E3S Web of Conferences 136 , 0 2012 (2019) https://doi.org/10.

1051/e3sconf/2019136020 12
ICBTE 2019

Research on Power System Relay Protection Method Based on

Machine Learning Algorithm
Jingying Fang1*, Xiangyun Zhang2
1School of Automation, Central South University, Changsha, Hunan, 410083, China
2 Departmentof Software Engineering, Harbin University of Science and Technology Rongcheng Campus, Rongcheng, Shandong,
264300, China

Abstract. With the development of power industry, there are gradually high permeability distributed energy
systems. However, the existing relay protection is difficult to be effectively applied in this type of power
system. To solve this problem, this paper applies machine learning algorithm to power system relay protection.
Firstly, the structure of power system with high permeability and distributed energy is analyzed, and the
challenges which current relay protection algorithms faced are introduced in detail. Then, the artificial
intelligence algorithm is introduced, and the machine learning algorithm in artificial intelligence algorithm
and its application in power system are mainly studied. Finally, the power system relay protection based on
machine learning algorithm is deeply studied, and the specific implementation method and implementation
flow are designed. The machine learning algorithm studied in this paper is helpful to the development of
technology in the field of power system relay protection.

1 Introduction Due to the limitation of power and voltage levels, new

energy generation usually forms a micro-grid and is
After the distributed energy is connected to the power connected to the distribution network by a contact switch,
system, the original one-way power flow mode has as shown in Figure 1. New energy sources such as micro
changed into two-way power flow mode, which has gas turbines, wind power and photovoltaic power
changed the structure and power flow of the power system. generation often produce direct current electric energy,
Especially when the distributed energy is highly which is not completely matched with the current
permeated into the power system, the operation mode of distribution network. Therefore, it is necessary to convert
the power system is more complex, and the traditional the electric energy into the matching electric energy with
relay protection method can not be fully applied to the the access distribution network through the power
power system with distributed energy[1-2]. In view of the electronic components as the core of the power converter.
fact that the traditional protection algorithm is difficult to
apply to the development of power system, this paper
applies machine learning algorithm to the relay protection
of power system.

2 New Problems of Relay Protection

2.1 Distributed Energy

With the continuous depletion of coal resources and

pollution problems caused by coal power generation, new
energy generation such as photovoltaic and wind power
has become a green energy mode with potential for
development, which has been paid more and more
attention by various countries. Depending on their Figure 1. Power system structure with distributed energy
geographical and climatic advantages, various forms of
new energy generation can be developed according to Because the new energy generation is easy to be
local conditions, and connected to the main power grid to constrained by the natural environment, it has the
complement each other[3]. characteristics of instability, intermittence and
fluctuation[4]. Usually it is equipped with energy storage
equipment such as supercapacitors, which absorbs excess

*Corresponding author’s e-mail: [email protected]

© The Authors, published by EDP Sciences. This is an open access article distributed under the terms of the Creative Commons Attribution License 4.0
(http://creativecommons.org/licenses/by/4.0/).
E3S Web of Conferences 136 , 0 2012 (2019) https://doi.org/10.1051/e3sconf/2019136020 12
ICBTE 2019

power when the power is surplus, and provides power With the development of national economy, internet, big
when the power is insufficient, so as to maintain the data and supercomputer, the development of artificial
stability of the power system. Traditional main power grid intelligence technology has entered a new stage featuring
has relatively fixed operation mode, and the size and in-depth learning, cross-border integration, man-machine
direction of power flow are easy to determine, which is cooperation and self-control of group intelligence
conducive to the realization of power system relay development. These artificial intelligence technologies
protection. The new energy is connected to the main with new characteristics will have a significant and far-
power grid in the form of micro-grid, which makes the reaching impact on human production, life and even
operation mode of power system more complex, increases thinking mode.
the randomness and uncertainty of power flow distribution,
and makes the traditional relay protection method of
3.2 Machine Learning Algorithms and Their
power system not necessarily fully applicable, so a new
Electric Power Applications
method is needed for in-depth analysis.
As the core of artificial intelligence, machine learning
2.2 Challenges of Relay Protection Algorithms (ML) has been applied more and more in dealing with
complex systems. Artificial intelligence algorithms such
Distributed energy is connected to the main grid by high as machine learning specialize in how computers simulate
permeability, which makes the protection mode of power or implement human learning behavior in order to acquire
system more complex. The improvement measures of new knowledge or skills, reorganize existing knowledge
relay protection can help to reduce the difficulty of relay structures and continuously improve their performance.
protection, but it can not completely eliminate the impact Machine learning relies on good data sources and needs
of distributed energy on traditional relay protection, good data support. At present, with the rapid development
mainly in the following aspects. of large data and super-performance computer technology,
1)The short-circuit current in the network will machine learning has improved a larger number of better
decrease or increase with the access of distributed training methods, which is conducive to the use of
generation, which may lead to the maloperation, rejection machine learning to achieve complex power advanced
or loss of selectivity of protection in the distribution applications [5]. This paper designs an advanced
network. application of power machine learning algorithm based on
2)For the small current grounding system, the grid- power big data support platform. With the help of large
connected transformer should adopt ∆/Yg connection power data, it can provide data source for machine
mode, in which the system side is ∆ connection mode; for learning, so that advanced applications of power system
the small resistance grounding system, the grid-connected such as effective relay protection can be obtained through
transformer should adopt Yg/∆ connection mode, in which machine learning algorithm [6].
the system side is Yg connection mode. For low current With the development of smart grid, more sensor
grounding system, the access of distributed generation has elements and various types of smart electronic devices,
no effect on zero sequence protection based on steady- smart meters are equipped in the power field. The smart
state characteristic fault detection algorithm. electronic devices and smart meters collect real-time data
3)The type of distributed generation, short circuit of current, voltage and switching status in the power
capacity, grid-connected position and neutral grounding operation site. After pretreatment, the collected data will
impedance of grid-connected transformer are all important be sent to the large data support platform of power through
factors affecting the protection of distribution network. the communication network. The power big data support
The main power grid with distributed energy has changed platform first carries on the power big data collection, the
the power distribution and impedance distribution of the power big data analysis and the power big data storage.
traditional power grid. The classical relay protection On the one hand, the stored large power data are stored in
method is difficult to fully adapt to the relay protection the knowledge database, and the functions of query,
with distributed energy. It is urgent to introduce advanced record and diagnosis of the knowledge database are
algorithms such as artificial intelligence into complex provided. On the other hand, the collection and analysis
applications such as relay protection of power system with results of the large power data are imported into the
distributed energy to improve the accuracy of protection. machine learning algorithm, and the advanced application
functions of the power system are completed by
combining the experience provided by the knowledge
3 Machine Learning Method database.

