An Architecture of a Multi-Agent System for SCADA

Dealing With Uncertainty, Plans and Actions

Sarah Calderwood, Weiru Liu, Jun Hong and Michael Loughlin

School of Electronics, Electrical Engineering and Computer Science,

Queen’s University Belfast, Belfast, U.K.

Keywords:

Autonomous Agents, Multi-Agent Systems, Sensors, SCADA, Uncertainty, Plans, Actions, Fusion.

Abstract:

This paper presents a multi-agent system approach to address the difﬁculties encountered in traditional

SCADA systems deployed in critical environments such as electrical power generation, transmission and dis-

tribution. The approach models uncertainty and combines multiple sources of uncertain information to deliver

robust plan selection. We examine the approach in the context of a simpliﬁed power supply/demand scenario

using a residential grid connected solar system and consider the challenges of modelling and reasoning with

uncertain sensor information in this environment. We discuss examples of plans and actions required for

sensing, establish and discuss the effect of uncertainty on such systems and investigate different uncertainty

theories and how they can fuse uncertain information from multiple sources for effective decision making in

such a complex system.

1 INTRODUCTION

Uncertainty (Graham and Jones, 1988; Halpern,

2003) is pervasive in sensor network based systems

and control systems. Information may be incomplete

or approximated due to limitations in the equipment

used to collect data or introduced by the algorithms

that process the data. It is sometimes infeasible to

exactly describe an environment, a particular situa-

tion, a future outcome or the possibility of multiple

outcomes. Plans are derived from system goals and,

taking into account the current environmental con-

ditions, they deﬁne a series of actions that aim to

achieve those goals. For example a grid connected so-

lar powered house may need to take an action such as

demand additional electricity from the National Grid

based upon information received from other sensors

or based upon other speciﬁc plans for that particu-

lar situation. A sensor may have accuracy limitations

that might introduce doubt whether this action should

take place. Uncertain information may cause a catas-

trophic error if this uncertainty is not dealt with prop-

erly.

Robust uncertainty modelling theory is required to

represent, reason and adequately represent reliabili-

ties associated with the sources of uncertain informa-

tion. Examples of theories include Dempster-Shafer

theory of evidence (Shafer, 1976) and possibility the-

ory (Dubois and Prade, 2011).

In a multi-agent system, agents communicate, co-

operate, compete and co-ordinate accordingly with

the other agents in order to minimize the degree of

uncertainty that arises to accomplish their individual

and global goals. Since agents collect uncertain in-

formation from different sources, it becomes neces-

sary to fuse (or combine) uncertain information from

multiple sources in order to determine the most plau-

sible model of the environment. Therefore developing

suitable fusion algorithms within each chosen theory

is important. In this paper we will investigate these

issues and illustrate how these uncertainty modelling

and fusion approaches can be integrated into multi-

agent systems.

The rest of the paper is structured as follows.

Section 2 provides a background stating the basic

terms. Section 3 describes a simpliﬁed power sup-

ply/demand scenario. Section 4 deﬁnes goals, plans

and actions for normal operation of three agents. Sec-

tion 5 describes the sensing environment and factors

contributing to uncertainty. Section 6 models uncer-

tainty and how to fuse uncertain information using

theories. Section 7 summarises conclusions and fu-

ture work.

300

Calderwood S., Liu W., Hong J. and Loughlin M..

An Architecture of a Multi-Agent System for SCADA - Dealing With Uncertainty, Plans and Actions.

DOI: 10.5220/0004594603000306

In Proceedings of the 10th International Conference on Informatics in Control, Automation and Robotics (ICINCO-2013), pages 300-306

ISBN: 978-989-8565-70-9

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

2 MULTI-AGENT SYSTEMS

AND SCADA

Multi-Agent Systems (Jennings and Wooldridge,

1998; Wooldridge, 2002) are comprised of interact-

ing autonomous agents that are well suited for appli-

cations in a dynamic, unreliable situation. Such sys-

tems provide robust models for representing complex

and real world environments such as power systems.

Multi-agent systems gather information from sensors

and other agents through communication, and then

reason with dynamic environments to make rational

and timely decisions or actions.

Typical multi-agent systems adopt a centralised

architecture (Pipattanasomporn et al, 2009) but prob-

lems arise because of the need to integrate increas-

ingly diverse components and to scale to larger de-

ployments. In addition the complexity and unpre-

dictability of the environment present unique prob-

lems for centralised solutions.

