Real-Time Multi Agent System Modeling for Fault Detection in Power Distribution Systems Jawad Ghorbani, Muhammad A. Choudhry, Ali Feliachi Advanced Power & Electricity Research Center (APERC) Lane Department of Computer Science and Electrical Engineering West Virginia University, Morgantown, WV 26506-6109 Abstract—This paper presents a decentralized multi agent system (MAS) which works in real time with a power distribution system for fault detection applications. The agents use local voltage and current rms values. The multi-agent models are simulated in Matlab® Simulink using user defined sfunctions and the power system is modeled using the Simulink Simpower toolbox. The proposed method has been tested on a model of an existing Mon Power circuit. Both faulted zone and fault type have been successfully identified. Index terms-Real time multi agent system, power distribution system, fault detection and isolation.
I.
INTRODUCTION
Power distribution systems are facing major automation changes in modern world due to rising interest in renewables, distributed generators and storage devices and smart switches. Number of outages, voltage and frequency violations, and other power quality disturbances has increased due to increase in system complexity. Power interruptions and power quality problems are very costly for utility companies. For example authors in [1] estimated that the annual national cost of power interruptions is approximately 80 billion dollars. One of the long-term objectives for power distribution networks is to improve the quality of the electricity and its continuity of supply in a cost effective way. Uninterrupted power is the primary element of the quality of power distribution. The reliability of the supply can be improved in a fast and cost-efficient manner if we could do the fault location, isolation, reconfiguration and restoration operations faster. This results in improving the standard reliability indices such as SAIDI and SAIFI. Therefore, it is essential to locate and isolate the faults and then restore power as fast as possible by reconfiguring the electrical circuit. This paper introduces a multi agent approach for fault location and isolation which improves the reliability since it helps to locate and isolate the fault in less time.
The first author is a PhD student in Advanced Power & Electricity Research Center, West Virginia University, Morgantown, WV, USA. (E-mail:
[email protected]). This work was supported in part by the U.S. Department of Energy (Allegheny Power: West Virginia Super Circuit)
978-1-4673-2308-6/12/$31.00 ©2012 IEEE
A fault in the distribution network often interrupts the electric power to all the consumers along the feeder. One of the most common protection methods used for radial distribution systems is having a recloser at the substation which will isolate the feeder in case of short circuit faults. Whenever a fault occurs the recloser opens and closes three times to check whether the fault still exists in the feeder or not. If the fault does not clear after these attempts, it locks out. An agent is defined as “an autonomous computational entity such as a software program that can be viewed as perceiving its environment through sensors and acting upon this environment through its effectors” [2]. A multi-agent system is a group of agents, which sense the environment and acts in order to achieve its objectives. Due to the increasing speed and decreasing cost in communication and computation of complex matrices, multi-agent system promise to be a viable solution for today’s intrinsic network problems. The West Virginia Super Circuit (WVSC) project is part of the Modern Grid Initiative aiming at demonstrating and testing new technologies to enable the deployment and the implementation of Smart Grids. This work is part of the WVSC project which consists of designing MAS for fault detection and isolation applications. There are three types of agents in each feeder which constitutes the proposed MAS. The bottom-up order of feeder agents is: switch agents, zone agents and recloser agent. Switch agents are the lowest level of agents which have access to voltage and current rms values and can communicate with their higher level agents called zone agents. Switch agents calculate the impedance at their measurement points by using rms value for voltage and current which help them in their decision making process. Each zone agent covers some switch agents and all zone agents are at a lower level than the feeder recloser agent. The agents are intelligent and autonomous and in the case of recloser lock out they have the ability to make decisions and isolate the faults to prepare the condition for reconfiguration and restoration. Agents communicate and consult with their higher level agents and determine the faulted zone. In the case of loss of communications, agents
