28-08-2014, 03:46 PM
Power Line Carrier (PLC) Signal Analysis of Smart Meters for Outlier Detection Project Report
Power Line Carrier.pdf (Size: 799.42 KB / Downloads: 31)
Abstract
A method for identifying outliers among a set
of smart meters by using the power line carrier (PLC) signal
strength is presented in this paper. The broad goal is to use
the PLC signal as a predictor of transmission problems to
proactively avert local power outages. The proposed method uses
the PLC signal strength measured between the communication
node (transformer) and residential smart meters. The paper
presents four metrics based on the distribution of signal strengths,
with each metric identifying a class of outliers. After detecting
the set of outliers using these metrics, we identify the set of
representative good meters
INTRODUCTION
Power Line Carrier (PLC) is a standard for carrying data
on a conductor used for electric power transmission. PLC,
also called Power Line Communication, uses the existing
electricity infrastructure to carry communication signals by
providing a standardized interface [10]. The principle behind
PLC is to modulate a carrier frequency on top of existing
50/60 Hz energy carrier lines [6]. PLC has been in use since
at least the 1920s [11], and recently its use is expanding
into the electricity distribution area for load management,
control of heating, lighting, and air conditioning as well as
for broadband communication. With the recent advances in
smart grid technology, many utility companies are developing
technologies for system control of smart grids and to provide
realtime information services for customers. Automatic meter
reading (AMR) has been used to retrieve customer data using
PLC [9]. PLC has also been used to measure the quality of
power [8]. Cavdar [4] has described how PLC can be used in
remote detection of illegal electricity usage.
This paper describes a technique for identifying outliers
among a set of smart meters by using PLC signals from the
smart meters. This approach categorizes outlier meters into
four groups depending on the characteristic classes of signal
strength behavior exhibited by the smart meters. Each group
of outliers is identified by metrics described in Section III.
The application of these metrics to identify the outliers and
the complementary set of representative meters is described in
Section IV
SIGNAL STRENGTH ANALYSIS
We analyzed the Power Line Carrier (PLC) signals from
about 15,000 smart meters in residential buildings covering a
period of 54 weeks. The data was obtained from a leading
power utility company. Each week’s data is a once-a-week
snapshot of signal strength for these smart meters; the smart
meter data is actually available at 15-minute intervals. Initial
analysis of the mean signal strength and standard deviation
of signal strength (Table I) indicated that there is not much
weekly variation for the aggregate set of meters. Therefore
we looked into the distribution of signal strengths to help us
gain better insights. We focused on a data-driven approach
Z-Score Signal Analysis
Z-scores are normalized scores to compute how many
standard deviations each signal strength is away from the
mean so that we can track how much of an outlier the meter’s
signal strength is from week to week. Since we are primarily
interested in low signal strength readings, we compute the
z-score based on how many standard deviations each signal
strength is below the mean signal strength.
CONCLUSION
This paper presents metrics for detecting the set of outliers
among a set of smart meters, and shows they can be used toidentify different types of connection anomalies. The result
of applying the techniques on a fifty-four week set of data
from 14524 meters showed that roughly 677 smart meters
exhibited potential connection anomalies. Further analysis of
the anomalous meters showed that some of them were assigned
to incorrect parent nodes which caused the signals to hop
across neighboring nodes resulting in large signal drops. Note
the improvement in signal strength in Figure 1 after this was
corrected.