.
Voltage flicker is a problem that has existed in the power industry for
many years. Many types of enduse equipment can create voltage flicker, and
many types of solution methods are available. Fortunately, the problem is not
overly complex, and it can often be analyzed using fairly simple methods. In
many cases, however, solutions can be expensive. Perhaps the most difficult
aspect of the voltage flicker problem has been the development of a widely
accepted definition of just what "flicker" is and how it can be quantified
in terms of measurable quantities.
To electric utility engineers, voltage flicker is considered in terms of magnitude
and rate of change of voltage fluctuations. To the utility customer, however,
flicker is considered in terms of "my lights are flickering." The
necessary presence of a human observer to "see" the change in lamp
(intensity) output in response to a change in supply voltage is the most complex
factor for which to account. Significant research, dating back to the early
twentieth century, has been devoted to establishing an accurate correlation
between voltage changes and observer perceptions. This correlation is essential
so that a readily measurable quantity, supply voltage, can be used to predict
a human response.
The early work regarding voltage flicker considered voltage flicker to be
a singlefrequency modulation of the power frequency voltage. Both sinusoidal
and square wave modulations were considered as shown mathematically in Equations
1 and 2, with most work concentrating on square wave modulation.
(eq.1)
(eq.2)
Based on Equations 1 and 2, the voltage flicker magnitude can be expressed
as a percentage of the rootmeansquare (rms) voltage, where the term "V" in
the two equations represents the percentage.
While both the magnitude of the fluctuations ("V") and the "shape" of
the modulating waveform are obviously important, the frequency of the modulation
is also extremely relevant and is explicitly rep resented as _m. For sinusoidal
flicker (given by Equation 1), the total waveform appears as shown in FGR.
1 with the modulating waveform shown explicitly. A similar waveform can be
easily created for squarewave modulation.
To correlate the voltage change percentage, V, at a certain frequency, _m,
with human perceptions, early research led to the widespread use of what is
known as a flicker curve to predict possible observer complaints. Flicker curves
are still in widespread use, particularly in the United States. A typical flicker
curve is shown in FGR. 2 and is based on tests conducted by the General Electric
Company. It’s important to realize that these curves are developed based on
square wave modulation. Voltage changes from one level to another are considered
to be "instantaneous" in nature, which may or may not be an accurate
representation of actual equipmentproduced voltage fluctuations.
The curve of FGR. 2 requires some explanation in order to understand its application.
The "threshold of visibility" corresponds to certain fluctuation
magnitude and frequency pairs that rep resent the borderline above which an
observer can just perceive lamp (intensity) output variations in a 120 V, 60
Hz, 60 W incandescent bulb. The "threshold of irritation" corresponds
to certain fluctuation magnitude and frequency pairs that represent the borderline
above which the majority of observers would be irritated by lamp (intensity)
output variations for the same lamp type. Two conclusions are immediately apparent
from these two curves: (1) even small percentage changes in supply voltage
can be noticed by persons observing lamp output, and (2) the frequency of the
voltage fluctuations is an important consideration, with the frequency range
from 6 to 10 Hz being the most sensitive.
Most utility companies don’t permit excessive voltage fluctuations on their
system, regardless of the frequency. For this reason, a "typical" utility
flicker curve will follow either the "threshold of irritation" or
the "threshold of visibility" curve as long as the chosen curve lies
below some established value (2% in FGR. 2). By requiring that voltage fluctuations
not exceed the "borderline of visibility" curve, the utility is insuring
conservative criteria that should minimize potential problems due to volt age
fluctuations.
===
FGR. 1 Sinusoidal voltage flicker.
===
To other customers, Zxfmr, Zsource, Vsource, Fluctuating load
FGR. 3 Example circuit for flicker calculations.
===
 Threshold of irritation
 Threshold of visibility
 "Typical" flicker curve
 Fluctuations per hour
 Fluctuations per minute
 Fluctuations per second
FGR. 2 Typical flicker curves.
===
For many years, the generic flicker curve has served the utility industry
well. Fluctuating motor loads like car shredders, wood chippers, and many others
can be fairly well characterized in terms of a duty cycle and a maximum torque.