should be able to decide independently and by using available communication links. A multi agent system is one of the popular approaches for decentralized management of power systems and is applied to different areas like fault diagnosis, voltage stability, electricity market pricing, protection coordination, power system reconfiguration and restoration. Most approaches for fault location in the literature consist of a master agent which makes the final decision based on the data received from other agents [3]. In some other works [4] each agent is considered to have its special functionalities e.g. acquiring data, analyzing data, managing situations, etc. In other words each step of decision making is done by a special agent. In [5] the algorithm constantly monitors the network and from the available observations identifies the fault location. In [6] sequence current magnitudes and current direction during a fault were used for fault detection. The proposed multi agent system is different in the way that each agent has a decision making capability. The method uses voltage and current rms values and the calculated impedance to monitor the network and identify the faulty zone and the fault type. The advantage of this method is that each agent can work independently in case of loss of connection and it needs less communication capacity because agents just talk to their upper level adjacent agent. There are two main approaches for modeling MAS in power system. MAS and power system model could work together in a real time mode or offline. An interface plays a key role in connecting the two softwares. Developing interfaces between power simulation models and software agents is a key part in the works described previously. In [7, 8] authors describe the Electric Power and Communication Synchronizing Simulator (EPOCHS); EPOCHS is a simulation system, which links the power simulators with an event-driven communication network simulator. Another implementation to integrate software agents with simulation model of electrical system is presented in [9] and Common Object Request Broker Architecture (CORBA) is used to communicate MAS with an EPDS model in the Virtual Test Bed (VTB) simulation environment.
This paper is organized as follows. Section II presents the West Virginia Super circuit (WVSC) and the problem statement. Section III introduces multi agent systems. Section IV details the proposed approach and how we implemented our model for the WVSC. Section V presents simulation results. Finally, section VI concludes the paper and points out future research directions. II.
WEST VIRGINIA SUPER CIRCUIT AND PROBLEM STATEMENT
The distribution systems are easily exposed to faults, most of which are temporary faults. Fault location and isolation plays an important role in power systems security and reliability. Traditional approaches are mostly centralized and all the information is sent to a data center to be processed and actions are taken. At the same time, sending such a large amount of data demands high communication capabilities [10]. On the contrary, Multi-Agent System (MAS) can realize distributed control through cooperation between agents with less required communication capacity. Focus of this paper is on application of advanced technologies like MAS in fault detection and isolation processes. The West Virginia Super-Circuit (WVSC) project will demonstrate improved performance, reliability, and security of electric supply through the integration of distributed resources and advanced technologies such as the Multi Agent System [11]. The West Virginia super-Circuit is divided into different zones connected through controllable switches as shown in Fig.1.
Softwares used for modeling the MAS and power system can work together in two ways: a) Real time In this case the power system model and the MAS should be able to communicate in real time and the data is accessible to agents without any delay in simulation time. Since most previous works [6-9] use two different softwares for modeling MAS (JADE, EPOCHS…) and power system (OpenDss, EPDS, …), there are some problems with interfacing the two softwares. In this work both MAS and power distribution system are simulated in MATLAB and there is no need for interfacing the two softwares. b) Offline Some authors use offline data for power system simulation as inputs for MAS. In this case the model cannot consider communications in exact time like the real world.
Figure 1.
West Virginia Super Circuit
The West Run Substation has five feeders, two of which (WR-3 and WR-4) are monitored for faults. Once a fault is detected, the MAS will locate and isolate the faulted zone and restore power to the unaffected zones. In this work, the fault location and isolation part of the project are investigated.
Fault location isolation algorithm is applied to the two feeders (WR-3 and WR-4) of WVSC. West Run substation feeders are equipped with reclosers. When there is a fault in the system, the recloser will go through its trip and close operations as configured. Fault location and isolation system will start its fault location, isolation processes following a recloser lock-out signal sent to MAS. Fault location isolation system will not take any action when recloser trips on a temporary fault and remains closed after a close operation. Time Current Characteristic (TCC) information at two current levels for WR-3 and WR-4 reclosers is given in Table 1. These reclosers are programmed to operate three times before locking out. The first reclosing operates on A-curve and the following two reclosings operate on D-curve. The approximate clearing times for A-curve and D-curve are also listed. The reclosers wait for 60 cycles between the each reclosing operation. TABLE I.