Decentralised systems offer the necessary control

and level of integration to help components work to-

gether and provide a methodology that helps to dis-

tribute tasks (Jennings et al, 1995).

There are many advantages in using multi-agent

systems including increased fault and noise tolerance,

increased ﬂexibility in design and scalability, en-

hanced security and increased efﬁciency. To demon-

strate and validate these advantages simulations will

be created using a simpliﬁed power supply/demand

scenario. This is beyond the scope of this paper.

SCADA (Supervisory Control and Data Acqui-

sition) is a proven and successful technology consist-

ing of a centralised system that monitors and controls

industrial processes, particularly in large scale critical

infrastructures such as power generation and power

transmission (Daneels and Salter, 1999; McArthur et

al, 2007; Arghira et al, 2011). SCADA systems are

attractive solutions in these application domains by

virtue of their ﬂexibility, simplicity, reliability and

ability to work autonomously in real time. SCADA

does however have a number of restrictions due to the

centralised control architecture especially in the areas

of ﬂexibility, scalability and resilience to failure or at-

tack (Yang et al, 2006).

Uncertainty in SCADA systems arises when sensor

data or inferred knowledge cannot be deemed accu-

rate. Applications must deal with inherent noise/error

in sensor data or knowledge as well as uncertainty,

incompleteness and inconsistent or conﬂicting data

from multiple sources (Sobh and Mahmood, 2002).

Humans have traditionally supervised such problems

to reason and resolve issues. In traditional SCADA

deployments the human supervisor takes the role

of encapsulating and handling inherent uncertainties

arising from incompleteness and inconsistencies. In-

telligent multi-agent systems need to perform this role

autonomously.

Domain knowledge captures environmentally

speciﬁc regulations, norms, restrictions, exceptions

and other domain-speciﬁc information. This knowl-

edge is used to generate plans of actions for a system

in a given state to achieve particular goals. Agents

need this knowledge to operate and reason with col-

lected sensory information for achieving their objec-

tives. Each agent may have its own speciﬁc knowl-

edge in addition to some common knowledge shared

amongst agents.

Domain knowledge is either elicited from domain

experts who are specialists in a chosen application

area or learned from domain speciﬁc datasets.

A common approach to capturing domain knowl-

edge is to use production rules that can readily be

represented and executed using Prolog. For example

a rule reacting to the reading of the voltage and fre-

quency of an overhead service line in a normal state

can be formatted as

If Voltage(Reading) ∈ (228V, 252V) ∧

Frequency(Reading) = 50Hz

Then Transmit(Electricity) ∧ Update(Meter)

Goal and action modelling helps to achieve au-

tomated intelligent control, reducing the need for hu-

man interaction and executing in real-time to suit var-

ious environmental conditions that may arise in the

system such as uncertainty, security issues or com-

ponent failure. From a description of possible states

of the world, the desired goals and a set of possible

actions, a plan can be determined that is guaranteed

from any of the initial states to generate a sequence of

actions that leads to one of the goal states.

3 POWER SCENARIO

For the purpose of illustration, this paper will use in

its simplest form, a grid connected solar system for

successful power generation and supply (see Figure

1). This is part of a larger scenario for power gen-

eration, transmission and consumption. It consists of

three agents (a substation, a distribution transformer,

a house) and multiple sensors to measure voltage, fre-

quency etc. Other agents required will aid decision

making, record status and events, retrieve data from

equipment as shown in Figure 2. The approach can

be readily transferred to similar applications operat-

ing in other SCADA-related domains.

Power from Grid. We assume electricity has

been generated at a power station and transmitted over

AnArchitectureofaMulti-AgentSystemforSCADA-DealingWithUncertainty,PlansandActions

301

transmission lines to a substation located near to con-

sumers. The AC frequency of the power supply is 50

Hertz (Hz).

Figure 1: A simple grid connected solar system.

Substation transformers reduce high voltage electric-

ity to lower voltage electricity. Medium/low volt-

age distribution transformers convert the power from

main distribution lines to 240 volts (V) to serve resi-

dential loads. Electricity is delivered on low voltage

overhead service lines to a meter located inside the

building. A fuse box will trip to protect from prob-

lems such as short-circuits or appliance faults.

