IEEE Power & Energy Magazine - May/June 2018 - 64

random matrix, cluster analysis, svM, ann, and expert system. finally, the development trend, locations, and severity
of the power equipment fault are forecast.
taking transformer fault diagnosis as an example, the
procedures can be divided into five steps, as shown in figure 8: 1) collect multidimensional state monitoring data;
2) calculate the correlations among these data; 3) collect a
large number of transformer fault samples; 4) use correlation analysis to obtain key feature parameters under different faults, original failure modes, final failure modes, and
fault locations; and 5) apply ann classification methods to
diagnose faults in transformers based on the multidimensional feature parameters.

Challenges in China
the challenges for big data analytics are multifaceted. there
are both technical issues originating from features of the
power system and nontechnical issue such as training or
locating sufficient technical personnel in this field. furthermore, the fast transition of china's power industry including market deregulation and high penetration of renewable
energy integration also presents new challenges.

Data Fusion and Security
Different types of data are often stored in several database
systems because the various parts of the power system are
usually run by different departments. for example, the realtime load data and monthly electricity consumption are
stored in the power dispatch and marketing departments,
respectively. the design of traditional database management
systems in power systems does not consider sharing and
interaction across departments, which presents barriers for
enabling big data thinking in power systems. this is not only
a technical issue but can also pose management difficulties
for grid companies. in the future, the multiple data fusion in
the current electric power industry will likely produce more
valuable information and lead to further applications. such
value would improve management and data sharing among
different departments.
in addition, the security issue of electricity consumption
data and power grid operation data is raising increasing concerns. cyber-attacks on smart meters and PMu devices harm
billing and privacy on the demand side. the impacts on the
stability and security of the whole power system are more significant today. ensuring the security of the massive amounts

Input
Monitoring Data

Power Grid Data

Dissolved Gas in Oil
Oil Temperature
Neutral Point Ground Current
...

Current
Voltage
Active Power
Reactive Power

Correlation Analysis Algorithms

Fault Data
Failure Mode:

Fault Location:

Overheat
Discharge
Short Circuit...

Winding
Iron Core
Casing...

Other
Data

Fault Diagnosis Model of Multidimensional Feature Parameters

Output
Key Feature Parameters

Development Trend of Fault

Overheat: DGA, Temperature, Load...

Original/Final Failure Mode

Aging: Furfural, DGA, PD...

Fault Location

Damp: Moisture in Oil, PD...

Failure Probability Distribution

Fault Diagnosis Conclusions
Failure Mode
Fault Location
Fault Severity

figure 8. A framework for transformer fault diagnosis. DGA: dissolved gas analysis; PD: partial discharge.
64

ieee power & energy magazine

may/june 2018

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - May/June 2018