Download as pdf or txt
Download as pdf or txt
You are on page 1of 5

Self-healing of Active Distribution Network With

Inverter-Based Distributed Generators Based on Big


Data
Zhiyong Zhu Bin Duan, Member, IEEE, Dandan Liu, Yujie Feng, and
College of Information Engineering, Xiangtan University Zhuang Yang
Xiangtan 411105, Hunan, China Collaborative Innovation Center of Wind Power Equipment
zhuzyxtu@163.com and Energy Conversion
Xiangtan, China

Abstract—Considering the increasing penetration rate of the improve the reliability of ADN service restoration is very
inverter-based distributed generators (IBDGs) in the power important. In [6], solve the problem of distribution network
system, IBDGs operated in islanded mode have a new effect on self-healing by using the branch and bound method. In [7],
self-healing of distribution network, and self-healing scheme in solve self-healing problems by using immune ant colony
active distribution network (ADN) with IBDGs based on big algorithm. In [8], the voltage direct and quadrature axis
data is proposed. The big data storage and mining technology is components are used as features for the fault detection and
used to accomplish fault discrimination, fault localization, isolation in a IBDG microgrid.
isolation and service restoration. In this paper, the use of a new
type of the relay which takes advantage of the enhanced During the operation of AND, massive heterogeneous and
processing techniques and communication infrastructure, both mu lti-state data, namely the big data, are generated. Store the
of which are recently becoming available for power networks big data efficiently, reliab ly and cheaply and access and
application. The function layout and communication process analyze them rapidly are important. However, the traditional
among relays in autonomous region are in detailed study, and data processing technology cannot meet the requirements.
the fault section can accomplish self-healing through Therefore, big data processing technique was adopted to
communication and negotiation among relays. As illustrated in process the massive heterogeneous and multi-state data in this
the simulation result, the proposed scheme can ensure the paper.
reliability and security of ADN service restoration.
The paper studies the influences caused by IBDGs on
Index Terms—Active distribution network, data mining, big ADN. Simultaneously, an ADN fault optimization model is
data, self-healing. proposed combined with line capacity and priority of power
supply restore and also operation mode and control strategies
I. INT RODUCT ION of IBDGs. The objective function and constraints are
demonstrated in the proposed model. Based on this , the AND
With the development of smart grid technology, more and fault self- recovery scheme is designed according to the three-
more distributed generations (DGs) are connected in phase grounding fault of feeders. Results of simulation show
distribution network. DGs which are connected to the grid that the proposed scheme can accomplish AND fault self-
casually in a large scale may exert adverse effects on steady recovery quickly, proactively and effectively.
operation and power quality of the entire power system. When
fault occurs caused by severe external interference (e.g., The rest of this paper is structured as follows. Section II
natural disasters, equipment failures, and demand / supply shows the system design and methodology. Section III
fluctuations), it is urgent need to protect the power supply of presents the self-healing process. Section IV is simulat ion
important loads through nearby power source [1], [2]. experiment. Concluded with a summary is presented in
Section V.
ADN refers to distribution network which has the ability
of co mbinable controlling various IBDGs, and effectively II. M ET HODOLOGY AND T HE SYST EM DESIGN
connects generation with consumers' need through active
controlling and management [3], [4], [5]. Co mpared with the A. Test System
traditional distribution network, the most notable feature of Fig. 1 shows the architecture of ADN with IBDGs. As
ADN is the ability to take the initiat ive to adjust and control shown in Fig. 1, each feeder power supply area can be
the access of IBDGs. In order to reduce the impact caused by regarded as an autonomous region, a number of autonomous
power failure to society, study of the self-healing of ADN and regions constitute the entire ADN. Autonomous region

T his research was supported by National Natural Science Foundation of


China (NSFC) (No.61170191, 61379063)