Solar Power. When the sun shines, solar pho-

tovoltaic panels generate direct current (DC). An

inverter converts DC into 240V alternating current

(AC) suitable for household appliances. A meter mea-

sures electricity production and consumption. Excess

electricity is fed into the National Grid where credit

is given by a utility company. If solar becomes un-

available then accumulated credits are used to offset

electricity drawn from the grid. Reverse power ﬂow

causes problems such as voltage control, power qual-

ity issues, difﬁculty of isolation for maintenance etc.

Figure 2: A sample solution of the SCADA agents and sen-

sor architecture for simpliﬁed power scenario.

4 GOALS, PLANS AND ACTIONS

For the purpose of illustration we use, in its simplest

form, three agents for solar power generation and dis-

tribution (a substation, a distribution transformer, a

house) each equipped with the relevant sensors.

The goals, plans and actions are captured in the

AgentSpeak programming language (Rao, 1996; Bor-

dini et al, 2007). AgentSpeak is an agent-oriented

programming language based on logic programming

and the BDI architecture for autonomous agents.

Goals. The pro-active behaviour of agents is pos-

sible through the notions of goals. A goal is a de-

sired state of the world. Each agent will have indi-

vidual goals that they strive to independently achieve

depending on their state (for example normal, emer-

gency, failure, recovery).

For example the house agent has a goal to deliver

sufﬁcient power to household appliances at an accept-

able voltage level and AC frequency. More specially,

to achieve this goal the house (fusebox, meter etc.)

must be switched on with the power supply operat-

ing at 50Hz and the voltage between 228V and 252V

(nominal voltage allows ±5%).

Similarly a goal to switch off a working trans-

former for maintenance requires informing the par-

ticipating components e.g. substation, consumers etc.

of the goal in order that they can react appropriately

to the overall system goals for this situation.

Plans. Each emerging situation (determined from

the information provided by sensing/observing agents

in the environment) must be complemented by a plan

of actions to follow to transform the initial situation

to one that satisﬁes some goal.

An AgentSpeak plan has the general structure of:

triggering event : context <- body

where the triggering event denotes the events that the

plan handles or a goal an agent needs to see achieved.

The context represents the circumstances in which the

plan can be used and the body is the course of action

to be used to handle the event if the context is believed

true at the time a plan is executed.

Substation Agent:

+!startSystem: true <- switchOnSystem;

runSystem.

+!runSystem: on(system) <- !startHouse;

!startTransformer.

Solar House Agent:

+!startHouse: on(system) <- solarHouseOn;

!runHouse.

+!runHouse: on(house) <- checkHouseSensors;

!runningHouse; !runHouse.

+!runningHouse: on(house) & frequency(50) &

(voltage(228) | voltage(252)) & on(fusebox) &

on(appliance) & meter(credit+1) <-

workSolarHouse.

+!runningHouse: on(house) <- switchOffHouse;

.send(system, untell, on(house));

.send(transformer, untell, on(house)).

ICINCO2013-10thInternationalConferenceonInformaticsinControl,AutomationandRobotics

302

Transformer Agent:

+!startTransformer: on(system) & on(house) <-

switchOnTransformer; !runTransformer.

+!runTransformer: on(transformer) <-

checkTransformerSensors; !runningTransformer;

!runTransformer.

+!runningTransformer : on(transformer) &

on(house) & (voltage(228) | voltage(252)) &

frequency(50) <- workTransformer.

+!runningTransformer: on(transformer) &

on(solarPower) <- switchOffTransformer;

.send(system, untell, on(transformer));

.send(house, untell, on(transformer)).

Actions. For the normal operation of the genera-

tion and distribution of solar power these typical ac-

tions may be executed for the above plans. These ac-

tions are speciﬁc to each agent and may rely on further

actions taken by other agents simultaneously.

switchOnSystem

Preconditions: ¬on(system)

Effects: on(system)

solarHouseOn

Preconditions: ¬on(house)

Effects: on(house)

workSolarHouse

Preconditions: on(house)

Effects: on(solarPower)

checkHouseSensors

Preconditions: on(house)

Effects: on(solarPower) & frequency(50) &

(voltage(228) | voltage(252)) &

on(fusebox) & on(appliance) & on(meter)

switchOnTransformer

Preconditions: ¬on(transformer)

Effects: on(transformer)

workTransformer

Preconditions: on(transformer)

Effects: on(gridPower)

checkTransformerSensors

Preconditions: on(house) & on(transformer)

Effects: on(gridPower) & frequency(50)

(voltage(228) | voltage(252))

These plans and actions can be further reﬁned to

account for the real complexity in a working power

system.

