set forth
in the January 1989 Special Issue of the IEEE Proceedings, on “Dynamics of discrete event systems” (Ho, 1989),
we attempt to delineate several power system control problems that may be amenable to analysis by DEDS
methodologies. While the author is not qualified to judge whether the latter can speak to any of the examples, it
may be relevant that in all cases the problems are non-trivial and in need of further contributions toward improved
solutions, especially in view of the ongoing radical restructuring of the industry.
The continuous process that is instantiated in the functioning of a power system is observed via three more or
less diverse variables. Two of these, the scaler “system” frequency and the vector system voltage profile, are in
principle mensurable physical phenomena; the third, system security, is a quality represented by a set membership
conditioned on an inferred relationship between the two physical system variables and future realizations of one
or more classes of autonomous stochastic phenomena. (These latter may be internal or external to the physical
system, such as, e.g., respectively, equipment failure mechanisms and weather events.)

Keywords: discrete systems, mixed systems, power system characterization, power system control, power system
security, voltage profile control, system frequency control

The Power System Trajectory

The Physical States

The power system, in its simplest representation, comprises a set of lines intersecting at
nodes (buses). Energy is injected at buses, with loads considered as negative injections;
electromagnetic fields, although they are ostensibly ignored in this model, are of basic
importance. The phenomenon of primary interest is the flow of power along lines, to and
from buses. Basic operation consists of energy extracted by loads being replaced by energy
injected by generators under automatic control, ostensibly a continuous process. At this level
of operation, loads being uncontrolled by the system operator and not available for direct
measurement, the control problem may be considered as a pursuit problem, with generation
tracking load. Since the inertia of loads is significantly less than that of generation, the
system is in balance only instantaneously (at times of zero duration) as generation overtakes
or falls behind load fluctuations. This is the crux of the power system control problem.
Lowering the level of abstraction, two phenomena are of concern: prime mover rotational
velocities (frequencies) and bus voltages. The state vector is conventionally defined as
comprising the magnitudes and angles of all bus voltages (with angles referred to an arbitrary
bus). While the state vector consists of the magnitudes and angles (with respect to an
arbitrary reference) of bus voltages, the latter are of interest to the system operator in terms of
their derivatives (frequencies), and those only of the generator buses. Furthermore, because
prime movers are (by design) herded into approximate synchronism by the interchange of
320 FINK

synchronizing energy, a prime object of interest for system control is the average system
frequency, which is a reflection of the fluctuating mismatch between energy generated and
consumed. If this mismatch gets out of control, synchronism will be lost and the system,
under control of a variety of protective devices, will begin to come apart. Thus, control
of system frequency, which is the average of the derivatives of a subset of one component
of the system state vector (generator voltage angles, rather than all bus voltage angles), is
necessary, but not sufficient, to continued system operation.
Bus voltage magnitudes (collectively, the system voltage profile) represent the other
component of the state vector. Unfortunately, perhaps, there is a dissymmetry between the
control subproblems posed by the two components of the state vector. System frequency is
available as a surrogate control objective because the interchange of synchronizing energy
between generators serves continuously to move them towards a common speed. This is
possible because the flow of active power over a line is a function of the angular difference
between the voltage vectors at the two ends of the line, and the low resistance of transmission
lines means that little active energy is lost in the process. A similar phenomenon governs
the flow of reactive energy, which is necessary to maintain bus voltage magnitudes. In this
case, the flow is governed by the difference in those magnitudes at the two ends of a line,
but unfortunately the high reactance of transmission lines means that a great deal of reactive
energy (as well as additional active energy) is lost in transit, so that voltages of individual
buses, or of small local groups of buses, must be controlled individually, or at least locally.
The two physical variables, system frequency and system voltage profile, are necessary,
but not sufficient, indicators of the third variable of interest, namely, system security. The
overriding objective of power system operation is to maintain the system in a secure region
of its state space. Security is defined as membership in a defined (bounded) region of state
space conditioned on the probability of imminent occurrence of some member(s) of one
or more classes of autonomous events ( contingencies ). The third indicator, then, is that
of a conditional probability (of one or more critical variables exceeding defined limits)
exceeding a defined threshold.
It is perhaps relevant to note that a major determinant of security is stability, which
implies not just containment of frequency within established bounds, but avoidance of an
unbounded departure of frequency beyond those bounds. Such loss of stability would result
from the system state trajectory moving (or being displaced) beyond its region of attraction.
Such displacement could be due to uncontrolled removal of a major component from the
system. Thus, in considering a possible loss of stability, whether of a major generator or of
a portion of the system, we are dealing with a problem of structural stability—retention of
stability following an abrupt change in system topology, obviously a discrete event.