978-1-5090-5167-0/16/$31.00 ©2016 IEEE


composed of IBDGs, distribution management system, power practice, the relays can send and receive informat ion which
electronics equipment, communications equipment, and power may help detect any abnormal events in the network. In this
consumers. paper, communication among relays using fiber self-recovery
network based on GOOSE. In addition, it is assumed that
Communication line
Power transmission line Control Center
communication channels are available to transmit informat ion
Active control fro m each relay to the other relays in the network, and that any
delays and errors are not significant enough to affect the
System Supporting Platform
proposed scheme [10].
= = =
Large B. System Supporting Platform Based on Big Data
power
Bus
network Data Acquisition
Fault
SCADA GIS Sensor knowledge
base
HV / MV
Large data connector

Load Data Storage


Relays = = = Distributed database
IBDGs
Distributed file system

Figure 1. Model of AND with IBDGs based on relays De-noising, Normalized, Feature extraction,
Data dimension reduction

Fig. 2 shows the autonomous region (show in Fig. 1) Data Calculation


Stream data Memory batch
MapReduce
processing processing
interconnection diagram of any three identical bus voltages.
The system consists of three substations, three bus system,
Training samples to obtained classification
three feeders, four tie-line, twenty-four digital relays, three models by the random forest algorithms.

loads and two identical inverter-based DGs.


Data Analysis
Substation 1 Substation 2 Substation 3

B1 B2 B4 B5 B7 B8 Application Fault discrim ination Fault localization


Bus
B3 B6 Fault isolation Service restoration
S4 B9
S1 S7
A
TS1 Load 1 Figure 3. System supporting platform based on big data
I1 Control Unit
IBDGs 1
(CU) S8 Fig. 3 shows the system supporting platform based on big
S2 S5 data. The platform uses data mining technology, distributed
I2 computing technology, distributed storage technology,
Relays IBDGs 2
Load 2 TS3
TS4 forming the data processing scheme based on big data
technology. The scheme can process the massive
Emergency
S3
feeder S9
heterogeneous and mult i-state data which generated during
S6
ADN operation effectively and proactively.
TS2 Load 3

Feeder 1 Feeder 2 Feeder 3


C. Data Generation and Simulation
1) Distributed Statistical Classifier Relays
Figure 2. Interconnection structure diagram of ADN’s autonomous region
Local Time series To/From
Other Relays
Each IBDG incorporates a three-phase voltage-sourced Feature
inverter (VSI), an LC filter, and a coupling inductor. The Extraction
control interface implemented in each IBDG is based on the Local Features
droop control technique proposed in [9]. The tie-line among
autonomous regions is emergency feeder. In the system during Differential
Communication
Features
normal operation, the relays (TS1~TS4) is turned off in order Statistical Channel
X
to the system open-loop operation. During the fault of the ADN, Classifier
either through the healthy part of distribution network working Remote
in the isolated mode, and can also through collaboration Features
among autonomous regions, to accomplish service restoration. Fault No-fault
Figure 4. General structure of a distributed statistical classifier relay
The relays (show in Fig. 1 and Fig. 2) are a new relay
protective device. For such a system, short circuit currents are Fig. 4 shows the general structure of a distributed
not high enough to operate traditional protective devices such statistical classifier relay. The system consists of four modules:
as fuses and overcurrent relays due to the interface control of local features are processed by a feature extraction module
the inverters which are equipped with current limiters. In fro m the voltage and current time series. Remote features,