5 SENSING ENVIRONMENT FOR

MULTI-AGENT SYSTEMS

In a real-world multi-agent system, the environment

plays a prominent role. Constantly sensing the envi-

ronment provides situational awareness and allows an

agent to adapt to environmental changes by selecting

or generating new plans as needed to ultimately en-

sure agents meet their own goals.

The environment can be considered to include all as-

pects of a system that are not owned or hosted within

an agent, including the infrastructure by which agents

communicate.

Certain environmental aspects may become un-

available to an agent but which are required for deci-

sion making about further actions. Informed guesses

may be required to progress. Uncertainty is an issue

that may arise and will have an impact on plan se-

lection. The following four aspects are factors that

contribute to uncertainty in a SCADA system such as

power distribution.

Supply and Demand Issues. Supply and demand

is a dynamic balance made more difﬁcult because

there are no scalable methods of storing electricity. It

is difﬁcult to predict and generate exactly the amount

of electricity required for consumer needs and to pro-

tect the grid from power overload and damage. Black-

outs occur if demand exceeds supply and brownouts

occur if power supply drops below demand. Main-

tenance of this balance is challenging. Output from

solar panels and other renewable sources are favoured

over fossil fuel based resources and generating power

stations may suspend generation temporarily until any

surplus electricity has been redistributed.

Future Popularity of Electricity and Con-

sumers Behaviour. It is predicted that electricity will

be increasingly relied upon to meet new demands for

electric heating, air conditioning and electric vehicles.

The unpredictable behavioural effects of consumers,

changes in their geographical distribution, the effect

of economic incentives etc. will also impact the po-

tential demand making it hard to estimate and plan

for future needs. For example if the price of electric-

ity becomes more expensive then consumers may be

more careful with usage.

Data Acquisition. Uncertainty arises from sens-

ing instrumentation or techniques used to make the

measurement. Moreover there is the possibility of hu-

mans, systematic errors or external deliberate intru-

sions compromising the reliability of data. For ex-

ample the measurement or updating of a voltage sen-

sor could be broken or a human may introduce an er-

ror when interpreting a sensor reading. The precision

and accuracy of the sensor could be signiﬁcant to the

problem i.e. voltage is within a certain range and fre-

quency is an exact value.

Unpredictability of External Events. Uncer-

tainty arises if components required to distribute elec-

tricity are tampered. This can be accidental or mali-

cious. Projections of human behaviour are not eas-

ily amendable to prediction. Uncertainty may be

embedded into environmental factors that are out of

our control (e.g. humidity), weather conditions (e.g.

AnArchitectureofaMulti-AgentSystemforSCADA-DealingWithUncertainty,PlansandActions

303

wind, lightning), operational events (e.g. compo-

nent failure) and economic events (increased electric-

ity prices, costs to repair components) etc. Agents

need to predict future events based on many different

types of information and environmental factors. It can

be impossible to be certain that events will occur. Dif-

ferent sources will have different degrees of reliability

when agents make use of their information.

It is advantageous to handle properly uncertain

sensor information to enhance the performance of an

agent system. A multi-agent system that has adopted

measures to minimise uncertainty will reduce the

chances of failure/disaster occurring e.g. if new ev-

idence has been collected or observed and introduced

to the system an agent’s beliefs will need to be revised

or updated to take into account new information. It

will also result in better autonomous decision making

because robust uncertainty handling would limit the

need for human interaction and so allow for real-time

situational awareness.

6 UNCERTAINTY MODELLING

Many techniques have been developed for represent-

ing and processing uncertainty depending on the in-

formation given.

6.1 Dempster-Shafter Theory (DST)

This theory (Shafer, 1976) extends probability theory

and offers a mechanism to combine uncertain infor-

mation from distinct sources (represented by belief

functions) using Dempster’s combination rule. Mass

values are assigned to sets of possibilities rather than

singleton events. This model can cope with varying

degrees of precision.

Deﬁnition 1: A frame of Discernment

Set Ω={ω

,...ω

} is called a frame of discern-

ment(frame) if one and only one ω

is true at a time.

Example 1. data acquisition - assume there are

two voltage sensors on a distribution line and we want

to monitor the readings. For simplicity, assume the

reading

is either Normal(n) or Abnormal(¬n), then

Ω = {n, ¬n}.

