Beruflich Dokumente
Kultur Dokumente
also used. A method called Committee Decision Mechanism signal. After the comparison using several Wavelet
(CDM) is developed to manage each extraction algorithms so Transforms, it was decided to use the Stationary Wavelet
a unique solution can be found. Transform (SWT). Its main advantage relies on the fact that is
time-invariant. This allows a better noise cancelation. Fig. 4
III. DEVICE DEVELOPMENT depicts a signal decomposition using SWT. In this work it was
used a mother wavelet based daubechies 4 (db4), as it had a
A. Methodology & software assembly
better performance in the load analysis when compared with
As mentioned above, the use of harmonic content spectrum other mother wavelets such as Haar.
even when combined with artificial intelligence was not
enough for a more complete evaluation of the load behavior.
Although there is an increase of nonlinear residential loads,
their harmonic content is very similar, i.e., different equipment
may present very similar current waveforms and assuming a
similar voltage at the load terminals the residential load
presents a similar power output.
Thus in this paper, we adopted an alternative way. First the
measured signal, uR ( t ) , is converted to an analytic signal,
uS ( t ) , using the Hilbert Transform [12], the phase of the
analytic signal is a time varying quantity given by (1).
φ ( t ) = arg ( u S ( t ) ) (1)
The principal value in the convolution integral of the
Hilbert Transform may present some numerical issues, thus
for a more efficient numerical procedure the discrete
analytical signal is used. In this case, the discrete analytical
signal is obtained by means of the FFT algorithm ( see Fig. 1. Measured current of three distinct types of lamps.
Appendix A ). Using (1) is possible to obtain the so-called
instantaneous frequency given below.
dφ ( t )
ωinst = (2)
dt
One interesting aspect of an analytic signal is that
instantaneous frequency presents a waveform which is
affected by all the harmonic contents in the original signal.
To illustrate the impact of the analytic signal approach,
consider the measured currents of three different lamps, each
with a different bulb. The first lamp (L.B. #1) has a compact
bulb, and the other two have conventional bulbs (L.B. #2 and
L.B. #3, from two distinct manufacturers and using different
technologies). Fig. 1 depicts the measured currents of these
three lamps and Fig. 2 and Fig. 3 show time-varying angle
associated with the measured current from each lamps and the
absolute value of the analytic signal, respectively.
From these figures, it can be seen that using an analytic
Fig. 2. Time-varying angle of the analytic signal for the measured current of
signal associated with the measured currents it is possible to three lamps.
“cluster” similar devices. Both the phase and amplitude
information of the analytic signal provide an useful Fig. 5 shows the wavelet decomposition for the absolute
information, i.e., similar devices present similar behavior and value of the analytic signal associated with the measured
small differences in the equipment structure can provide very current of a compact lamp using SWT. Each level of the SWT
different waveforms. is further analyzed using the Burg algorithm. The Burg
For the equipment identification we could use either the algorithm allows to parametric estimation of the power
phase angle of the absolute value of the analytic signal. One spectrum of a signal.
peculiar characteristic found was that the absolute value would Fig. 6 shows the Burg spectrum of each SWT level for a
create waveforms with a more pronounce maxima, while loads compact fluorescent light with a filter of order eight. Using the
with low harmonic content tend to produce smooth curves. Burg spectrum is possible to identify the maxima in the
Thus we decided to further analyze the absolute value using a spectrum of each level of the SWT decomposition. This is an
Wavelet Transform. The Wavelet Transform is used as it can important feature as it then used as a comparison parameter in
efficiently deal with discontinuities and transients in the the load identification.
Authorized licensed use limited to: Instituto Federal da Paraiba (IFPB). Downloaded on May 11,2020 at 15:30:12 UTC from IEEE Xplore. Restrictions apply.
3
Fig.6. Burg Spectrum for a compact fluorescent lamp with AR filter of 8th
Fig. 3. Absolute value of the analytic signal for the measured current of three order
lamps.
A bandpass filter was used after the voltage and current
transducers it has 95dB of damping and it was synthesized in
order to allow that and the sampling ratio did not affect
significantly the measured results. Each channel used for the
measurements had an insulation of 1 kV. The measurements
results were transmitted using an UDP protocol in an Ethernet
connection for data processing. The number of samples for
60 Hz period was 256. Fig. 7 shows the front-end used for
Fig. 4. Structure of the SWT data acquisition and Fig. 8 shows ARM Cortex platform used
for data processing.
B. Hardware Assembly
For the implementation of this methodology in a standalone
device, all the routines were coded in C++ using the Eclipse
IDE. Linux was used as a framework for the implementation
of the hardware. For the hardware assembly some details of
signal processing were needed to be treated. First all current
and voltage transducers were based on resistors to maintain an
adequate bandwidth.
Fig.8. ARM Cortex A9 platform used
Authorized licensed use limited to: Instituto Federal da Paraiba (IFPB). Downloaded on May 11,2020 at 15:30:12 UTC from IEEE Xplore. Restrictions apply.
4
The next stage was to develop a database of measured data. without interference with the consumer. The device can be
Several loads were measured, to name just a few: air installed outside the consumer residence allowing a true
conditioner, coffee machine, LED light, water drinker, LCD noninvasive assessment of the energy consumption.