From this information, engineers can predict the magnitude and frequency of
voltage changes anywhere in the supplying transmission and distribution system.
Voltage fluctuations associated with motor starting events are also easily
translated into a point (or points) on the flicker curve, and many utilities
have based their motor starting criteria on this method for many years. Other
loads, most notably arcing loads, cannot be represented as a single flicker
magnitude and frequency term. For these types of loads, utility engineers typically
presume either worstcase or most likely variations for analytical evaluations.
Regardless of the type of load, the typical calculation procedure involves
either basic load flow or simple voltage division calculations. FGR. 3 shows
an example positive sequence circuit with all data assumed in perunit on consistent
bases.
For fluctuating loads that are best represented by a constant power model
(arc furnaces and load torque variations on a running motor), basic load flow
techniques can be used to determine the fullload and noload (or "normal
condition") voltages at the "critical" or "point of common
coupling" bus where other customers might be served. For fluctuating loads
that are best represented by a constant impedance model (motor starting), basic
circuit analysis techniques readily provide the fullload and noload ("normal
condition") voltages at the critical bus. Regardless of the modeling and
calculation procedures used, equations similar to Equation 3 can be used to
determine the percentage voltage change for use in conjunction with a flicker
curve. Of course, accurate information regarding the frequency of the assumed
fluctuation is absolutely necessary. Note that Equation 3 represents an oversimplification
and should therefore not be used in cases where the fluctuations are frequent
enough to impact the average rms value (measured over several seconds up to
a minute). Other more elaborate formulas are available for these situations.
%Voltage change = (3)
From a utility engineer's viewpoint, the decision to either serve or deny
service to a fluctuating load is often based on the result of Equation 3 (or
a more complex version of Equation 3) including information about the frequency
at which the calculated change occurs. From this simplified discussion, several
questions arise:
1. How are fluctuating loads taken into account when the nature of the fluctuations
is not constant in magnitude?
2. How are fluctuating loads taken into account when the nature of the fluctuations
is not constant in frequency?
3. How are static compensators and other high response speed mitigation devices
included in the calculations?
===
FGR. 4 Poorly timed motor starter voltage fluctuation.
FGR. 5 Adaptivevar compensator effects.
===
Input transformer Block 1 Block 2 dB 1 3
0.05 0 8.8 Range selector
0.5 1.0 2.0 5.0 10.0 20.0
%
?V V Hz Weighting filters 35 Hz Block 3 Block 4 Block 5 A/D converter sampling
rate =50 Hz
Squaring multiplier Squaring and smoothing Square rooter ?  v
1 min integrator Statistical evaluation of flicker level Programming of short
and long observation periods Output and data display and recording Output
5 recording Output 4 short time integration Output 3 range selection Output
2 weighted voltage fluctuation 64 level classifier Output interfaces 1st order
sliding mean filter Simulation of lampeye brain response Detector and gain
control Demodulator with squaring multiplier Signal generator for calibration
checking Input voltage adaptor rms Meter Output 1 half cycle rms voltage
indication
FGR. 6 Flicker meter block diagram.
===
Changes (min)
FGR. 7 Threshold of irritation flicker curve and Pst = 1.0 curve from a flicker
meter.
===
FGR. 8 Shortterm flicker severity example plot.
===
As examples, consider the rms voltage plots (on 120 V bases) shown in FGR.
4 and 5. FGR. 4 shows an rms plot associated with a poorly timed twostep reducedvoltage
motor starter. FGR. 5 shows a motor starting event when the motor is compensated
by an adaptivevar compensator. Questions 13 are clearly difficult to answer
for these plots, so it would be very difficult to apply the basic flicker curve.
In many cases of practical interest, "rules of thumb" are often
used to answer approximately these and other related questions so that the
simple flicker curve can be used effectively. However, these assumptions and
approaches must be conservative in nature and may result in costly equipment
modifications prior to connection of certain fluctuating loads. In modern environment,
it’s imperative that endusers operate at the least total cost. It’s equally
important that enduse fluctuating loads not create problems for other users.
Due to the conservative and approximate nature of the flicker curve methodology,
there is often significant room for negotiation, and the matter is often not
settled considering only engineering results.