Deﬁnition 2: Basic Belief Assignment (BBA)

Let m be a function on Ω as m: 2

Ω

→ [0,1] then m is

a BBA iff

∑

A m(A)=1. When m(∅)=0 is required, it

is called a mass function.

Example 2. Each sensor (S

) will contribute its

observations by assigning beliefs over Ω. The mass

It is possible to deﬁne a set of voltage values as the

elements of a frame.

function derived from sensor S

reading is denoted

by m,

m({n})=0.5, m({¬n})=0.3, m(Ω)=0.2

If m(∅)6=0, m is referred to as self-inconsistent,

which is equivalent to the open world assumption.

This means this is something we do not know there-

fore it is not modelled in Ω or the sensor information

is wrong.

Deﬁnition 3: Pignistic Transformation

In order to aid decision-making we use pignistic

transformation (Smets, 2004) which transforms a

BBA to a probability function.

Let m be a BBA on Ω. Its associated pignistic proba-

bility function BetP

: Ω→[0,1] is deﬁned as:

BetP

(ω) =

∑

A⊆Ω,ω∈A

|A|

m(A)

1 − m(

, m(

0) 6= 1

where |A| is the cardinality of subset A.

Example 3. Using Ex. 2,

BetP

(n) =

m({n})

|{n}|

m({Ω})

|Ω|

= 0.6

Similarly BetP

(¬n) = 0.4.

This means that it is highly likely voltage on the cable

is Normal, so no immediate actions are needed.

6.2 Possibility Theory

Dubois and Prades (2011) theory is suitable for mod-

elling an agents knowledge that is incomplete, there-

fore complementing probability theory.

Deﬁnition 4: Possibility Theory

Let Ω be a frame of discernment consisting of a set of

possible solutions.

A fundamental function in possibility theory is a

possibility distribution π : Ω → [0,1]. π is said to be

normal iff ∃ω

∈ Ω such that π(ω

) = 1. π(ω) means

the maximum likely possibility of ω being true.

Example 4. Using Ex. 2, π(n)=1, π(¬n) = 0.5

A possibility distribution is more like a quantita-

tive comparison of its alternatives. This is more suit-

able when evidence is less precise to deﬁne either a

probability function or a mass function. It has been

proved that any possibility distribution π can be con-

verted into a mass function.

6.3 Fusing Uncertain Information

In a complex system, certain information collected

from different or multiple sources such as data from

sensors or agents can be fused to derive more precise

beliefs and help achieve situation awareness. Percep-

tions can be built which allows the selection of appro-

priate plans. By retrieving agent goals a decision can

be made on the actions to execute.

ICINCO2013-10thInternationalConferenceonInformaticsinControl,AutomationandRobotics

304

Different theories will deploy different combination

mechanisms. For example, Dempsters combination

rule for DS theory (Shafer, 1976), states that the rule

cannot be applied if pieces of information are almost

conﬂicting or coming from sources that may have in-

ﬂuenced one another.

Deﬁntion 5: Dempster’s Combination Rule

Given two mass functions m

and m

, their combined

mass function is:

1⊕2

(C)=m

⊕m

(C)=(1/1-k)

∑

A∩B=C6=∅

(A)

(B)

k =

∑

A∩B=∅

(A) m

(B) where k is the degree of

conﬂict.

For each proposition such as Normal(n), this the-

ory gives a rule of combining sensor S

and S

obser-

vation m

and m

respectively.

Table 1: This table shows the results of using the combina-

tion rule.

↓ m

→ {n},0.6 {n,¬n},0.4

{n},0.4 {n},0.24 {n},0.16

{n,¬n},0.6 {n},0.36 {n,¬n},0.24

Therefore m

1⊕2

({n})=0.76, m

1⊕2

({n,¬n})=0.24.

This reading indicates the voltage on the line is

normal.

For heterogeneous uncertain information, we can

merge mass functions (m) with possibility distribu-

tions (π), after π can be converted into another m’.

This then allows m and m’ to be combined using DS

values (Hunter and Liu, 2006).

As a result of uncertainty, agents have chances to

take different plans to achieve the same goal. They

need to evaluate which plan to take under such utility

through a selection function that is still to be devel-

oped.

7 CONCLUSIONS AND FUTURE

WORK

Uncertainty models are approximations because it is

infeasible to eliminate uncertainty entirely. It be-

comes necessary to model a problem domain, incor-

porating appropriate fusion algorithms, in such a way

that is suitable for the type of uncertainty evident in

the system to capture the true nature of the real world

domain.