978-1-5090-5167-0/16/$31.00 ©2016 IEEE


received via wireless communication channel module, local
and remote features are combined together by a differencing
operation. A statistical classifier will make a decision
(Fault/No-Fault) based on differential and local features [10].
At each of the relay points, electrical features such as
symmetrical components, total harmonic distortion, magnitude
and phase angles of current and voltage are among the system
parameters measured. Relays use both local and remote
electrical information to independently carry out two
fundamental tasks of protective relaying: fault discrimination
and fault localization. Since these tasks are implemented using
advanced data analysis algorithms, the relays are assumed to
be microprocessor-based throughout the paper [10].
Figure 5. Classification accuracy based on RF
In this paper, the statistical classifier is implemented using
the random forest (RF) [11] algorithm. RF are a comb ined
III. SELF-HEALING PROCESS
classifier based on statistical learning theory. Through
combining bootstrap re-sampling method and decision tree A. Self-healing Optimization Model of ADN with IBDGs
algorithm, its essence is to build a co llection tree
During the fault of the ADN , the first concern of the operators
classifier {hk (x), k  1, 2, ?··} , and then use the set by voting is how to fast and maximu m restoration of power supply on
for classification and prediction. Because the algorithm breaks condition that ensuring the priority of important load power
through the bottleneck that the single classifier cannot supply. As shown in Fig. 2, taking the restoration of the AND
improve in performance, it has good performance and can be feeders fault problems as an example , the optimization model
applied to various kinds of classification filtering and of the service restoration can be described as:
prediction. The final classifier’s decision (Fault/No-Fault) is
n m
generated by taking the majority vote over the decisions of max(  i xi Li    j L j ) 
each single tree in the forest. The RF has been shown to be i 1 j 1
more effective than individual decision trees with comparable
overall size, as it is less likely to overfit the data [11]. Here, Li is the power of load i in the feeder of
The training of the statistical classifiers follows a learning- autonomous region. xi is the power state of load i ( xi  0
fro m-data approach. The feature selection assists identify the means the power-losing o f load, xi  1 means the power-
most important features by ranking the information gain (IG)
of all input features. supplying of load); L j is the power of isolated island j when
2) Data Generation it runs in an islanded mode; n is the sum of load in the feeder
The dataset used in the analysis is generated by of autonomous region after service restoration; m is the sum
numerically simulating the ADN with IBDGs shown in Fig. 2 of isolated island after service restoration;  i ,  j are different
using Matlab and Simulink. Simu lation is necessary since weight coefficient of load which represents premier-level of
faulty events are harmful to the distribution system and they load.
cannot be safely reproduced in a real system for the learn ing
purposes [10]. This paper acquires 500 pieces of fault data The constraint conditions of ADN service restoration
records and 500 pieces of normal data records through optimization model:
simulation. We select 250 pieces of fault data and 250 pieces (1’ ) Line capacity constraints:
of normal data at random to form an original train dataset.
Then, spot-check 500 sa mples fro m it by bootstrap; Next ,  Pi  Ck,max  
build 500 decision trees models singly for those 500 samples i Ak
and achieve 500 kinds of classification results .Finally , Here, Ck,max is the maximu m transmission capacity of line
according to the results, take a vote of every record to
determine the final classification (Fault/No-Fault). k in AND; Ak is the set of the node which gains the electric
energy from line k ; Pi is the power that transmitted from
3) Simulation
Fig. 5 shows the classification results . In practice, the line k to node i .
classification error should decrease with the increase of the
(2’ )The power-supplying constraints in isolated island zone:
number of decision trees until the classifier performance levels
off. As it can be seen fro m the Fig. 5, RF can build efficient L  CL,max 
relays with the accuracy rate of ~99.96%. The results show
that the relays can reach the accuracy of ~99.96% in detecting Here, L is the total load capacity in isolated island;
faults on the generated dataset. CL,max is the maximu m capacity which the BDGS can
provide.