For roughly three decades, certain engineering groups have recognized the
limitations of the flicker curve methods and have developed alternative approaches
based on an instrument called a flicker meter. This work, driven strongly in
Europe by the International Union for Electroheat (UIE) and the International
Electrotechnical Commission (IEC), appears to offer solutions to many of the
problems with the flicker curve methodology. Many years of industrial experience
have been obtained with the flicker meter approach, and its output has been
wellcorrelated with complaints of utility customers. At this time, the Institute
of Electrical and Electronics Engineers (IEEE) is working toward adopting the
flicker meter methodology for use in North America.
The flicker meter is a continuous time measuring system that takes voltage
as an input and produces three output indices that are related to customer
perception. These outputs are: (1) instantaneous flicker sensation, Pinst,
(2) shortterm flicker severity, Pst, and (3) longterm flicker severity, Plt.
A block diagram of an analog flicker meter is shown in FGR. 6.
The flicker meter takes into account both the physical aspects of engineering
(how does the lamp
[intensity] output vary with voltage?) and the physiological aspects of human
observers (how fast can the human eye respond to light changes?). Each of the
five basic blocks in FGR. 6 contribute to one or both of these aspects. While
a detailed discussion of the flicker meter is beyond the scope of this section,
the function of the blocks can be summarized as follows.
Blocks 1 and 2 act to process the input voltage signal and to partially isolate
only the modulating term in Equations 1 or 2. Block 3 completes the isolation
of the modulating signal through complex filtering and applies frequencysensitive
weighting to the "pure" modulating signal. Block 4 models the physiological
response of the human observer, specifically the shortterm memory tendency
of the brain to correlate the voltage modulating signal with a human perception
ability. Block 5 performs statistical analysis on the output of Block 4 to
capture the cumulative effects of fluctuations over time.
The instantaneous flicker sensation is the output of Block 4. The short and
longterm severity indices are the outputs of Block 5. Pinst is available as
an output quantity on a continuous basis, and a value of 1.0 corresponds with
the threshold of visibility curve in FGR. 2. A single Pst value is available
as an output every 10 min, and a value of 1.0 corresponds to the threshold
of irritation curve in FGR. 2. Of course, a comparison can only be made for
certain inputs.
For square wave modulation, FGR. 7 shows a comparison of the "irritation
level" given by IEEE Std. 141 (Red Book) and that level predicted by the
flicker meter to be "irritating" (Pst = 1.0). For these comparisons,
the lamp type used is a 120 V, 60 Hz, 60 W incandescent bulb. Note that the
flicker curve taken from IEEE Std. 141 is essentially identical to the "borderline
of irritation" curve given in FGR. 2.
As FGR. 7 clearly demonstrates, the square wave modulation voltage fluctuations
that lead to irritation are nearly identical as predicted by either a standard
flicker curve or a flicker meter.
The real advantage of the flicker meter methodology lies in that fact that
the continuous time measurement system can easily predict possible irritation
for arbitrarily complex modulation waveforms.
As an example, FGR. 8 shows a plot of Pst over a 3day period at a location
serving a small electric arc furnace. (Note: In this case, there were no reported
customer complaints and Pst was well below the irritation threshold value of
1.0 during the entire monitoring period.) Due to the very random nature of
the fluctuations associated with an arc furnace, the flicker curve methodology
cannot be used directly as an accurate predictor of irritation levels because
it’s appropriate only for the "sudden" voltage fluctuations associated
with square wave modulation. The tradeoff required for more accurate flicker
prediction, however, is that the inherent simplicity of the basic flicker curve
is lost.
For the basic flicker curve, simple calculations based on circuit and equipment
models in FGR. 3 can be used. Data for these models is readily available, and
timetested assumptions are widely known for cases when exact data are not
available. Because the flicker meter is a continuoustime system, continuoustime
voltage input data is required for its use. For existing fluctuating loads,
it’s reasonable to presume that a flicker meter can be connected and used to
predict whether or not the fluctuations are irritating. However, it’s necessary
to be able to predict potential flicker problems prior to the connection of
a fluctuating load well before it’s possible to measure anything.
