A Case Study of Multi-Agent Interoperability in

IEC 61850 Environments

Arshad Saleem Nicholas Honeth Lars Nordström
Technical University of Denmark KTH Royal Institute of Technology KTH Royal Institute of Technology
{[email protected]} {[email protected]} {[email protected]}

Abstract—The IEC 61850 is the most promising stan- implemented decision making logic they determine optimal
dard for design of substation communication and automa- operational conditions for the power system given boundary
tion systems. On the other hand multi-agents systems conditions.
are attracting growing interest for different applications
of substation automation systems. In multiagent systems Interoperability is of course a necessary pre-requisite
agents represent different stake holders in the power sys- for such architectures. Here we identify two aspects
tem and based on implemented decision making logic they of interoperability; horizontal and vertical. Horizontal
determine optimal operational conditions for the power interoperability is relies on a common semantic models of the
system’s given boundary conditions. Interoperability is of power system that the agents can use to form decisions. One
course a necessary pre-requisite for such architectures. such semantic model is presented in the IEC 61970 Common
Here we identify two aspects of interoperability; horizontal Information Model (CIM). The CIM has the constructs
and vertical. Horizontal interoperability is relies on com- necessary to represent knowledge about the complete power
mon semantic models of the power system that the agents system that the agents need for optimized decision making.
can use to make decisions. One such semantic model is Vertical interoperability is concerned with making the agent-
presented in the IEC 61970 Common Information Model based architecture interact with contemporary automation,
(CIM). At this level, the IEC 61850 standard provides protection and control systems in substations and power
a model for access to information and control functions plants. At this level, the IEC 61850 standard provides a
that has the necessary flexibility needed. In this paper model for access to information and control functions that
we discuss the mapping between a multi-agent based has the necessary flexibility needed. In this paper we discuss
architecture for power system control and the IEC 61850 the mapping between a multi-agent systems MAS based
standard for utility automation. The mapping is based architecture for power system control and the IEC 61850
on a use-case drive approach, in which the information standard for utility automation. The mapping is based on a
exchange need is defined by the multi-agent system. use-case drive approach, in which the information exchange
Index Terms— multi-agent systems, power systems con- need is defined by the multi-agent system. Based on this
trol and protection, industrial application multi-agent need, interaction patterns with logical nodes (LN) as defined
systems, power systems control and protection, industrial in the IEC 61850 are identified. The mapping also enables
applicationa interaction at different levels of abstraction with the IEC
61850 based systems.
I. I NTRODUCTION
The development towards a sustainable energy system in Rest of this paper is organized as follows:
the electric power industry has lead to the emergence of a set Section II introduces fundamental concepts of agents, multi-
of market models and new concepts for optimized operation agent systems, IEC 61850 and related standards. It also intro-
and control of power systems, e.g. Virtual Power Plants and duces the problem of protection and control in electric power
Microgrid. In these new concepts, the traditional stake holders systems with distributed generation which provides the study
are complemented by new actors that take roles such as case used in this paper. Section III describes our approach for
aggregator, prosumer, dispatchable load etc. Common to all MAS to IEC 61850 mapping. Section IV presents application
these concepts is that they assume a more flexible and loosely of our approach in a study case and discusses the results.
coupled ICT system architecture. In such architectures, ICT Section V concludes the paper.
components communicate to implement optimization, control
and protection functions. One approach to such architectures II. R ELATED W ORK
is the use of multi-agent system. The agents represent This paper combines the formalisms of agent-based control
different stake holders in the power system and based on with the nomenclature of IEC 61850. In this section, these
topics are dealt with separately after which other work that
combines them is discussed.