The “Operating States,” System Status

Over the past thirty years, the importance of system security has led to the characterization
of system status in terms of a number of “operating states” (which must be distinguished
from the elements of the physical state vector, but since the use of a term as awkward as
“statuses” does not commend itself as the way to do it; generally, we will use “status” in
the singular, and “operating states” in the plural). These operating states characterize the
DISCRETE EVENTS IN POWER SYSTEMS 321

security of the system at a given time. Although DyLiacco originally defined three states
(DyLiacco, 1967), five have more generally come to be defined in terms of the equality and
inequality constraints that define system operation:
x 0 = f (x, y, p, u)
0 = g(x, y, p)
0 ≤ h(x, y, p)
(Following Venkatasubramanian et al. (1995), these equations are written as differential-
algebraic equations, where x and y are, respectively dynamic and instantaneous variables, p
represents system parameters, and u control inputs. This distinction is necessary in treating
combined modeling of mechanical (power plant) and electromagnetic (transmission plant)
performance.)
More specifically, the equality constraints require a balance between power generation and
load, and the inequality constraints require that critical variables not exceed rated normal
or emergency limits. The concept of operating states involves definition of a “security
threshold” as one of the limits to be observed (see below, under Security Control). The five
operating states are distinguished by violations of one or more of the constraints (Fink and
Carlsen, 1978), as follows:

Status Equality Constraints Inequality Constraints


Normal All satisfied All satisfied
Alert All satisfied All except security threshold
(ST) satisfied
Emergency All satisfied Additional constraints violated
Extremis Violated (system breaking up) Violated
Restorative Violated (not all load demand All except ST satisfied
being satisfied)

Disturbances (“contingencies”) may cause uncontrolled transitions from Normal status to,
successively, Alert, Emergency, or Extremis status; subsequent control actions, automatic
or manual are required to return the system to Normal. Manual system restoration is usually
required after the system has entered Extremis status.
In terms of these operating states, then, the first objective of system operation is to keep
the system trajectory in Normal Status, or if violated, to return it as quickly as possible.
Least cost operation is secondary to this primary objective.

Relevant Discrete Processes

The continuous trajectory of the system state vector is, in general, with two exceptions,
controlled and otherwise affected by discrete control actions and uncontrolled disturbances,
which may be frequent or infrequent. Relevant time scales range from milliseconds, through
322 FINK

seconds and minutes, to hours, days, and weeks or longer. The two major exceptions are the
primary control action of prime mover governors, and the control of voltages by exciters
or by static VAr (reactive power) compensators (SVCs) and other power electronic de-
vices (flexible AC transmission—FACTS—or variable impedance power system—VIPS—
devices), which may be controlled essentially continuously within feedback loops to control
a number of parameters. Due to the range of time scales that are involved, however, frequent
control actions that are inherently discrete (such as resetting of governor set points) may
reasonably be considered, over longer time scales, as continuous.
Frequent, automatic control actions include resetting of governor and exciter setpoints,
the former for frequency control, the latter for voltage control. Less frequent automatic
control actions include automatic transformer tap changing and, rarely, switching of shunt
capacitor banks to control local voltages. Infrequent but high speed automatic control
actions include operation of a wide variety of protective relays to de-energize (disconnect
from the system) equipment that is endangered by significant disturbances, or in the case
of underfrequency relays, to disconnect portions of the load that the system is losing the
ability to supply.
Infrequent, manual control actions include connection or disconnection of major equip-
ment, including generators, transmission lines, and shunt capacitors, mostly reconfiguring
the system to follow diurnal changes in load.
Frequent, autonomous disturbances, usually minor but multitudinous, are imposed by
additions to or subtractions from the system load, including residential, commercial, and
industrial classes. Mostly, load changes, which may be considered individually as micro-
scopic random point processes, aggregate into a slowing varying macroscopic random pro-
cess superimposed on a nearly periodic diurnal process. Non minor load changes include,
notably, arc furnace loads, which can impose significant, nearly instantaneous, frequent
changes in load when they are in operation.
Infrequent autonomous disturbances ( contingencies ) include unexpected sudden losses
from service of major pieces of equipment, such as generators, transformers, or transmission
lines, whether due to internal failures or damage from external causes, e.g., lightning strokes.
At this point, it may be useful, at least tentatively, to keep in mind the possible distinctions
between (i) manual discrete control actions, (ii) automatic discrete events which may be
triggered causally at apparently random intervals, and (iii) autonomous discrete events
which are or may be considered essentially random.