RECLOSER SETTINGS FOR WR-3 AND WR-4
TCC
A curve
D curve
Current
2000A
6000A
2000A
6000A
Clearing Time
5 cycles
3 cycles
43 cycles
8 cycles
III.
B. Multi agent system structure The proposed multi agent system structure is shown in Fig.3. According to this architecture, there are three types of agents in each feeder. In the bottom-up order these agents are: a. Switch agent These agents have direct access to measured voltage and current rms values with the resolution of 16 samples per cycle and also these agents calculate the equivalent impedance. b. Zone agent The zonal agents are higher in level than switch agents and cover switch agents. Zone agents gather the switch agents’ data and at the same time communicate with the other zonal agents. Each zone agent can determine whether the fault is in its zone or not even in the case of loss of communication and decide to isolate the fault. Fig.2 shows the Zones for the WR-3 and WR-4 feeders of WVSC. WR-3 zones are with blue color and WR-3 with green. c. Re-closer agent Recloser agent is the master agent of each feeder that keeps track of the status of the recloser at the substation and informs the other agents in various zones when the recloser locks out. The data gathered during the three last closings of the recloser are the latest available data after recloser locks out. The agents should be aware of the recloser operating time to save the data with highest possible resolution for fault detection process.
MULTI AGENT SYSTEM
A. Basics of multi agent system Historically, multi-agent systems technology is a sub-field of distributed artificial intelligence, which itself is a sub-area of artificial intelligence. Nowadays, the term ‘multi-agent systems’ mostly is used to refer to all types of systems composed of multiple autonomous components [12]. In multi-agent systems, individual problem-solving entities are called agents; agents are grouped together to form communities which cooperate to achieve the goals of individuals and the whole system. It is assumed that all agents are capable of a range of useful problem-solving activities in their domain. As distributed systems, multi-agent architectures have the capacity to offer several desirable properties over centralized systems. MAS technology can allow us to distribute and localize the control of power systems. By incorporating intelligence in all agents the reliability of the system will improve dramatically since there is a no single point of failure as compared to the centralized control. There are lots of designs for agent’s environments and there is no single design which provides an infrastructure for specifying communication and interaction protocols [10]. One of the most common used protocols is the Foundation for Intelligent Physical Agents (FIPA) which is related with agent technology and MAS are standardized [13].
Figure 2. Zones in WVSC WR-3 and WR-4 feeders
using user defined S-functions. In the following both parts are discussed in detail.
A block is designed to generate reclosers operation. Whenever recloser senses overcurrent, it locks out after three operations according to the time periods mentioned in Table.1, if the fault is not cleared. Lines are modeled based on the positive, negative and zero sequence impedance values. Different types of faults such as single line to ground, line to line, and three phase faults are modeled with a fault block in Simpower Toolbox and ground resistance is considered to be 0.1ohm. Feeder loads are modeled with active and reactive power. Westrun #1 a b c
A B C
RC
Bb
SW14
CB14
RMS discrete
SW 1-I
RMS discrete
Aa
Bb
Cc
CB1
a A
b B
c C
A B C
SW1
A B C
A B C CB15 SW15
A B C
a A
A B C
Aa
b B
Bb
c C
Cc
A B C
A B C
RMS discrete
CB16
SW16
a A
Aa
b B
Bb
c C
Cc
A B C A B C
A B C
Cc
A B C
30 MVA 138 kV2
A B C
A B C
Bb
c C
+ v -
RMS discrete
A B C
CB11
Aa
b B
Display1
A B C
CB2
SW2
a A
Aa
b B
Bb
c C
Cc
A B C
A B C
SW3
CB3
Aa
a A
Bb
b B
Cc
c C
+ RMS v discrete
Display2
A B C
A B C
A B C
A B C CB4
Aa
a A
Bb
b B
Cc
c C
Figure 4. Finite state machine for MAS after fault occurance
In the case of normal operation, the difference between the input and output current of each zone is related the power usage of the zone. The switch agents at beginning and end of each zone monitor the current changes and in the case of abnormal changes report the fault. In addition to the current rms values the impedance changes are also considered as a criterion for fault detection. In the simulation results section some current and impedance changes due to some fault scenarios are presented. IV.