A. Multi-Agent Systems
The fundamental concept of software agent is defined as:
An agent is an encapsulated computer system that is
situated in some environment and can act flexibly and au-
tonomously in that environment to meet its design objec-
tives.[10]
Agents are elaborated by a metaphore commonly known as
BDI (Belief, Desire, Intention). Beliefs represent knowledge
of an agent about its environment. The Beliefs are captured
through sensors of the agent and stored in an internal data
base. This data base (also commonly called knowledge base)
should be properly organized, updated and synchronized to Fig. 1. Functional levels and logical interfaces defined in the IEC
other functions, e.g. decision making of the agent architecture. 61850 standard.
The Desires are goals or design objectives of an agent (or of
the systems agent is part of). Desires not only sets the criteria
for rationality of an agent but also defines the nature and level C. IEC 61850 and related standards
of autonomy for agents. Intentions is the way agents attempt The IEC 61850 series of standards for communication
to achieve their design goals. In agent oriented software networks and systems is intended to provide interoperability
engineering intentions are modeled as behaviors. A behavior between Substation Automation Systems (SAS) [1]. It pro-
of an agent may consists of a single or multiple actions and vides a specification for the communication between Intelli-
lead to a achievement of a goal or a sub goal. gent Electronic Devices (IEDs) and related SAS equipment.
Multi-agent systems (MAS) are systems consisting of more Furthermore, it describes the requirements of the functions
than than one agent. MAS are useful to implement in applica- implemented in SAS, not to attempt to standardize the func-
tion areas that are naturally distributed, decentralized and are tions themselves but to specify the communication between
easy to be decomposed in their design. A system architecture them.
based upon MAS provides a natural way of decomposing a A subset of the functions that are described in [1] fall
software system into subsystems and to model interactions into the category of distributed functions. The standard defines
between these subsystems and individual components (agents) these as the set of functions where it’s subparts, called Logical
within the subsystems. Nodes (LNs), are located on different physical devices. While
all functions communicate with each other, the process that
they are controlling, monitoring or protecting, distributed
B. Protection and Control in Electric Power Systems functions are dependent on the execution of a set of defined
with Distributed Generation functional steps for their functionality. The loss of any of
Introduction of distributed generation in medium and low the constituent LNs could mean that the function would be
voltage grids has brought challenges in functions of pro- blocked or that it would be functionally degraded.
tection and control. New approaches suggest increasing use The IEC 61850 standard also defines a functional hierarchy
of modularity and communication. The methods presented where functions are classified in terms of how closely they
in [6], [9], [5], [8], [4], [7] are among many works that are situated to the substation process. Three main levels are
propose the use of multi-agent systems for protection and defined:
control in electric power systems. Both [6] and [9] propose
methods that utilize agent-based zoning for use in distribution • bay level functions - refers to the group of functions that
networks. In these schemes, agent-controllers are placed at are predominantly associated with a specific bay in the
the zone borders and at DG sources withing the network. By substation instance.
interactively comparing measurement data and coordinating • process level functions - interface directly with the pro-

effector capabilities at different locations in a distribution cess, namely I/O functions such as data acquisition and
network, various protection, monitoring and control functions issuing of commands.
can be implemented. Such distributed functions can include • station level functions - refer to functionality that con-