Interactions of Discrete Events and Continuous Processes

As was stated earlier, power system functioning is observed via three more or less diverse
variables. Two of these, the scalar system frequency and the vector system voltage profile,
are quantitative phenomena; the third, system security, is qualitative.
The problems arising from the mixed nature of the power system process are due in most
part to the interaction of discrete and continuous control processes. One major class of
such problems involves interactions or tradeoffs between costs associated with the control
functioning of the respective processes. Depending on the time scales, one is faced, in
the unit commitment problem, with the effect of start-up costs of units that are brought on
DISCRETE EVENTS IN POWER SYSTEMS 323

line to anticipate diurnal load increase, with costs (due to increased wear and consequent
maintenance) of switching energized shunt capacitors, or throttling losses incurred when
steam prime movers with sequential valving are moved to the next successive valve position.
In all of these cases, the problem is that of minimizing production cost over an appropriate
horizon when the longer run savings to be anticipated from the discrete control actions must
be traded off against the cost incurred by that action.
Another class of problems involves the counter-intuitive effects of some discrete control
actions. One example is that, in some circumstances, depending on system topology, an
overload on one transmission line may be relieved by switching off another line. It follows,
of course, that the inverse may also be true under some circumstances—that energizing one
line may result in overloading another. Another example is the sometimes counterproduc-
tive consequences of the operation of underfrequency load shedding relays. Inescapable
time delays are involved in underfrequency load shedding actions, and other delays char-
acterize the response to those actions of the system trajectory, due to system inertia. The
former delays may be allowed for in the design of load shedding programs, but the latter
will vary dependent on system load, system generation mix, and system topology at the
time of the initiating disturbance. A third example involves the potentially counterpro-
ductive effects of automatic tap changing under load of transformers during a period of
declining system voltage. Given an extensive (greater than local) shortage of reactive en-
ergy, such automatic action may increase reactive losses and accelerate a decline in system
voltage levels, but freezing such action may or may not be beneficial, depending on other
circumstances.

Mixed Control Problems

In choosing the following problems for presentation, we are motivated by several observa-
tions noted in review of the aforementioned Special Issue of the IEEE Proceedings (Ho,
1989). Ho, in “Scanning the Issue,” makes several observations characterizing DEDS. He
states that “For DEDS, the [presumably, state] trajectory is piecewise constant and event
driven. . . . Thus a sequence of two numbers (state and state holding time) basically char-
acterizes the trajectory.” He also notes that the discipline deals with “man-made dynamic
systems which are not easily described by ordinary or partial differential equations . . .
where the evolution of the system in time depends on the complex interactions of the timing
of discrete events” (emphasis added). He adds further that “the concepts of dynamics,
such as time constants, time and frequency response, controllability and observability, have
played and will continue to play important roles in the development of models and tools
for DEDS.” (Ho, 1989, pp. 1f)
Further in the Special Issue (Ho, 1989), Zeigler, discussing a possible common modeling
framework for continuous and discrete event systems, notes that any mathematical system
is characterized by “a time base, inputs, states, and outputs, and functions for determining
next states and outputs, given current states and inputs. Discrete event systems represent
certain constellations of such parameters, just as continuous systems do. For example,
the inputs in discrete event systems occur at arbitrarily spaced moments, while those in
continuous systems are piecewise continuous functions of time.” (Ho, 1989 p. 72)
324 FINK

These observations suggest that power systems, as we have described them, may be
approached, complementarily, as either continuous or discrete events system, in either
case ignoring or approximating (at the risk of seriously distorting) certain critical features,
or, comprehensively, as mixed systems at the risk of posing intractable problems. In
what follows, we attempt to characterize certain critical problems in (barely) enough detail
to encourage readers versed in DEDS studies to determine for themselves whether that
discipline can make any substantial contributions. The first two examples, in particular,
exhibit the primacy of state and state holding times together with the importance of timing
that, according to Ho, are quintessential DEDS characteristics.

