Impacts of Communication Failure on PowerSystem Operations

GU ChaojunDepartment of Electrical and Computer Engineering

National University of SingaporeSingapore

Email: chaojun@nus.edu.sg

Panida JirutitijaroenDepartment of Electrical and Computer Engineering

National University of SingaporeSingapore

Email:elejp@nus.edu.sg

Abstract—Communication technology serves important func-tions in the power system energy management system (EMS).Recent studies found that cyber-attack and terrorist attack cancause failure of power system communication which affectsEMS in various ways. This paper analyzes two impacts ofcommunication failure on EMS, namely, system observability andsystem control. The first impact on system observability is studiedusing DC state estimation to identify unobservable states whenloss of communication occurs at various locations. The secondimpact on system control is studied using DC OPF to find the lossof load caused by communication failure. Simulation studies onthe two analyses are conducted on IEEE 14 bus and IEEE 30 bussystems. Results show that certain substations are more criticalfor system observability. Optimal upgrade strategy is providedin this paper to improve system observability. It is also shownthat failure of communication will cause extra loss of load. Byprotecting certain substations, the system can expect less loss ofload. It is concluded that communication system is crucial tothe power system operation. This paper also gives suggestion tooptimally improve communication systems.

I. INTRODUCTION

Communication technology is widely used in supervisorycontrol and data acquisition (SCADA) system of Power Grid.SCADA serves the function of monitoring and controlling thegrid. SCADA system generally includes control centers andremote terminal units (RTU) [1]. RTUs gather system data atsubstations. These data will be sent to the control center. Insidethe control center, data from RTUs will be passed to EMSfor several important functions, including state estimation,contingency analysis, etc [1]. If necessary, EMS will suggestthe control center to send commands to RTUs to reconfigurethe grid. Communication between RTUs and the control centercan be in different forms, failure of communication will makethe overall system operation at risk.

A. Power System Communication

Power systems can be considered as containing two layers,current carrying layer and information layer [2]. Electricityflows in current carrying layer, which includes generators,transmission lines, circuit breakers, etc. Information layerwhich is also known as SCADA system, includes multiple

This work was supported by Singapore Ministry of Education-AcademicResearch Fund, Grant No. WBS R-263-000-691-112.

978-1-4799-2522-3/13/$31.00 c©2013 IEEE

RTUs and control centers. The information layer relies oncommunication technology to monitor and control the grid.There are multiple forms of communication between RTU anda control center. It can be fiber, radio or satellite communica-tion. These communications use different protocols. The mostwidely used protocols are DNP3.0 for US and IEC 60870-5 for Europe [3]. These protocols are publicly accessible,which makes it a potential threat to SCADA security. Oncethe information layer is broken, system operator will losecontrol of the grid. In this paper, we focus our analysis on theinformation layer of the system, which is explained below.

1) RTU: RTU is usually attached to a substation. Differentsubstations can be geographically far apart. In each substation,there are several measuring devices. These devices measureanalog and discrete data. Analog data includes bus voltagemagnitude, active power flow. If the substation has a phasemeasurement unit (PMU), it can measure the bus phase angle.Discrete data includes circuit breaker status and alarm status.All these measured data will be sent back to the controlcenter for processing. Different types of data need to meetdifferent communication requirements. Some of them are moreurgent than others. For example, measurement data are sentin a higher frequency than log data. Apart from sendingmeasurement data, RTU also receives commands from thecontrol center to reconfigure substations. These commandsinclude opening circuit breakers, load shedding and changingtransformer tap positions.

2) Control Center: Control center gathers data from RTUs.These data will be processed in EMS. Inside the EMS, rawdata first go through state estimation to filter bad data. Afterthat, EMS uses these data to do contingency analysis andseveral other functions. EMS helps the grid to operate atnear optimal state. All these functions rely on a functioningSCADA system. Failure of communication between RTUs andthe control center disables EMS. History shows that large-scale blackouts are usually not initiated by communicationfailure, but after the initial fault, cascading effects are mostlycaused by communication system failure [4], [5] . Thesecommunication failures can be caused by malfunctions ofcommunication equipments or from cyber-attacks.

B. Cyber Threats on Power System

Given the importance of SCADA systems to power grid,cyber security of SCADA becomes a crucial issue. The currentSCADA system was designed decades ago as an internalnetwork with limited access points. SCADA vendor used toprovide only limited security features. With the developmentof smart grid, which relies more heavily on informationsystem, cyber security issue becomes even more severe. Gov-ernments and authorities are therefore working on developinga security framework for power systems [6], [7] .