fault location and restoration [9], [4], current differential cerns the substation as a whole.
protection [6], islanding [7], adaptive load shedding [2] and Figure 1 from [1] illustrates the functional hierarchy as well
voltage regulation [8]. as shows the numbering of the standard interfaces between
LNs in different levels and between LNs situated on the same with SCADA or other high-level control.
functional hierarchy level. Interfaces 2 and 10, shown in gray,
are not defined in the standard. Interface 2 is reserved for use The formalism proposed here is that a set of LNs
in remote protection functions on the same level on the control are implemented in each MAS agent, this allows MAS
plane while interface 10 is the undefined vertical communica- integration to remain consistent with the IEC 61850 standard.
tion to SCADA or other remote control. The specification of More specifically, LNs that are implemented in the MAS
communication via the remaining interfaces is the core of the (mostly at the station functional level) appear as standard
IEC 61850 standard. LNs to all other LNs and communicate via the same
Communication between LNs at the station and bay levels interfaces using the protocols specified by the IEC 61850
occurs through interfaces 1 and 6, while between LNs within standard. This allows complex distributed functions to utilize
a station or bay are 9 and 3 respectively. Interface 8 supports the cooperative, autonomous and pro-active capabilities of
direct communication between LNs in different bays, this is MAS-based control interoperably with IEC 61850-based SAS.
used to support functions such as interlocking.
Finally, interfaces 4 and 5 provide the communication
channel between process and bay level functions. The use of A. Roles and control plan
a process bus specified in IEC 61850 is discussed in detail in The control strategy presented in [9] specifies that an agent
[3]. The benefits that are pointed out are the increased level of can assume a set of different roles. These roles fill different
interoperability between low-level devices that is achievable as functionalities defined in a control plan. The role assigned to
well as the possibility for cost and operational optimizations an agent can change due to changes in the state of the system
that are not possible using more traditional methods where control plan such as disappearance of an agent, appearance
extensive copper wiring is required. of a new agent, changes in agent capability, external trigger
events or scheduled activities. The roles assumed by the agent
D. Multi-Agent Systems and IEC 61850 and LNs assigned to each role are defined in the control plan.
General objectives of this paper is similar to that of the
work done in [2] which proposes the view of IEC 61850 The processes of generating a control plan and allocating
and CIM to provide a standardized framework for application roles proposed in [9] makes use of a transition function.
of MAS to electrical power protection, control, monitoring The transition function maps a control plan, defined as a set
and recording. In [2] the author proposes a 1:1 mapping of of related roles, to specific world situations. The transition
agents to LNs in a SAS. Vertical interoperability is achieved function is based on domain principles such as the laws
by implementing LN functionality as an agent while the of electromagnetism and control theory. Transitions are
horizontal agent interoperability is by definition maintained. determined at design time. At run time, role assignment
Agents are categorized by the functional level at which they to agents based on the transition function are determined
are placed, these include the process, logical device, bay and by means of an auction mechanism. A coordinating agent
substation level. The type of inter-agent communication and mediates the communication required to perform the role
the interfaces used are defined by the functional level at which assignment. The assignment and realization of the roles to
the distributed function implemented by the MAS is situated. agents depends on the capabilities of the agent physical host
devices.
III. MAS TO IEC 61850 M APPING
When considering the method for the extension of [2]
in this paper, we use [9] as the basis for deployment of B. Logical Node Assignment
agents and the assignment of agent functionality. Figure 3 Depending on the required functionality of the agent, sets
illustrates the agent placement used in the example in [9] that of LNs that implement the functionality must be assigned to it
is described in section IV where a Distributed Generation accordingly. IEC 61850 LNs are selected such that they form
(DG) agent is placed at each DG source and a relay agent the low-level ”atomic” functional units of control, monitoring
is deployed at each zone border. The geographical locations and protection. Each agent must therefore implement IEC
of the the agents are also the location of the host physical 61850 LNs for all of the low level functionality required by
devices or servers as they are referred to in [2]. all roles that the agents are capable of assuming. Not all LNs
are used simultaneously, different subsets of an agent’s LN
The horizontal communication in the functional hierarchy set are used depending on the current role.
occurs through logical interfaces 2, 3, 8, 9 and 10. All except
interfaces 2 and 10 are specified by the IEC 61850 standard The LNs that the agent is capable of implementing will
for communication between LNs. Interface 2 is allocated depend on the capabilities of the agent host physical device
for implementation of remote protection functions. This as well as which LNs lower in the functional hierarchy it
bay-level horizontal communication should by definition be can communicate with. Figure 2 illustrates the relationship
both reliable and low latency. The remote control interface between the agent itself, the roles it can assume, the LN
10 for vertical communication is intended for communication set associated with a specific role and finally, how the
 Agent
Relay Agent A1

Station A1

Role
Role Transformer Agent 1

Zone 1 Relay Agent B1

Load Agent 2 Station B3 Station B2 Station B1

Load 2
LN LN
LN
LN

Load 1
Load Agent 1
Relay Agent B3
Relay Agent C1

Station C3 Station C2 Station C1

Zone 2
Capability Set

DG 2
G

Load 4
G

Load 3
DG 1

DG Agent 2
DG Agent 1
Load Agent 4 Load Agent 3
Fig. 2. Illustrates how an agent can assume various roles containing Relay Agent C3

