PMU Data Analytics (continued)
.
We also proposed a simple
signal processing approach to
achieve data compression and
data privatization simultaneously
for PMU data. Random noises
are added to the measurements
to protect data privacy, and the
quantization is applied after-
wards to reduce the amount
of information to transmit. The
privacy of each PMU is enhanced
because an intruder can only
observe highly quantized val-
ues even if it eavesdrops the
data communication from some
PMUs to the operator. Our major
contribution is a data recovery
method for the operator to re-
cover actual values from quan-
tized measurements of multiple
PMUs. Therefore, the reduced
data transmission, privacy en-
hancement of individual utilities,
and the information accuracy for
the central operator are achieved
simultaneously. Figure 3 shows
the original data, highly noisy
and quantized values, and the
recovered data from the quan-
tized values of two bus voltages
in recorded PMU datasets.
~end~
Faculty News
Figure 3: Original, quantized, and recovered data of
two voltage magnitudes in a recorded PMU dataset
For instance, we developed OLAP, an online miss-
ing data recovery algorithm that can fill in the
missing values in the streaming PMU data. Fig. 1
shows the interface of our OLAP method imple-
mented on openPDC. We further exploited the
low-rank property of the Hankel matrix of the
data matrix such that consecutive and simultane-
ous data losses can be correctly recovery without
modeling the power system.
We developed a novel data-driven method to
identify and locate events without modeling the
power system. The critical innovation is to charac-
terize an event by a low-dimensional row subspace
spanned by the dominant singular vectors of the
data matrix that contains spatial temporal blocks
of measurements from multiple PMUs. This sub-
space characterization is a compact representation
of system dynamics. Then an event is identified by
comparing the obtained data with a pre-computed
event dictionary with each dictionary atom cor-
responding to a row subspace of an event. Fig-
ure 2 illustrates the dictionary construction. The
subspace representation signifi cantly reduces the
dictionary size, leading to a fast and efficient event
identification method.
Dr. Joe Chow, RPI Campus Di-
rector, was elected to the Na-
tional Academy of Engineering
on February 8, 2017. Election to
the National Academy of Engi-
neering is among the highest
professional distinctions ac-
corded to an engineer. Dr. Chow
and the newly elected class will
be formally inducted during a
ceremony at the NAE’s annual
meeting in Washington, D.C., on
Oct. 8, 2017.
Above: Dr. Joe Chow
Spring 2017
CURENT Newsletter
4