Bus Voltage Control

Reactive Injection

One interaction of interest may be discussed in the context of the control of bus voltages.
Bus voltages may be controlled by the injection of reactive power (“reactive”), either locally
or at an adjacent bus. Reactive may be provided either discontinuously by switching banks
of capacitors, or continuously by controlling static VAr compensators (SVC) or, at generator
buses, by controlling generator excitation.
There is a significant difference, however, in the “quality” of the reactive provided by
each of these three sources. The amount of reactive provided by a capacitor is a direct
function of the voltage applied to that capacitor. Thus, a capacitor at a load bus may supply
reactive to local reactive loads (e.g. induction motors), and this local supply will forestall
the drawing of that reactive from the network (which would impose a greater total system
use of reactive due to reactive losses in transmission). However, if the voltage starts to sag
due to increased real power transmission, the reactive output of the capacitor will decrease,
thus aggravating the problem. In contrast, generator reactive is independent of bus voltage.
SVC output is similarly independent of bus voltage over its output range, but once it reaches
its upper limit, it reverts to the characteristics of a capacitor bank, and its output may drop
precipitously.
There are two reasons for preferring to support bus voltages insofar as possible by capacitor
output. First, considering the planning horizon, the capital cost of capacitors is less than that
of generators or SVC installations. Second, considering the operating horizon, the control
of capacitors is less responsive than that of either generators or SVCs; thus, preserving
available reserve capacity on the latter by relying on controlled capacitor banks for normal
control needs, can be vital in providing response to system emergencies, when rapid increase
in reactive injections may be the means of avoiding system instability. Thus, it becomes
desirable to consider control regimes that will take maximum advantage of the characteristics
of each source.
Besides this being a mixed regime, it is also a problem in dynamic optimization, since
frequent switching of capacitor banks becomes extremely expensive due to excessive wear
of switch contacts. Therefore, while diurnal variations in reactive demand should be tracked
by switching of capacitors, such switching should not be done in response to short term
DISCRETE EVENTS IN POWER SYSTEMS 325

fluctuations. At the same time, demands or generators for reactive in response to fluctuations
should be kept to a minimum over a reasonable time frame by resort to capacitor switching.

Unit Commitment

Unlike the preceding problems, which play out over time scales of microseconds through
minutes, unit commitment involves times scales of days, weeks, or even longer. The
problem is that of “committing” generating units (i.e., starting and connecting them to
the system) in a timely manner to satisfy anticipated load demands. Hence, we have a
problem that seems, on the surface, to be sufficiently decoupled from the continuous nature
of power systems operation as to constitute a valid candidate for DEDS treatment. The
system state is characterized by units being “on” or “off,” and the time they are in each
state is of great significance. Commitment of a unit is discrete, costly, and once effected,
not immediately reversible (physical characteristics impose both minimum “up-times” and
minimum “down-times”). The ideal solution, “just in time” commitment, is precluded, not
only by uncertainties in load forecasting, but also by the need for committed reserve (on-line
but not loaded) generation that will accommodate load imbalances that may be caused by
potential disturbances. Obviously, the unit commitment problem would not exist if units
could be “committed” partially and/or with no discrete start-up or shut-down costs, i.e., if
commitment were not discrete.
The problem has been attacked using a wide variety of optimization techniques. Sheble
and Fahd (1994) cite 15 techniques, ranging from heuristic (priority listing, exhaustive enu-
meration) through popular (dynamic programming, Lagrangian relaxation) to speculative
(simulated annealing, risk analysis, decision analysis).
We should note that, as restructuring of the industry unfolds, unit commitment may, under
some possible scenarios cease to be a concern of centralized system control, becoming
instead a matter for decision by individual generating plant owners as to whether, or when,
they wish to place their plants in operation.