TV, Stereo sound system, blender, refrigerator, conventional Even though the device has presented an excellent
(tube) TV, pump, electric shower, toaster, desktop computer, identification rate there are still some issues that might
notebook computer, blender, microwave, DVD player, lamps improve its performance. There is a difference between the
(conventional, compact ), iron, etc. right identification rate in the simulation results and
For the building of the database, the following structure experimental ones and this occurs essentially in the electronic
was used: loads. This difference needs to be explained and understood.
1. Acquire current measurements, i(t) Differences in the operation mode that may appear in the
2. Apply Hilbert transform in the measured current, i.e, electronic loads may be responsible for that mismatch and
I H ( t ) = i ( t ) + jH ( i ( t ) ) (3) further research is still need to clarify this subject.
3. Apply the SWT to the absolute value of I H and obtain Another issue not considered here is the impact of the
power factor in the identification. This information may be
the Burg spectrum of each component of the SWT
helpful to improve the identification of loads with a similar
4. Identify the peeks in the Burg Spectrum and then create a
power profile.
“dictionary” with this data
Although not shown in details in this paper we have also
VI. APPENDIX A – HILBERT TRANSFORM
considered alternative analyses using the measured power and
the “instantaneous” admittance (a time-domain transfer The Hilbert Transform, Η (.) , allows to turn a real signal in
function of the current over voltage). The results are shown in time-domain, uR ( t ) into an analytic signal, i.e., a complex
the next section.
one, uS ( t ) , i.e.,
IV. TEST CASES us ( t ) = uR ( t ) + jΗ ( uR ( t ) ) = uR ( t ) + juI ( t ) (4)
For the test cases, first we considered loads that were where the imaginary part of the analytic signal can be obtained
already in the dictionary and despite their differences it was via the principal value of the following convolution.
found that there are some similarities between all these loads 1 ∞ u R (τ )
allowing us to create two main groups, namely: uI (t ) = ∫ dτ (5)
π −∞ t − τ
A. Pumping and compressors: contains refrigerator,
The analytic signal is then a function in which its frequency
water drinker, pump, air conditioner
domain response is null for negative frequencies.
B. Electronic devices: DVD player, desktop
The discrete analytical signal is obtained as follows: let the
computer, wireless phone, notebook computer.
discrete version of the original signal, uR ( t ) , sampled through
Table 1 summarizes the results of the simulations, the level
of identification reached very good results using only currents a time T be a vector u in which the FFT is given by another
results. vector U . The discrete analytic signal is a vector W whose k
components are given by (5).
TABLE 1: Simulation results as a function of measured data considered ⎧ U [0] k =0
⎪
Group Equipment ⎪ 2U [k ] −1 ≤ k ≤ N − 1
Input data W ⎡⎣k ⎤⎦ = ⎨ 2 (6)
identification identification ⎪U ⎡⎣ N 2⎤⎦ k=N 2
Current 99% 97% ⎪ 0 N 2 +1 ≤ k ≤ N −1
Time-domain 36,28% 17,24% ⎩
admitance
Power 97,9% 95,7%
VII. REFERENCES
For the real time identification only current measurements [1] C.M. Ho, W. K. Lee, and Y.S. Hung, “Signature Representation of
data were considered. Using 10 measurements for equipment Underground Cables and its Applications to Cable Fault Diagnosis”. In:
Proc. of IEEE Int. Conf. on Advances in Power System Control,
identification, the rate of correct identification was in average Operation and Management. Dec. 1993, pp.861-865.
97.22% considering the groups and 88.8% for the equipment [2] G. P. Hanke, and D. Very, “Electric Load Monitoring and Control in the
identification. Domestic Environment”. In: Proc. of IEEE Conf. on Instrumentation
and Measurement Technology, vol.2, May 1994, pp.560-562.
[3] E. Tapia, S. Intille, and K. Larson, “Activity Recognition in home setting
V. CONCLUSIONS using simple and ubiquitous sensors”. In: Ferscha, A., Mattern, F. (eds)
Pervasive 2004. LNCS, vol.3001, pp.158-175. Springer, Heidberg
For the implementation of any smart system, a small but yet (2004).
fundamental object is the area known as smart-metering. It is [4] E. Tapia, S. Intille, L. Lopez, and K. Larson, “The design of a portable
in this area that this paper is focused. It has presented a device kit of wireless sensors for naturalistic data collection”. In: Fishkin, K.,
Schiele B., Nixon, P., Quingley, A (eds). Pervasive. LNCS, vol.3968,
that was able to obtain a very high rate of correct identification pp. 117-134. Springer, Heidelberg, 2006.
and using only current measurements the main advance of the [5] J. Fogarty, and S. Hudson, “Sensing from the Basement: A
developed equipment relies in the fact that it can be carried out Recognition”. In: The Proc of ACM Symposium on User Interface
Software Technology, ACM Press, New York, 2006.
Authorized licensed use limited to: Instituto Federal da Paraiba (IFPB). Downloaded on May 11,2020 at 15:30:12 UTC from IEEE Xplore. Restrictions apply.
5
Authorized licensed use limited to: Instituto Federal da Paraiba (IFPB). Downloaded on May 11,2020 at 15:30:12 UTC from IEEE Xplore. Restrictions apply.