978-1-5090-5167-0/16/$31.00 ©2016 IEEE


(3’ ) Active power balance constraints: After the implement of fault localization and isolation, the
original region that no-fault and is connected with the bus
n n
continues supplying the normal power, then the fault section
 y  i  Li  determines service recovery schemes. The determination of
P  D 
yBa i 1 i 1
service restoration schemes (Due to limited space, the paper
Here, Ba is the set of emergency feeders which supply taking the fault in Fig. 2 as an examp le to detail explain.)
includes the following steps :
power to autonomous region a where faults happened; Py is
the power that can be provided from emergency feeders y to Step 1: Relay S4 sends a request to the relay S5 which
belongs to the CU to obtain the parameters e.g. load capacity,
autonomous region a  Di is the capacity of the load i on source capacity and the transmission state of near autonomous
feeders which is provided by IBDGs . region (can provide by TS3) of CU.
(4’ ) No electromagnetic coupled power loop constraints : Step 2: Relay S5 inquire these parameters of the CU, then
response relay S4 with the information that inquired.
  yk  1  Step 3: Relay S4 collects all the information, then
kRi synthesis the constraint condition of optimizat ion model to
analysis the conditions of load supply power recovery and
Here, Ri is the set of emergency feeders which to supply then executes service restoration operation according service
power for load i  yk is the state variable of emergency restoration schemes.
feeders k  yk value is 0 or 1. IV. SIMULAT ION AND RESULT ANALYSIS
To test and verify the feasibility of service restoration
B. Self-healing Scheme design of ADN with IBDGs schemes, according to interconnection structure of ADN’s
In order to restore the electricity supply of the loads in autonomous region show in Fig. 2, build a simu lation model in
fault section quickly and effectively, the electrical equipment, Matlab/Simu link. This paper takes three-phase grounding
load and IBDGs near the fault section can be considered as a fault of feeder 2 as an example to proceed the simulation of
control unit (CU). Considering the constraint conditions, the ADN service restoration, and then cut out the current
load in the CU has three service restoration schemes: waveform of relay S5 and S6 and load 1,2 and 3 separately.
(1’ ) No emergency feeders for CU: If the load capacity in During the fault isolation period, relay S5, S6 are cut off
the CU is less than or equal to the IBDGs capacity, IBDGs can instantly in 0.5s. To avoid reclosing, this paper testes the
supply power to the load in the CU alone. Meanwhile, IBDGs current waveform of relay shown in Fig. 6 which shows the
will run in an islanded mode as a standalone system. If the current of relay S5 and S6 is zero after 0.5s.
load capacity is greater than IBDGs capacity, consider
uninstalling partial load. 80

60

(2’ ) CU containing emergency feeders: If the load capacity 40

in the CU is greater than IBDGs capacity, through rational 20

coordination mechanisms to implement IBDGs and 0

emergency feeders jointly restore electricity supply for the -20

load in the CU. -40

(3’ ) No IBDGs for CU: If there are emergency feeders, the -60

feeders supply power to the load in the CU alone; if not, and -80
0.45 0.5 0.55 0.6

then combine the CU with nearest one. The load recovery


power supply of the recombinant CU can refer the options (1’ ), (a) Relay S5
(2’ ). 10

0
C. Self-healing Process of ADN with IBDGs
-10
1) Fault discrimination, Localization and Isolation -20
In this paper, relays can independently carry out two -30
fundamental tasks of protective relaying: fault discrimination
-40
and fault localization. The relays should be capable of
distinguishing between fault and no-fault conditions and of -50

recognizing when it is their zone of interest (the line to they -60

are attached) that is affected by a fault [11]. The relay is -70


0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
trained to trigger only when a fault strikes the line that the
relay is attached to. For examp le, if a fault occurs at point A in (b) Relay S6
Fig. 2, relays S4 and S5 will open to isolate the fault.
Figure 6. Current waveform of the relays
2) Determination of Service Restoration Schemes