SIMULATION MODEL
The simulation model has two parts. The first part is for simulating power system distribution network which is the WVSC by MATLAB Simpower Toolbox. The second part is the multi agent system which is implemented in Simulink by
a A
Aa
b B
Bb
c C
Cc
A B C
A B C
A B C
A B C
Aa
Bb
Cc
b B
c C
CB8
A B C
a A
A B C
b B c C
CB7 a A
SW7 Aa
b B
Bb
c C
Cc
SW8 Aa Bb Cc
CB5
A B C
A B C
A B C
A B C
+ v -
A B C
Aa
Bb
Cc
CB6
A B C A B C
A B C
A B C A B C
SW6
SW5
RMS discrete
A B C
A B C
A B C
SW9
A B C
CB9
A B C
A B C
A B C
a A
A B C
A B C
c C
Display3
b B
A B C
RMS discrete
a A
+ v -
c C
b B
CB10
a A
A B C
Cc
Bb
SW10
Aa
A B C
SW4
A B C
A B C
Fault Breaker
A B C
A B C
A B C
A B C
A A B B C
A B C
A B C
A B C
A B C
a b c
Bb
c C
SW11
a A
A B C A B C
Aa
Cc
b B
A B C
a A
Cc
A B C
Bb
c C
SW12
A B C
Aa
b B
Pierpoint #2
A B C
A B C
A B C
A B C
A B C
M AS
CB12
SW13
a A
RMS discrete
SW3-I
#3
#5
CB13
SW 3-V
SW18
A B C
Aa
Cc
A B C
c C
A B C
#7
RCW1
#8
Re-closer Control
SW 1-V
SW17
A B C
a b c
#4
acom A b B c C
A B C
Bb
A B C
30 MVA 138 kV RC Pulse
Aa
Discrete, Ts = 0.001 s powergui
A B C
30 MVA 138 kV1
Cc
Westrun #2
b B
C. Agents communications Fig.4 shows the Finite State Machine (FSM) for fault detection isolation and finally restoration processes. According to this FSM after the recloser locks out, agents start their negotiations to locate and isolate the fault. Agents can just communicate with their lower level agents and in normal situations lower level agents should not take any action before consulting with their higher level agents. But in the case of loss of communications with higher level agents, the lower level agents can make their decisions to isolate the fault and restore the power.
a A
Figure 3. Multi agent system architecture
A. Power System Model Two feeders of West Virginia Super Circuit (WR-3 and WR-4) which are shown in Fig.1 are modeled in MATLAB using Simpower Toolbox. There are 16 switches and 2 reclosers installed in these two feeders to enable the system. Switches are Cooper DAS-15 type three-phase vacuum switches with 15kV, 630 A ratings. 7 switches are normally closed and the other 9 switches will be used for reconfiguration and restoration applications. Fig. 5 shows the Simulink model for WR-3 and WR-4 feeders. Transformers in each feeder are 138/12.5 KV and 33.6 MVA. Simulation results are calculated based on per unit values. The base MVA, voltage values are 6 MVA, 8.5 Kv.
Display4
Figure 5. Power distribution system Simulink model
The power system model could simulate in both discrete or continues models. Since continues model is more accurate the power system is modeled using the continuous mode. B. Multi agent system Model The multi agent system is implemented using the Sfunction blocks of MATLAB Simulink. Each agent is modeled by an S-function. S-functions (system-functions) provide a powerful mechanism for extending the capabilities of the Simulink® environment. An S-function is a computer language description of a Simulink block written in MATLAB®. S-functions use a special calling syntax called the S-function API that enables you to interact with the Simulink engine. They follow a general form and can accommodate continuous, discrete, and hybrid systems [14]. Fig.6 shows the MAS model.