Forthcoming research will involve the design and

implementation of a multi-agent SCADA system tak-

ing the power system as an example, dealing with un-

certainty, multi-source information, as well as their

effects on agent goals and plans. To recognise dif-

ferent environments, goals, plans and actions will be

incorporated using sensor data and domain knowl-

edge. A selection function will be developed to decide

which plan is most appropriate to achieve the goal

depending upon the prevailing circumstances. This

can help in future decision making and planning. The

AgentSpeak language has been used to model plans

and goals. Jason (Bordini et al, 2007) will be used

to develop a BDI agent architecture for the SCADA

power control. We plan to integrate our BDI architec-

ture with the Electronic Institutions framework (Ar-

cos et al, 2005) in order to model norms/regulations.

REFERENCES

Arcos, J. L., Esteva, M., Noriega, P., Rodrguez-Aguilar, J.

A. and Sierra, C. (2005). Engineering open environ-

ments with electronic institutions. In Engineering ap-

plications of artiﬁcial intelligence. 18(2), 191-204.

Arghira, N., Hossu, D., Fagaraan, I., Iliesc, S.S. and Cos-

tianu, D.R. (2011). Modern scada philosophy in

power system operation A survey. In UPB Scientiﬁc

Bulletin, Series C: Electrical Engineering. 73(2), pp.

153-166.

Bordini, R. H., Hubner, J. F., Wooldridge, M. (2007). Pro-

gramming Multi-Agent Systems in AgentSpeak using

Jason. John Wiley and Sons.

Daneels, C. and Salter, W. (1999). What is SCADA? In In-

ternational Conference on Accelerator and Large Ex-

perimental Physics Control Systems. Italy, 339-343.

Dubois, D. and Prade, H. (2011). On possibility theory,

formal concept analysis and granulation: Survey. In

Applied and Computational Mathematics. 10(1), pp.

10-19.

Graham, I. and Jones, P. (1988). Expert systems: knowl-

edge, uncertainty and decision. Chapman and Hall.

Halpern, J. (2003). Reasoning about uncertainty. The MIT

Press, Cambridge, Massachusetts, London, England.

Hunter, A. and Liu, W. (2006). Fusion rules for merging

uncertain information. In Information Fusion Journal.

7(1):97-134.

Jennings, N. R., Corera, J. M. and Laresgoiti, I. (1995). De-

veloping industrial multi-agent systems. In 1st Int.

Conf. on Multi-Agent Systems (ICMAS ’95). 423-430.

Jennings, N. and Wooldridge, M. (1998). Agent Tech-

nology: Foundations, Applications, and Markets.

Springer, Berlin.

McArthur, S. D. J., Davidson, E. M., Catterson, V.M.,

Dimeas, A. L., Hatziargyriou, N. D., Ponci, F.

and Funabashi, T. (2007). Multi-agent systems for

power engineering applications - Part I: Concepts, ap-

proaches, and technical challenges. In IEEE Transac-

tions on Power Systems. 22(4), pp. 1743-1752.

Pipattanasomporn, M., Feroze, H. and Rahman, S. (2009).

Multi-agent systems in a distributed smart grid: De-

sign and implementation. In IEEE/PES Power Sys-

tems Conference and Exposition. PSCE 2009.

AnArchitectureofaMulti-AgentSystemforSCADA-DealingWithUncertainty,PlansandActions

305

Rao, A. (1996). Agentspeak(l): BDI agents speak out in

a logical computable language. In Agents Breaking

Away. Springer.

Shafer, G. (1976). A Mathematical Theory of Evidence.

Theory of Evidence. Princeton University Press.

Smets, P. (2004). Decision making in the TBM: the neces-

sity of the pignistic transformation. In International

Journal of Approximate Reasoning. 133-147.

Sobh, T. and Mahmood, A. (2002). Recovering structure

uncertainties from noisy sense data. In Computers

and Electrical Engineering. IEEE/PES Power Systems

Conference and Exposition. 28(2), pp. 127-141.

Wooldridge, M. (2002). An Introduction to Multiagent Sys-

tems. Wiley, New York.

Yang, Z., Ma, C., Feng, J. Q., Wu, Q. H., Mann, S. and Fitch

J. (2006). A multi-agent framework for power system

automation. In International Journal of Innovations

in Energy Systems and Power. Vol. 1, no. 1.

ICINCO2013-10thInternationalConferenceonInformaticsinControl,AutomationandRobotics

306