a set of LNs each which use different capabilities. Relay Agent D1

Station D3 DG Agent 5 Station D2 Station D1

DG 5
Zone 3
G

DG 3
G

LNs are interact with the capabilities of the physical host G

Load 5
Load Agent 6
DG 4
device. Communication between LNs strictly follows IEC

Load 6
Load Agent 5 DG Agent 3 DG Agent 4
Relay Agent D3
61850 specification using for instance, logical interface 3
if the communication occurs between bay-level LNs in the
same bay or interface 8 if the communication is between bays. Fig. 3. Power system network used in example indicating assignment
of agents.
The scope of the functionality available to an agent LNs
is determined by the capabilities accessible to the agent.
Most of the capabilities are enabled through the process level brief explanation of agent decision models and description of
LNs like XCBR, TVTR and XSWI. These LNs do not need the agent types used in the system. The portection and control
to be implemented on the same device but need to be in system model is then mapped to the IEC 61850 formalism
communications with the agent physical host device using described in the preceding section.
the IEC 61850 specified process interfaces. Communication
between the process level LNs and bay level LNs occurs
A. Agent Decision Models
through the process bus labeled as IF4 and IF5 in the standard. This subsection describes the decision models of three types
Some bay level function LNs could be implemented by IED of agents.
bay controllers as per the current norm while more complex 1) Relay Agent: The relay agent has a central role
distributed functions become well suited to implementation in proposed schema. There is one Relay agent at the start
as agent LNs. and end of each zone in the network e.g. R2 and R3 for
zone II. They continuously monitor the state of the network,
This method is consistent with the IEC 61850 standard for identify and respond to any changes or transition triggers.
substation automation but allows complex functionality or Relay Agents work as zone disconnecters with responsibility
functionality where stakeholders should be represented to be to separate a zone from the network. In normal condition,
implemented on an agent platform that supports a high level there is a steady state current flowing into the network and
of local control intelligence incorporated into LNs as well whenever a fault occurs due to, e.g., a short circuit, a high
as the ability to negotiate and cooperate with other agents. current (fault current) flows into the network. The value
Some larger distributed function LNs could be implemented of this fault current is significantly higher than the normal
on a group of individual but related agents while similarly, current value in steady state. The relay agent gets triggered
a group of closely related LNs could be implemented on a upon observing an unusual high current value and has three
single agent platform logical device. main tasks:

i: Direction of fault current:

IV. E XAMPLE AND VALIDATION
In order to describe and validate the mapping formalism In a fault scenario, current always flows form the current
presented here, we apply it to the example agent-based pro- source to the fault location. Thus, to ensure that fault is
tection and control scheme presented in [9]. Beginning with a inside the primary zone of a relay, the relay agent has to
 TABLE I
D ESCRIPTION OF AGENTS , SATES , ROLES AND CAPABILITIES

Agents States Roles Capabilities Actions

stressed P++
average generator produce power P- -
DG relaxed regulator freq. control disconnect
disconnected reconnect
faulted primary
functioning facilitator monitor current close
Relay faulted-zone neutral monitor voltage reclose(open)
cleared-zone blocked

critical connected (re)connect

Load non-critical disconnected self-shed disconnect

make sure that the direction of the fault current is into previous section. Figure 4 describes the decision model of
its primary zone at two zone connecting breakers i.e. the Relay agent.
breakers which connect a zone to its neighboring zones,
and at the DG connection breakers for all DGs inside the zone.

ii: Magnitude of fault current:

The relay agent has to ensure that one of the fault current
measurements either from the zone connecting breakers or
from the DG connecting breakers is greater than a certain
threshold. This is necessary because in case some of the
loads in the zone are served from DGs outside the zone, the
current flows into the zone even in normal situation when
there is no fault.

iii: Role assignment:

After a fault has been confirmed inside one of the zones,

the job of each relay agent with the zone of its primary
responsibility isolated from the main grid is to calculate
energy balancing in its primary zone and assign new roles to
DGs and loads inside the zone. This requires calculation of
total generation and consumption of energy inside the zone
and negotiation with DG and Load agents for participation in
balancing. DG agents and Load agents calculate local cost
functions based upon their current state and capabilities and
communicate it with Relay agent. The relay agent based upon
the value of cost function of each of these agents assigns
them new roles. Thus, the job of Relay agent, in this case, is
to determine a mapping function that takes current state and
maps roles to specific agents based upon their capabilities, i.e.,

ftr (Scur , Ti , CPini ) =⇒ CPf n (1)

