Identification of power quality disturbances using the

MATLAB wavelet transform toolbox

Resende, J.W., Chaves, M.L.R., Penna, C.
Universidade Federal de Uberlandia (MG)-Brazil
e-mail: [email protected]

Abstract: This paper presents an approach that is able to in Fig. 1: the f(n) signal is passed through a low-pass digital
provide the detection and location in time as well as the filter (hd (n)) and a high-pass digital filter (gd (n)). After that,
identification of power quality problems present in both half of the signal samples are eliminated. This is indicated by
transient and steady-stable signals. The method was the symbol in Fig. 1.
developed by using the discrete wavelet transform (DWT)
analysis. The given signal is decomposed through wavelet
transform and any change on the smoothness of the
signal is detected at the finer wavelet transform
resolution levels. Later, the energy curve of the given
signal is evaluated and a relationship between this energy
curve and the one of the corresponding fundamental
component is established. The paper shows that each
power quality disturbance has unique deviations from the
pure sinusoidal waveform and this is adopted to provide
a reliable classification of the type of disturbance.
Key words: power quality disturbances, discrete wavelet
Fig. 1 – Sub-band codification scheme of a signal
transform, signal processing.
Basically, the DWT evaluation has two stages. The
1. INTRODUCTION first consists on the wavelet coefficients determination. These
coefficients represent the given signal in the wavelet domain.
It is well known that the main power quality deviations From these coefficients, the second stage is achieved with
are caused by short-circuits, harmonic distortions, notchings, the calculation of both the approximated and the detailed
voltage sags and swells, as well as transients due to load version of the original signal, in different levels of
switching. In order do correct such problems, it is required, resolutions, in the time domain.
in general that, firstly, they should be detected and identified. At the end of the first level of signal decomposition
Nevertheless, whenever the disturbance lasts for only for a (as illustrated in Fig. 1), the resulting vectors yh (k) and yg (k)
few cycles, a simple observation of the waveform in a busbar will be, respectively, the level 1 wavelet coefficients of
may not be enough to allow one to recognise that there is a approximation and of detail. In fact, for the first level , these
problem in there or, more difficult yet, to identify the sort of wavelet coefficients are called cA1 (n) and cD1 (n),
the problem. respectively, as stated bellow [1]:
On the other hand, the wavelet transform has been
adopted if different fields, such as telecommunications and
cA1 ( n ) = ∑ f ( n ). hd ( −k + 2 n ) (1.a)
k

acoustics. In the last decade the wavelet transform has been

studied to analyse voltages and currents during short duration cD1 ( n) = ∑ f ( n). gd ( − k + 2 n) (1.b)
k
disturbances. The main purpose of this paper is to show an Next, in the same way, the calculation of the
approach in which the MATLAB wavelet transform toolbox approximated (cA2 (n)) and the detailed (cD2 (n)) version
is adopted not only to detect power quality problems but also associated to the level 2 is based on the level 1 wavelet
to classify them. coefficient of approximation (cA1 (n)). The process goes on,
always adopting the “n-1” wavelet coefficient of
2. DISCRETE WAVELET TRANSFORM approximation to calculate the “n” approximated and detailed
wavelet coefficients. Once all the wavelet coefficients are
The discrete wavelet transform (DWT) is one of the known, the discrete wavelet transform in the time domain
three forms of wavelet transform. It moves a time domain can be determined. This is achieved by “rebuilding” the
discritized signal into its corresponding wavelet domain. corresponding wavelet coefficients, along the different
This is done through a process called “sub-band resolution levels. This procedure will provide the
codification”, which is done through digital filter techniques. approximated (aj (n)) and the detailed (dj (n)) version of the
In the signal processing theory, to filter a given signal f(n) original signal as well as the corresponding wavelet spectrum
means to make a convolution of this signal. This is illustrated [4].
 3. THE APPROACH DEVELOPED en_ref(j): energy distribution concentrated in each
wavelet transform level of the correspondent
By using the DWT and observing the particular fundamental component of the signal in study;
en_ref(7): energy concentrated at the level 7 (which
features of the several decomposition levels of a signal, some
concentrates the highest energy) of the corresponding
important conclusions of it can be drawn. These information fundamental component of the signal in study
can be used to detect, to locate and to classify the
disturbance. A digital program was developed and As to be shown in the next section, the “dp(j) (%) x
implemented in the wavelet toolbox of the MATLAB wavelet levels” curve of every particular power quality
platform, through five steps, as follows: disturbance has an unique pattern that can be used to
Step 1: Evaluation of the wavelet coefficients of the signal in identify the problem in the voltage waveform.
study.
Step 2: Evaluation of the square of the wavelet coefficients 4. STUDY OF CASES
found at step 1.
Step 3: Calculation of the distorted signal energy, in each 4.1) Detecting and localizing in time a disturbance:
wavelet coefficient level. The wavelet adopted in this study was the “Daub4”
The “energy” mentioned above is based on the [4], which is shown in Fig. 2.
Parseval’s theorem: “the energy that a time domain function
contains is equal to the sum of all energy concentrated in the
different resolution levels of the corresponding wavelet
transformed signal”. This can be mathematically expressed
as [2,3]:

N N J N
(2)
∑ f ( n) = ∑ a J (n ) + ∑∑ d j (n )
2 2 2

n =1 n=1 j =1 n=1

Where:
f(n): Time domain signal in study Fig. 2 – Daub4 wavelet
N: Total number of samples of the signal The Fig. 3(a) shows a voltage which was distorted by
N
2: Total energy of the f(n) signal a capacitor bank switching which has occurred at t= 400 ms
∑ f (n ) (the simulation of this switching was done in the SABER
n=1
N
: simulator [5]). Fig. 3(b) till 3(g)) summarise the steps 1 and
∑ a J (n )
2
2 of the approach, as follows. Figs. 3(c), (d), (e), (f), (g)
n =1
Total energy concentrated in the level “j ” show the squared wavelet detailed coefficients for levels 5, 4,
of the approximated version of the signal. 3, 2 and 1, respectively. Fig. 3(b) illustrates the approximated
J N : version of level 5.
∑∑ d j (n )
2

j=1 n =1
The level 1 of the transformed signal (fig. 3(g)),
Total energy concentrated in the detailed clearly shows a peak at t= 400 ms. The other wavelet levels
version of the signal, from levels “1” to “j”. have also experienced variations at this same instant. This
implies that some transient phenomena has occurred here.
Step 4: In this stage the steps 1, 2 and 3 are repeated for the Therefore, it can be said that the disturbance has been
corresponding “pure sinusoidal version” of the signal in detected and located in time. However, so far, there is no
study. sufficient evidences of what sort of disturbance occurred at
Step 5: The total distorted signal energy of the signal in this signal. This will be discussed in the following
study (found in step 3) is compared to the corresponding one subsection.
of the pure signal version (evaluated in step 4). The result of
this comparison is a deviation that can be evaluated by (3): 4.2) Identifying the disturbance:
In order to try to identify the type of disturbance present in
 en_dist( j) − en_ref( j)  (3) the voltage signal of Fig. 3(a), the the step 3 has to be work
dp( j ) (%) =  *100 out. In this step, through equation (2), the energy
 en_ref(7) 
concentrated in 10 wavelet coefficient levels is calculated
where:
j: wavelet transform level and plotted. The results are in Fig. 4(a). The 7th level holds
the biggest part of the signal energy. The 6th level also keeps
dp(j) (%): Deviation between the energy
distributions of the signal in study and its corresponding also a important parcel of it. The remaining levels practically
fundamental sinusoidal wave signal, at each wavelet do not add any important parcel to the signal energy.
transform level After this, in step 4, the fundamental component of
en_dist(j): energy distribution concentrated in each Fig. 3(a) voltage is evaluated as well as the corresponding
wavelet transform level of the signal in study energy distribution. This is illustrated in Fig. 4(b). After this,
if a visual comparison was made between Figs. 4(a) and 4(b), level The result of this is the deviation curve, illustrated in
no important differences could be observed. However, these Fig. 4(c). This curve “magnify” the deviations of the signal
figures are not equal. In order to spot these differences, the with disturbance from the corresponding pure sinusoidal one.
equation (3) is required. This equation allows the calculation The curve of Fig. 4(c) is within the “pattern” for capacitor
of the deviation between the energy distributions of the bank switching. This will be shown in the next section.
signal in study (Fig. 4(a)) and its corresponding fundamental
sinusoidal wave signal (Fig. 4(b)), at each wavelet transform

Fig. 3– Detecting a disturbance, with Daub4 – wavelet coefficients squared.;

(a): Voltage signal in study; (b): wavelet approximated coefficient for level 5,
(c) (d) (e) (f) and (g): wavelet detailed versions for levels 5, 4, 3, 2, and 1.