There are three possible solutions to the apparent "prediction" dilemma
associated with the flicker meter approach. The most basic approach is to locate
an existing fluctuating load that is similar to the one under consideration
and simply measure the flicker produced by the existing load. Of course, the
engineer is responsible for making sure that the existing installation is nearly
identical to the one pro posed. While the fluctuating load equipment itself
might be identical, supply system characteristics will almost never be the
same.
Because the shortterm flicker severity output of the f licker meter, Pst
, is linearly dependent on voltage fluctuation magnitude over a wide range,
it’s possible to linearly scale the Pst measurements from one location to predict
those at another location where the supply impedance is different. (In most
cases, voltage fluctuations are directly related to the supply impedance; a
system with 10% higher supply impedance would expect 10% greater voltage fluctuation
for the same load change.) In evaluations where it’s not possible to measure
another existing fluctuating load, other approaches must be used.
If detailed system and load data are known, a timedomain simulation can be
used to generate a continuoustime series of voltage data points. These points
could then be used as inputs to a simulated flicker meter to predict the shortterm
flicker severity, Pst. This approach, however, is usually too intensive and
timeconsuming to be appropriate for most applications. For these situations, "shape
factors" have been proposed that predict a Pst value for various types
of fluctuations.
Shape factors are simple curves that can be used to predict, without simulation
or measurement, the Pst that would be measured if the load were connected.
Different curves exist for different "shapes" of voltage variation.
Curves exist for simple square and triangular variations, as well as for more
complex variations such as motor starting. To use a shape factor, an engineer
must have some knowledge of (1) the magnitude of the fluctuation, (2) the shape
of the fluctuation, including the time spent at each voltage level if the shape
is complex, (3) rise time and fall times between voltage levels, and (4) the
rate at which the shape repeats. In some cases, this level of data is not available,
and assumptions are often made (on the conservative side). It’s interesting
to note that the extreme of the conservative choices is a rectangular fluctuation
at a known frequency; which is exactly the data required to use the basic flicker
curve of FGR. 2.
Using either the flicker curve for simple evaluations or the flicker meter
methodology for more complex evaluations, it’s possible to predict if a given
fluctuating load will produce complaints from other customers. In the event
that complaints are predicted, modifications must be made prior to granting
service. The possible modifications can be made either on the utility side
or on the customer (load) side (or both), or some type of compensation equipment
can be installed.
In most cases, the most effective, but not least cost, ways to reduce or eliminate
flicker complaints are to either (1) reduce the supply system impedance of
the whole path from source to fluctuating load or (2) serve the fluctuating
load from a dedicated and electrically remote (from other customers) circuit.
In most cases, utility revenue projections for customers with fluctuating loads
don’t justify such expenses, and the burden of mitigation is shifted to the
consumer.
Customers with fluctuating load equipment have two main options regarding
voltage flicker mitigation. In some cases, the load can be adjusted to the
point that the frequency(ies) of the fluctuations are such that complaints
are eliminated (recall the frequencysensitive nature of the entire flicker
problem). In other cases, direct voltage compensation can be achieved through
highspeed static compensators.
Either thyristorswitched capacitor banks (often called adaptive var compensators
or AVCs) or fixed capacitors in parallel with thyristorswitched reactors (often
called static var compensators or SVCs) can be used to provide voltage support
through reactive compensation in about one cycle. For loads where the main
contributor to a large voltage fluctuation is a large reactive power change,
reactive compensators can significantly reduce or eliminate the potential for
flicker complaints. In cases where voltage fluctuations are due to large real
power changes, reactive compensation offers only small improvements and can,
in some cases, make the problem worse.
In conclusion, it’s almost always necessary to measure/predict flicker levels
under a variety of possible conditions, both with and without mitigation equipment
and procedures in effect. In very simple cases, a basic flicker curve will
provide acceptable results. In more complex cases, however, an intensive measurement,
modeling, and simulation effort may be required in order to minimize potential
flicker complaints.
While this section has addressed the basic issues associated with voltage
flicker complaints, prediction, and measurement, it’s not intended to be allinclusive.
A number of relevant publications, papers, reports, and standards are given
for further reading, and the reader should certainly consider these documents
carefully in addition to what is provided here.