Security Control

Earlier in this discussion, we defined the primary objective of system operation to be that of
maintaining the system state trajectory within Normal operating status. Broadly conceived,
this objective may be taken to characterize an overall security control problem, to which
operating cost minimization, which pertains to operation in Normal status, is subsidiary. It
follows that specific operating objectives differ according to system status at any point in
time, and that determination of the current system status is a critical function.
A formalism that defines the security assessment problem has the following features: (i) a
(vector valued) random process x(t) that describes the trajectory of system components
(machine voltages, speeds and angles, line status, compensator states, etc.); (ii) a random
process z(t) that describes the status of exogenous disturbances to the system (weather
conditions, uncontrollable load elements, accident occurrence, etc.); and (iii) a vector u(t)
of control inputs (generation dispatching, reserve actions, controllable loads, line switching,
326 FINK

etc.). Assume that these three random processes completely determine system operation
and that they can be related through
x(t) = F(x−, z−, u−)
y(t) = H (x−, z−, u(t))
u(t) = g(y(t))
The first relationship, a causal function, expresses the dependence of the current state on
past values (denoted by the “minus” signs) of the random processes. The second and third,
functional relationships, stipulate that the state and disturbance processes are only available
through noise corrupted observations, of which the controls are functionals.
At this juncture, it is relevant to point out that the three random processes, x(t), z(t),
and u(t), are all mixed. The trajectory of system components, x(t), includes machine
voltages, speeds and angles, which are continuous in nature, and line status and capacitor
status, which are discrete. Exogenous disturbances, z(t), include weather conditions, which
are continuous over short time frames (seconds, minutes) but may be considered discrete
over time frames of an hour or more, and uncontrollable load changes and accidental
disturbances which are discrete in nature. Control inputs, u(t), include normal generation
control, which is continuous, as well as unit commitment, generator tripping, load shedding,
and line switching, which are all discrete.
Within the framework of the security control problem, the set X of all operating states
consists of two distinct subsets: the subset N of “normal” and “alert” states wherein no
operating constraints are violated, and otherwise the subset M of the other three states.
Assuming that x(0) is in N , and given relevant statistics for the stochastic process z(t), we
may calculate the probability that x(t) will be in M at any future time, or as a function of
time:

P(t) = P[x(t)eM/x(t = 0), z−, u−, z(t), u(t)].

In this expression, P(t) represents the integral over the upper tail of the function z(t) for
which the joint function (z(t), u(t)) results in x(t) being in M. Calculation of P(t) over a
defined future horizon t (typically 10–15 minutes) of interest (the security horizon) requires
knowledge of the current state as well as of past and future values of the disturbance and
control processes. A Security Level of the system, given the present state and potential
future disturbances and resultant controls out to a specified horizon T , may then be defined
as

S(t = 0) = [1 − P(T )].

Define a time varying Security Threshold

d = d[t, P(z+ | z−)],

with values in the range [0, 1], where P(z+ | z−) is the distribution of the conditional
probability of z(t+) given z(t−). This in effect implies that the Security Threshold could
change if for example a major storm were approaching, which is how operators react
DISCRETE EVENTS IN POWER SYSTEMS 327

empirically today. Then, whenever S/d becomes less than unity, the system is deemed to
have entered the Alert (vulnerable, or insecure) State, and the security control problem is,
therefore, to construct the control u(t) in order to restore S/d > 1, or failing this, to achieve
the supremum of S(u+) over u+ and within future time segments.

Emergency Control

System control in the Emergency and Extremis states takes a variety of forms, some manual,
many necessarily automatic because of the rapidity with which events may unfold. Many
of the automatic controls are local; e.g., equipment that is in danger of damage may be
tripped “off line” by local relay action. We cite two examples. The first is that of one
control regime, involving automatic control of transformer taps, that may be invoked in
an attempt to forestall an incipient collapse of system voltage. The second example is
that of underfrequency load shedding, which, although in the present state of the art its
control is distributed among autonomous relays, is of system wide significance; this latter
example will be discussed in some detail because of the sparsity of its treatment in the
literature.