Since the MAS is mainly including agents communication, and communications are intrinsically discrete in time, MAS works in discrete mode and it also provide the base to model communication delays and latency. VS1 RCW1-V
ZM at
T o Workspace1
RCW1 T o Workspace
RCW1-I
IS1
2 S-Function5
T o Workspace2
RCW1
Zs1 Scope2 T o Workspace3
SW1-V
SA1
ZA1
ZA1 Goto
SW1-I
S-Function1
U
S-Function4
Y
Selector2
SW1
U
Y
Selector1 U
SW2-V
SA2
Y
Sel ector
Scope1
SW2-I
S-Function2 SW2
ZA3 SW3-V
Di splay3
SA3
S-Functi on9
SW3-I
S-Function6 SW3
SW4-V
SA4
ZA2
S-Function3
S-Function7
SW4-I
Di splay2
SW4
Figure 7. Recloser operation
Scope SW5-V
SA5 SW5-I
S-Function8 SW5
2
4
Figure 6. Multi agent system model in MATLAB Simulink
V.
SIMULATION RESULTS
3
2.5 Current
Since the measured data resolution is 16 samples per cycle, the agents can access to power system model simulation data with 16 samples per cycle resolution, the MAS discrete step simulation time should be 0.001 second. The switch agents have access to voltage and current measurement data and they calculate the impedance value based on the voltage and current rms value. The data packets that switch agents send to their corresponding zone agents have 9 elements. These elements are voltage, current and impedance of each of three phases.
3.5
1.5
1
0.5
0
2
4
6 Time
8
10
12 x 10
4
12
10
A. Case 1: Single phase to ground fault between switch 2 and 3
As Fig.8 and Fig.9 show, after fault occurrences and during the recloser operations the per unit values of current sensed at switch 1 is increased and impedance is decreased. The current value before the fault is 0.7 per unit and after the fault occurrences; it increases to 3.25 per unit.
0
Figure 8. Current changes for switch 1 during and before fault (Perunit)
In order to test the operation of the MAS and power system and to evaluate its performance, different fault scenarios were simulated, two of the simulated cases are presented here:
8 Im pedance
At time 20 second a single phase to ground fault occurs between switch 2 and 3 according to Fig.5. The recloser detects the over current and locks out. During the three recloser operations the data are communicated among the agents and the agents detect the fault type and zone based the changes in current and impedance values change. Fig.7 shows the recloser operation after fault occurrences. Fig.8 and Fig.9 show the current and impedance change for each phase from recloser point of view. The blue phase is the faulty phase and the other two phases are in blue and green colors.
2
6
4
2
0
0
2
4
6 Time
8
10
12 x 10
4
Figure 9. Impedance changes for switch 1 during and before fault (Perunit)
In this case, impedance values for the switches after the switch 3 in feeder WR-3 does not change after fault occurrences and by this way agents know that the faults is not below the switch 3. The difference between the input and output current of zone 1 where the fault occurs is different and it’s another criterion which agents use to locate the fault.
B. Case 2: phase to phase fault between switch 2 and 3 A phase A to phase B fault occurs at the time 20 second and Fig.10 and Fig.11 show the current and impedance changes before and after the fault and during the recloser operations. In this case both phase A and B current increased and the impedance decreased.
Our simulation results reveal that different faults can be distinguished and located by using the measured current and impedance values. The proposed method has been tested on a model of an existing Mon Power circuit. Both faulted zone and fault type have been successfully identified. In future works this simulation model will be used to support our research in using multi-agents for restoration and reconfiguration processes after fault location and isolation.