3.1Artificial Intelligence Algorithms

4 Power system relay protection based
The term artificial intelligence has been proposed since on machine learning algorithm
1956, and the emergence of electronic computers as a tool
to simulate human thinking has enabled the artificial
4.1 Implementation Method
intelligence technology to have a platform for its
exploration and development. Artificial intelligence Power system relay protection based on machine learning
technology is an innovative technology leading the future. algorithm can give full play to the advantages of machine

2
E3S Web of Conferences 136 , 0 2012 (2019) https://doi.org/10.1051/e3sconf/2019136020 12
ICBTE 2019

learning in dealing with the diversity and complexity of 3) The relay protection of power system based on
power system relay protection with distributed energy. machine learning algorithm is deeply studied, and the
Machine learning algorithm collects training data of relay realization method and process are introduced in detail.
protection called from knowledge database and real-time
operation data obtained from relay protection field.
The algorithms of probability theory, statistics,
approximation theory, convex analysis and algorithm
complexity theory are used to train power system model.
The fault characteristics of power system with distributed
energy are analyzed in detail, as shown in Figure 2.

Figure 2. Application of Machine Learning Algorithms in

Power System Relay Protection Figure 3. Implementation process

4.2 Implementation Flow References

The application process of power system relay protection 1. Blaabjerg, F., Teodorescu, R., Liserre, M., Timbus,
based on machine learning algorithm is shown in Figure 6. AV. (2006) Overview of control and grid
Firstly, it collects information related to relay protection synchronization for distributed power generation
such as electric quantity and switch quantity of power systems. IEEE T IND ELECTRON., 53(5): 1398-
system, pre-processes information, and eliminates the 1409.
misinformation generated during field transmission or
communication network transmission. Then, it is judged 2. Carrasco, JM., Franquelo, LG., Bialasiewicz, JT.,
whether the collected information is sufficient for the Galvan, E., Portillo, R., Prats, MM., Leon, JI.,
machine learning algorithm, and if the amount of data is Moreno-Alfonso, N. (2006) Power-electronic
not enough, the relay protection information required by systems for the grid integration of renewable
the machine learning algorithm is continuously collected; energy sources: A survey. IEEE T IND
once the relay protection information required by the ELECTRON., 53(4): 1002-1016.
machine learning algorithm is sufficient, the The machine 3. Lopes, JAP., Hatziargyriou, N., Mutale, J ., Djapic,
learning algorithm is performed to extract the power P., Jenkins, N. (2007) Integrating distributed
failure characteristics required for relay protection, and generation into electric power systems: A review of
further to identify the relay protection. Finally, it is judged drivers, challenges and opportunities. ELECTR
whether the power system has a fault, and if there is no POW SYST RES., 77(9): 1189-1203.
fault, it returns; if there is a fault, the relay protection 4. Huang, AQ., Crow, ML., Heydt, GT., Zheng, JP.,
action such as tripping or alarm is taken. Dale, SJ. (2007) The Future Renewable Electric
Energy Delivery and Management (FREEDM)
System: The Energy Internet. P IEEE., 99(1): 133-
5 Conclusion 148.
With the distributed energy and high permeability main 5. Wan, C., Xu, Z., Pinson, P., Dong, ZY., Wong, KP.
grid, the operation of power system becomes more (2014) Probabilistic Forecasting of Wind Power
complex and the relay protection is more difficult. In this Generation Using Extreme Learning Machine.
paper, the machine learning method is applied to the relay IEEE T POWER SYST., 29(3): 1033-1044.
protection of power system, and the large data method is 6. Morales, JA., Orduna, E., Rehtanz, C. (2014)
used to provide data support for machine learning, so as to Classification of lightning stroke on transmission
achieve effective and reliable relay protection of power line using multi-resolution analysis and machine
system. The main contents are as follows. learning. INT J ELEC POWER., 58: 19-31.
1) The structure of power system with distributed
energy is introduced, and the challenges faced by
traditional relay protection algorithms are studied.
2) Analyse the characteristics of artificial intelligence
algorithm, especially machine learning algorithm, and
study the application of machine learning algorithm in
power system.