Fig. 4. Relay agent decision model.
Where ftr is the function that takes current state Scur and a
transition property Ti to map a chosen control plan CPini into 2) DG Agent: DG agents represent distributed power
a final control plan CPf n with all roles assigned to specific generators in electrical network. Every DG agent, on receiving
agents. T is the set of transition triggers described in the message from Relay agent, calculates its cost function. The
cost function of DG agent is based upon its current state cost function of load is given as:
e.g. relaxed/average/stressed, and its capabilities e.g. ability
to control frequency. The cost function of a DG agents is
defined as: δc (Scur , Ccur ) =⇒ Urole (3)
it is a function that maps current state Scur and current
capabilities Ccur of load agent into a role utility.
δc (Scur , Ccur ) =⇒ Urole (2) After calculation of the cost function, load agent sends a bid
based upon value of this cost function to Relay agent. Relay
i.e., the cost function is a function that maps current state
agent cumulates bids from all Load agents and sends back a
of DG agent Scur , and current capabilities Ccur into a role
message with a new role. Load agent on receiving this message
utility Urole . DG agent sends a bid based on the value of
takes up the new role. The decision model of load agents is
this cost function. Relay agent cumulates bids from all DG
same as that of DG agent with only difference of different
agents and sends back a message with a new role. DG agents
set of capabilities and current states. Different possible states,
upon receiving this message takes up the new role and start
roles, capabilities and actions for Relay, DG and Load agents
executing actions related to this role. A flow chart for decision
are described in Table I.
model of DG agents is given in figure 5.
B. MAS Protection and Control Mapped to IEC 61850
This section details the allocation of LNs in order to
model the protection scheme from section IV-A. We begin
by describing the LNs that are of interest after which the LN
assignment and interaction is described.
The deployment of the various agents is shown in figure 3.
For the Relay agents we assign the station-level LNs RFLO
and PDIF, Load agents are assigned PIOC and MMTR while
DG agents are assigned ZGEN and ARCO as shown in Figure
7. Relay agents are placed at the zone borders in order to
monitor and control up and downstream flow from the zones.
Figure 6 shows the IEC 61850 style model of the collaborat-
ing parts of substations B1 and C1. It presents a modified ver-
sion of figure 15 in the IEC 61850 standard which illustrates
the LNs that define a distributed busbar protection system for
a single substation. To illustrate the mapping onto the test
scenario presented in this paper, the example is expanded to
include two electrically connected substations which interact
with each other in order to implement a distributed fault
location LN RFLO.
In this case it makes sense to implement the lower level
(process and bay level) LNs in dedicated logical device
hardware such as IEDs and MUs . The closely related sta-
tion level PDIF and RFLO LNs are collocated on the relay
agent platform. The distributed function RFLO requires agent
capabilities such as communication, negotiation, data consis-
tency/quality management and intelligent pro-active control.
The interaction between agents uses agent communication
language (ACL) and utilizes the utility’s IP-based wide-area-
network for communication.
The process level LNs define the sensor and actuator equip-
ment at the process level, a current measurement transformer
Fig. 5. DG agent decision model. in the example. Process level LNs are likely to use merging
units (MU) as logical devices . Current samples are collected
3) Load Agent: Load agents represent electric power at process level and send via the process bus (most likely
loads in the network. Load agent, on receiving message using GOOSE or GSE messages) to bay controller logical
from Relay agent, calculates its cost function. The cost devices that are subscribers to the current sample data. Bay
function of load agent is based upon its current state e.g. level LNs could be assigned to dedicated fast-response logical
critical/non-critical and the capabilities e.g. auto-shed. The devices, they often include protection and safety functions
 Substation B1 Substation C1
LN for station
PBDF PBDF
level functions Current
Samples
ACL

Relay Agent
Trip Current
Samples
Trip
Relay Agent
RFLO RFLO
B1 C1

LN for bay level

functions Bay B1a
PBDF PBDF Bay B1b PBDF Bay C1a PBDF Bay C1b

Trip Trip

Current Current
LN for process Samples Samples

level functions
TCTR XCBR TCTR XCBR TCTR XCBR TCTR XCBR

Fig. 6. Decomposition of functions into interacting LNs on different levels showing agent LN interaction.