Voltage Collapse

Bus voltage levels are maintained by supplying sufficient reactive energy to meet demand.
Sagging voltages indicate inadequacy of supply, and if not corrected, may result in total
collapse of voltages. A variety of mechanisms are involved, including bifurcations of the
stability equilibria in state space that result in loss of stability. Of interest here is the inter-
action of continuous processes and discrete events that, under certain conditions may lead
to loss of stability. Vu and Liu (1992) have analyzed the behavior of bus voltage trajectories
following a disturbance that results in a regime of imbalances between reactive supply and
demand. The (discrete) operation of transformer tap changers, the (differentiable) dynamic
relationship between voltage levels and reactive imbalances, and the hard limit on gener-
ator exciter voltages, together partition the state space into regions within which voltage
trajectories move toward or away from stable operating points, and within which operator
intervention, such as locking out automatic tap changing, may or may not be beneficial. The
authors point out explicitly the difficulty of combining continuous and discrete variables in
a high dimensional analysis.

Underfrequency Load Shedding

Underfrequency load shedding (UFLS) is a practice used throughout the power industry.
Power systems are designed to withstand the effects of an array of “credible” worst dis-
turbances. The function of UFLS is to serve as back-up protection for the system in cases
which might occur that were not covered by the design process.
328 FINK

Fortunately, even in the emergency state, the power system is usually robust enough to
enable carefully designed heuristic load shedding strategies to be effective. It is important
however, that these strategies be designed on the basis of mature understanding of the char-
acteristics of the system involved, including system topology and dynamic characteristics
of its generation and its load. A poorly designed load shedding program may be ineffec-
tive, or worse may exacerbate stresses on the transmission network leading to its cascading
disruption. Over the years, however, utility experience and extensive studies on a number
of systems have resulted in dependable guidelines for the design of effective load shedding
programs. When plans designed on the basis of these guidelines are validated by carefully
modeled simulations, they provide reliable tools for preservation of the system under severe
conditions.
Underfrequency load shedding is a coarse tool for use in an extreme situation. As a
last-resort expedient, it should be simple, rapid, and decisive. To be effective, it should be
automatic, be distributed uniformly across the system at each step to avoid aggravating line
overloading, be locally controlled in response to local frequency to be independent of system
splitting, and should not be dependent on communication links. The shedding schedule
should not require frequent modifications to follow the continual load and system changes,
as a well-designed and well distributed scheme will be sufficiently robust to accommodate
such changes.
The major parameters that must be considered in devising an UFLS program are the level
and distribution of spinning reserve and the specific load shedding parameters: the frequency
thresholds, the total amount of load to be shed, the MW step sizes, and intentional time
delays. An important characteristic of any schedule is the amount of undershedding and/or
overshedding in which it is likely to result over the range of cases that might be encountered.
The response of system frequency to a load shedding schedule will be characterized by
an average value within a band of actual values pertaining to individual machines and
determined by the intermachine oscillations. Frequency differences across the network can
result in randomness of load shed.

Amount of Load to Be Shed

Since underfrequency load shedding is a last resort that must work if the system is to be saved
from collapse, it does not pay to be timid. If the system goes down because a particular
load was not shed, that load goes down anyway, carrying the rest of the system with it.
For an isolated (non-interconnected) system, maximum load subject to and scheduled for
shedding should, in principle, include all load. As a practical matter, perhaps 80% should
be included, some with appreciable time delay.

Frequency Threshold

The first step threshold should not be too close to normal frequency, to avoid tripping
on severe but non-emergency frequency swings. However, frequency thresholds must be
coordinated with machine protective relaying (and vice versa).
DISCRETE EVENTS IN POWER SYSTEMS 329

Step Size & Number of Steps

A significant consideration in determining step sizes is the potential for excess load shedding.
Frequency steps must be far enough apart to avoid overlap of shedding due to (intentional
or inherent) time delays. Although average excess load shed might theoretically be reduced
by increasing the number (and reducing the MW size) of steps, the need for rapid, decisive
action is more important. Moreover, there are other important considerations weighing
against more than a minimal number of steps. Step frequency thresholds too close together
may in themselves cause overshedding: a second step at one bus may be triggered before
the first step at another, due to the transient differences in bus frequencies. Again, a second
step may be triggered before system inertia responds to the previous step. Finally, since
load shed at each step should be distributed across the entire system, many small steps
requires very many more relays.

Time Delay

Some delay is needed in order both to ride out short time transient frequency excursions, and
to accommodate the response time of the system to each step of load shedding. However,
time delays should be introduced cautiously, for many of the reasons just cited for separation
of frequency thresholds.