(a) (b) (c)

Fig. 4- Distribution of energy in 10 wavelet coefficient levels
(a) Energy Distribution for figure 2(a) voltage;
(b) Energy Distribution for the fundamental component of figure 2(a) voltage
(c) Deviation between figures 3(a) and 3(b) energy distributions
5. GENERALISING THE CLASSIFICATION PATTERN illustrate the most ordinary power quality disturbances, as
The analysis of the capacitor bank switching, shown follows: a short-circuit fault (Fig.5(b)), a notching distortion
in the previous section, can be expanded to other power (Fig.5(c)), a harmonic distortion (Fig.5(d)), a voltage sag
quality disturbances. Fig. 5 shows eight voltage signals that (Fig.5(e)), a voltage swell (Fig.5(f)), a capacitor bank
where analysed according to the steps 1 to 5. Fig. 5(a) switching transient (Fig.5(g)) and a inductive-resistive load
illustrates the perfect fundamental voltage waveform, which switching (P+jQ) (Fig.5(h)).
is adopted as a reference for the deviations experienced by
the other seven voltage signals of figure 5. The other figures

Fig. 5: Some voltage signals with a power quality disturbance

(a): Study reference: pure sinusoidal wave; (b): a short -circuit event; (c) a notching disturbance;
(d): a harmonic distortion; (e): a voltage sag; (f):a voltage swell;
(g): a capacitor bank switching transient; (h): an inductive-resistive load switching transient

All the procedures related in section 4 for the analysing 24 cases of power quality disturbance of that
capacitor bank transient switching were repeated to all Fig. 5 particular category. The amplitude and the shape of the
curves. The final results are summarised in the “Distribution curves can vary a little due to the fact that the intensity of the
of energy deviation” curves illustrated in Fig. 6. For instance, disturbance can change in accordance with the point of the
Fig. 6(a) does not show any deviation from the sinusoidal measurement. Nevertheless, each family of curves has an
signal because it refers exactly to the Fig. 6(a) pure unique signature that can be adopted for the power quality
sinusoidal. However, the other curves (Fig. 6(b),...,6(h)) disturbance recognition. This can be helpful, for instance, to
indicate different deviation patterns from the pure sinusoidal find the source of the disturbance as well as to correct it.
waveform. These features are unique for each disturbance
studied an they can be adopted as “patterns” or “signatures”
for each disturbance.
Fig. 7 can help to clarify this statement. This figure
shows seven different “curve families” of power quality
disturbances. Each one of these families were built by
 6. CONCLUSIONS pure sinusoidal waveform curve of energy. This feature is
adopted to provide a classification of the type of disturbance.
This paper has shown an approach that, by using The method presented in this paper should be also
some wavelet transform features, it is able to detect and to adopted to identify harmonic loads in distribution systems.
locate in time, as well as to classify a transient disturbance. It
was shown that the power quality disturbances have unique
deviations in their curves of energy from their corresponding

Figure 6: Devia tion in energy distributions related to figure 5 voltages

(a): Study reference: pure sinusoidal wave; (b): a short -circuit event; (c) a notching disturbance;
(d): a harmonic distortion; (e): a voltage sag; (f):a voltage swell;
(g): a capacitor bank switching transient; (h): an inductive-resistive load switching transient
 Fig. 7: Families of deviation in energy distribution, for voltages with disturbances
(a): Harmonics; (b): Notching; (c): Short-circuit transients; (d): Voltage sag; (e): Voltage swell;
(f): Capacitor bank switching transients; (g): Inductive-resistive load switching transients

7. REFERENCES
[3] – Burrus, C. S., Gopinath, A. R., Wickerhauser, M. V.,
[1] – Robertson, D.C., Camps, O. I., Mayer, J.S., Wavelets Wavelets and time frequency analysis , Proceedings of the
and electromagnetic power system transients , IEEE IEEE, vol. 84, No. 4, April 1996, pp. 523-540.
Transactions on Power Delivery, Vol. 11, No. 2, April 1996,
p.924-930. [4] – Penna, C., Detection and classification of power quality
disturbances using the wavelet transform – M. Sc.
[2] – Gaouda, A. M., Salama, M. M. A., Sultan M. R., Dissertation, june 2000, Universidade Federal de Uberlandia,
Chikhani, A . Y., Power quality detection and classification Brazil.
using wavelet-multiresolution signal decomposition, IEEE
Transactions on Power Delivery, Vol. 14, No. 4, October [5] – SABER design applications reference, Analogy Inc.,
1999, p. 1469-1476. july, 1997.