4.5
REFERENCES
4
[1]
3.5
Current
3
2.5
[2]
2
1.5
[3]
1
[4]
0.5
0
0
2
4
6
8
10
12 4
Time
x 10
[5] Figure 10. Current changes for switch 1 during and before fault (Perunit) [6]
2.5
2
Im pedance
[7] 1.5
[8]
1
0.5
0
0
2
4
6
8
10
Time
12
Figure 11. Impedance changes for switch 1 during and before fault (Perunit)
VI.
[9]
4
x 10
[10]
CONCLUSIONS AND FUTURE WORK
This paper presents a decentralized multi agent system (MAS) which works in real time with a power distribution system for fault detection applications. The agents use local voltage and current rms values. The multi-agent models are simulated in Matlab® Simulink using user defined s-functions and the power system is modeled using the Simulink Simpower toolbox. Both MAS and power system models are implemented in MATLAB Simulink, and therefore, there is no need for an interface between the two models. This advantage provides a simpler and more accurate simulation model for investigating MAS applications in power systems.
[11]
[12]
[13] [14]
Kristina Hamachi LaCommare, Joseph H. Eto, “Understanding the Cost of Power Interruptions to U.S. Electricity Consumers”, University of California Berkeley, Environmental Energy Technologies Division, September 2004. G. Weiss, “Multiagent Systems: A Modern Approach to Distributed Artificial Intelligence,” The MIT Press, 2000. L. H. Chiang, E. L. Russell, and R. D. Braatz, Fault Detection and Diagnosis in Industrial Systems, Springer, London, 2001. J. A. Crowder, “Multiple information agents for real-time ISHM: Architectures for real-time warfighter support”, International Conference on Artificial Intelligence, 2010. L. Liu, K. P. Logan, D. A. Cartes and S. K. Srivastava, “Fault Detection, Diagnostics and Prognostics: Software Agent Solutions,” IEEE Transactions on Vehicular Technology, Vol. 56, No 4, pp 1613 – 1622, July 2007. Seetaram Alwala, Ali Feliachi, M. A. Choudhry, “Multi Agent System based Fault Location and Isolation in a Smart Microgrid System”, Innovative Smart Grid Technologies (ISGT), IEEE PES 2012. J.S. Thorp; X.R. Wang; K.M. Hopkinson; D. Coury; R. Giovanini, “Agent Technology Applied to the Protection of Power Systems”, Autonomous Systems and Intelligent Agents in Power System Control and Operation. C. Rehtanz. S. Baden, Eds., Springer-Verlag. 2003. [4] K.M. Hopkinson; K.P. Birman; R. Giovanini; D.V. Coury; X. Wang; J.S. Thorp, “EPOCHS: Integrated Commercial Off-The-Shelf Software for Agent-based Electric Power and Communication Simulation”. Proc. Of the 2003 Winter Simulation Conference. 7-10 of December 2003, New Orleans, USA. G. Morejon, L. Sun, D. Cartes. CORBA Implementation for Interfacing Software Agents with Virtual Test Bed. ASNE Reconfiguration and Survivability Symposium 2005. Atlantic Beach, Florida. February, 2005. Janeth G. Gómez-Gualdrón, Miguel Vélez-Reyes, “Simulating a MultiAgent based Self-Reconfigurable Electric Power Distribution System”, 2006 IEEE COMPEL Workshop, Rensselaer Polytechnic Institute, Troy, NY, USA, July 16-19, 2006. Belkacemi, R.; Feliachi, A.; Choudhry, M.A.; Saymansky, J.E., “MultiAgent systems hardware development anddeployment for smart grid co ntrol applications”, Power and Energy Society General Meeting, 2011. Jusong Yang; Montakhab, M.; Davies, T.S.; Pipe, A.G.; Carse, B., “Fault diagnosis of power distribution systems using a multi-agent approach ”, Universities Power Engineering Conference. UPEC 2004. 39th International, 2004. The Foundation for Intelligent Physical Agents. 2005. http://www.fipa.org. http://www.mathworks.com/