Relay Agent C3 Relay Agent C1

RFLO PBDF RFLO PBDF

Station C3
Station C2 Station C1
DG 2

G
Load 4

G
Load 3

DG 1

Load Agent 4 Load Agent 3 DG Agent 2 DG Agent 1

PIOC MMTR PIOC MMTR ZGEN ARCO ZGEN ARCO

Fig. 7. Showing the mapping of high-level LNs to agents in zone C from Figure 3.

which must be verifiable in terms of reliability and response By modeling the structure and communication of multi-
time. Station level LNs are more likely to require a high level agent functionality using IEC 61850 nomenclature there is
of interaction and therefore are in some cases best assigned the potential for seamless interoperability between modern
to agent platform based logical devices where distributed SAS best-practices and sophisticated distributed intelligent
functions can be implemented across a set of agents. control.

V. C ONCLUSIONS
Applying the IEC 61850 functional hierarchy allows SAS
In this paper we have shown the applicability of multi- design engineers the flexibility to make optimal choices in
agent systems for control and protection in electric power terms of the allocation of dedicated hardware for predictable
systems can be augmented by the integration of IEC 61850 response times or integration of functionality in general
communication principles. hardware to save costs and allow for integration for high-level
distributed control.
 Traditionally, the protection systems in electric power
industry have utilized very little communication. With the
adoption and integration of IEC 61850-enabled devices and
the development of powerful, reliable distributed intelligent
control methods that inter-operate transparently with these
devices, the goal would be to realize a robust, scalable,
secure and interoperable future electric power transmission and
distribution system that adheres to well-developed and intuitive
standards and best practices. We have therefore provided a
mechanism which is robust to communication failure. In the
worst case of total communication failure the result will be as
good as that of from current common practice.

R EFERENCES
[1] International Electrotechnical Commission, Geneva, Interna-
tional Standard IEC 61850-1.10, 2003.
[2] A. Apostolov. Multi-agent systems and iec 61850. In Multi-
agent systems and IEC 61850, 2006.
[3] A.P. Apostolov and B. Vandiver. Iec 61850 process bus -
principles, applications and benefits. In IEC 61850 process bus
- Principles, applications and benefits, 2010.
[4] Ioannis S. Baxevanos and Dimitris P. Labridis. Implementing
multiagent systems technology for power distribution network
control and protection management. Power Delivery, IEEE
Transactions on, 22(1):433–443, Jan. 2007.
[5] K. Heussen, A. Saleem, and M. Lind. Control architecture of
power systems: Modeling of purpose and function. In Power &
Energy Society General Meeting, 2009. PES ’09. IEEE, pages
1–8, July 2009.
[6] T. Kato, H. Kanamori, Y. Suzuoki, and T. Funabashi. Multi-
agent based control and protection of power distributed system
- protection scheme with simplified information utilization -
. In Intelligent Systems Application to Power Systems, 2005.
Proceedings of the 13th International Conference on, pages 49–
54, Nov. 2005.
[7] M. Pipattanasomporn and S. Feroze, H.and Rahman. Multi-
agent systems in a distributed smart grid: Design and imple-
mentation. 2009.
[8] A. Saleem, K. Heussen, and M. Lind. Agent services for situa-
tion aware control of power systems withdistributed generation.
In Power & Energy Society General Meeting, 2009. PES ’09.
IEEE, pages 1–8, July 2009.
[9] Arshad Saleem, Morten Lind, and Manuela M. Veloso. Mul-
tiagent based protection and control in decentralized electric
power systems. 2010.
[10] M. Wooldridge. An introduction to multiagent systems. Wiley,
2009.