978-1-5090-5167-0/16/$31.00 ©2016 IEEE


Fig. 7 shows the single-phase current waveform of load 1, are capable of implementing advanced signal processing
2 and 3(IBDGs can supply power to the load 1 alone; IBDGs methods and which benefit fro m co mmunication technologies
and emergency feeders jointly restore electricity supply for the placed on top of the physical electric network. Achieve
load 2 through rational coordination mechanisms ; the feeder 1 storage and processing of the massive heterogeneous and
can supply power to the load 3 alone by emergency feeders). mu lti-state data by using big data technology in this paper. An
As shown in Fig.7, the current is zero in t ime period 0.5-0.55s ADN model is built in Matlab and afterwards a self-healing
and return to the pre-fault after 0.55s. simulation in Matlab is executed. Results of simulation verify
1 the feasibility of service restoration schemes. The proposed
service restoration schemes can reduce outage time effectively
0.5
and improve the efficiency of maintenance technicians greatly.
0
A CKNOWLEDGMENT
-0.5
The research is supported by National Natural Science
-1
Foundation of China (NSFC) (No.61170191,61379063).The
authors would like to show great thanks for the Collaborative
-1.5
0.45 0.5 0.55 0.6 0.65 0.7 Innovation Center of wind power equipment and energy
conversion.
(a) Load 1
150 REFERENCES
100 [1] Y. You, D. Liu, and Q. Zhong, “Multi-time scale coordinated control of
distributed generators based on active distribution network,” EI.
50
Automation of Electric Power Systems, vol. 38, pp. 192-198, May.
0 2014.
[2] D. D. Liu, B. Duan, J. Wang, “ A Multi-agent System for Self-healing
-50
of Distribution Systems in Active Distribution Network Based on IEC
-100 61850,” EI. Automation of Electric Power Systems, vol. 39, May. 2015.
[3] VF. Martins, CLT. Borges, “Active distribution network integrated
-150
0.45 0.5 0.55 0.6 0.65 0.7 planning incorporating distributed generation and load response
uncertainties,” IEEE Transactions on Power Systems, vol. 26, pp. 2164-
(b) Load 2 2172, 2011.
[4] N. Higgins, V. Vyatkin, N. K. C. Nair, and K. Schwarz, “Distributed Po
wer System Automation With IEC 61850, IEC 61499, and Intelligent C
ontrol,”Systems, Man, and Cybernetics, Part C: Applications and Revie
ws, IEEE T ransactions on, vol. 41, pp. 81-92, 2011.
[5] I. J. Balaguer, Q. Lei, S. Yang, “ Control for grid-connected and
intentional islanding operations of distributed power generation,” IEEE
T ransactions on Industrial Electronics, vol. 58, pp. 147-157, 2011.
[6] J. Nahman, G. St’rbac, “A new algorithm for service restoration in
large-scale urban distribution systems,” EI. Electric power systems
research, vol. 29. pp. 181-192.
[7] J. J. Xun, S. X. Huang, “Ant colony algorithm based on immune
(c) Load 3 principle for the fault restoration of distribution network,” EI. Power
Systems and control, vol. 36. pp. 192-198, Sep. 2008.
[8] H. Al-Nasseri, M. Redfern, and F. Li, “A voltage based protection for
Figure 7. Current waveform of loads
micro-grids containing power electronic converters,” in Proc. IEEE
Power Eng. Soc. Gen. Meet. 2006., p. 7.
The controlling flow of simulat ion: t = 0 ~ 0.5 s, load 1: [9] N. Pogaku, M. Prodanovic, and T. Green, “Modeling, analysis and
maximu m power 10 kW; load 2: maximu m power 1 MW; testing of autonomous operation of an inverter-based microgrid,” IEEE
Load 3: maximu m power: 100 kW, rated frequency: 50 Hz. t = T rans. Power Electron., vol. 22, pp. 613–625, 2007.
0.5 ~ 0.55 s, the system executes the schemes of service [10] E. Casagrande, W. L. Woon, H. H. Zeineldin, “A Differential Sequence
Component Protection Scheme for Microgrids With Inverter-Based
restoration. Load power is restored after 0.55s. Distributed Generators,” IEEE Transactions on smart grid, vol. 5, Jan
2014.
V. CONCLUSIONS [11] L. Breiman, “ Random forests,” Mach. Learn., vol. 45, pp. 5–32, 2001.
This paper was motivated by the problem of self-healing
in ADN with IBDGs. Modern microprocessor relays, which

978-1-5090-5167-0/16/$31.00 ©2016 IEEE

You might also like