Other Considerations

It might seem that, to be optimally effective, UFLS should be based on an accurate estimate
of the amount and location of the generation that caused the system upset, and control means
should shed that amount of load in that area. Both of these objectives are unattainable. While
system frequency in the steady- state is a sensitive indicator of any small mismatch between
load and generation within a closed system, it is not an unambiguous immediate indicator
of the magnitude of a sudden change in the load/generation balance, and no indicator at all
of the location. Even the initial rate of change of frequency is ambiguous, since individual
generators react individually according to their own inertia, and to their electrical distance
from the disturbance and from other machines. A third problem, related to the first, is
inability to determine, in the available time frame, the exact topology and dynamic state of
the system. Finally, the amount of load actually shed by relays will depend on the actual
load demand at the time of shedding, which is not likely to be equal to the potential (peak
connected) load on those feeders.
Rate of change of frequency, while it is in principle an earlier indicator of an instantaneous
change in load/generation balance than absolute frequency, is even more subject to distortion
by local dynamics than is the latter. This is another aspect wherein, because of the emergency
conditions under which UFLS is invoked, simplicity and robustness are to be preferred
to more refined, carefully tuned designs. In general, a reliable UFLS schedule can be
developed without resorting to rate of change conditioning. If for some reason it seems
330 FINK

desirable to resort to the rate of change, measurements must be regressed in order to avoid
erratic performance, which may introduce unacceptable delays.
In effect, UFLS supplements spinning reserve. For small load-generation mismatch and
without load shedding, maximum deviation from normal frequency increases sharply as
the magnitude of the mismatch approaches the amount of spinning reserve. Conversely,
as the amount of spinning reserve approaches several times the disturbance magnitude,
the maximum deviation approaches the associated steady state frequency offset. Spinning
reserve also serves to reduce expected excess shedding for any given schedule, but the effect
diminishes as size of disturbance increases. (It should be noted that the magnitude of a
frequency swing following a disturbance is very dependent on the size of the disturbance
((amount of actual generation, not generating capacity, that is lost)) relative to the concurrent
level of the load.) The effectiveness of spinning reserve is strongly affected by its distribution
over the (equivalent electrical) area of the network; and the number of, and its distribution
among, responsive generators.

Conclusions

A number of power system control problems have been described which individually exhibit
characteristics of discrete or mixed continuous/discrete system problems. Of these, perhaps
the cleanest example of a discrete problem is that of unit commitment. The reason for this
is that of the discrete nature of the problem per se, and the clean separation between the time
scale on which the unit commitment problem unfolds, which is hours to weeks, and that on
which (continuous) system operation proper unfolds, which is subseconds to hour(s). An
even cleaner separation is that between such “operation planning” problems and “system
planning” problems, which extend over years, but these have not been discussed in this
paper.
Bus voltage control and emergency control are clearly mixed problems, since no such
clean time scale separation is possible. Security control, however, is somewhat tantalizing
in this respect. The primary subproblem is that of security assessment, and that in turn
exhibits many of the distinguishing characteristics of DEDS. The primary factors are the
system status, and its dwell time within each state—that within the Normal state to be
maximized, and that within all other states to be minimized. Unfortunately, no time scale
separation can be invoked vis-a-vis other system processes.

References

DyLiacco, T. E. 1967. The adaptive reliability control system. IEEE Trans. PAS-86(5): 517–531.
Fink, L. H., and Carlsen, K. 1978. Operating under stress and strain. IEEE Spectrum. 15(3): 48–53.
Ho, Y. C. 1989. Guest Editor. Special Issue on Dynamics of Discrete Systems. IEEE Proceedings. 77(1).
Sheble, G. B., and Fahd, G. N. 1994. Unit commitment literature review. IEEE Trans. PS-9(1): 128–135.
Venkatasubrmanian, V. et al. 1995. Dynamics of large constrained nonlinear systems—a taxonomy theory. IEEE
Proc. 83(11): 1530–1561.
Vu, K. T., and Liu, C-C. 1992. Shrinking stability regions and voltage collapse in power systems. IEEE Trans.
CAS-39(4): 271–289.

Você também pode gostar