This paper studies the impacts of communication failure onpower system operations. Two types of impacts are analyzedin this paper: system observability and load shedding control.Section II discusses system unobservability caused by commu-nication failures. If a control center cannot receive data fromcertain RTUs, the system will become unobservable. We enu-merate a list of communication scenarios that lead the systemto become unobservable. Results show that certain substationsare more important than others for system observability. Wethen suggest the locations for substation upgrade to reduce therisk of the system becomes unobservable. Section III discusseshow communication failure will affect power system control.We utilize DC OPF to find the optimal loss of load whenthere are communication failures between control center andRTUs. Result shows that failure of certain communicationlinks between the control center and RTUs causes extra lossof load. Section IV provides case study for the above twostudies using IEEE 14 and 30 bus system. Case study showsthe importance of communication network to power systems.Conclusions are given in Section V.

II. POWER SYSTEM UNOBSERVABLE ANALYSIS

One impact of communication failure is system observ-ability. The power system control center uses data gatheredfrom substations to perform state estimation. If communicationbetween the RTU and the control center is blocked, thecontrol center may not be able to perform state estimation.This section of paper will discuss how communication failureaffects state estimation and what the best strategy to improvesystem observability is. There are mainly two methods forsystem observability study, one is network topology, and theother is numerical analysis which is applied in this paper.

A. Assumption

We assume that a remote terminal unit (RTU) is attached toeach bus. Each RTU is capable of measuring active power flowin all transmission lines connected to that bus. An RTU canbe either working or failed. A working RTU at bus i measuresactive power flow of all transmission lines that connect to busi. This information will be sent back to the control center.A failed RTU will not send any information to the controlcenter. It is assumed that the control center knows the system’sstructure. Bus phase angles measured by PMU devices arenot considered in the case study, but can be included by littleadjustment of the method used in the analysis.

B. DC State Estimation

The states to be estimated in DC state estimation are busphase angles. If the state estimator can determine all bus phaseangles, the system is observable. Otherwise, the system isunobservable. DC power flow is used for formulating DC stateestimation, z = Hx+e, where x is the vector of bus phase an-gles to be estimated. x = (θ1, θ2, ..., θn)′. z is the active powerflow at each transmission line. We use Pij = Bij × (θi − θj)where Pij : active power flow in transmission line betweenbus i and j. Bij : branch susceptance between bus i and j. θi: phase angle at bus i.

The H matrix is initially set to be a m × n zero matrix,where m = transmission lines no.+1. Note that +1 standsfor setting bus 1’s angle as reference angle, H(m, 1) = 1. n =total buses. The ith row of matrix H will present the activepower flow of the ith transmission line. Suppose that the ithtransmission line connects bus j and k, then if at least one ofthe two RTUs at bus j or k is functioning, set H(i, j) = Bij ,H(i, k) = −Bij .

After formulating the H matrix, rank of H is checked. Ifrank(H) equals to the bus number, the system is observable,otherwise the system is unobservable.

C. Unobservable List

Given that each RTU can be either working or failed, for an bus system, there is a total of 2n communication scenarios.These scenarios cover from all RTUs working to all RTUsfailed. If we do a state estimation rank check for all thesescenarios, we can generate a list of scenarios that the systembecomes unobservable. We call this list full unobservable listfor the n bus system, Un.

For a large system, it is impractical to check all these2n scenarios. One possible approximation is to analyze upto a limited number of RTUs failure; for example, up to5 RTUs failure. This approximation will be acceptable ifthe probability of having more than 5 RTUs failure at thesame time is very small. By doing so, instead of checking2n scenarios, we only need to check

(n0

)+(n1

)+ ... +

(n5

)scenarios. Passing these scenarios to the rank check, we getan unobservable list which has up to 5 RTUs failure, Un

5 .Similarly, we can generate an unobservable list of up to kRTUs failure, Un

k .

D. Probability of System Becoming Unobservable

If we assign each RTU with a failure probability, we can findthe probability of the system become unobservable. Define pfii = 1, 2, ..., n as the probability of RTU at bus i fails. For eachscenario in the unobservable list, we can find its probability ofoccurring. For example, if the unobservable list contains thescenario that all RTUs fail, this scenario occurs at a probabilityof pf1 × pf2 × ...× pfn. By adding up all these scenarios in theunobservable list, we can find the probability of the systembecoming unobservable.

E. Optimal Upgrade on RTUs

One way to improve the system observability is to upgradethe RTUs. Upgrades can be done by replacing old communi-cation equipment with new ones. In this analysis, we assumethat an upgrade on RTU will decrease the probability ofRTU failure, thus decrease the probability of the system be-comes unobservable. We should notice that upgrade differentcombinations of RTUs give different results. To find whichcombination of RTUs to upgrade for optimal improvement onobservability, we can do it mathematically or by enumeration.Enumeration method is given by a flow chart Fig.1.

Fig. 1. Enumeration Method Flow Chart

Mathematically, the problem can be transformed into anonlinear optimization problem. The original probability ofRTU failure is defined as pfi . After upgrading, the probabilityof RTU failure becomes pfi − δpfi where δpfi is the improve-ment of reliability as the result of investment. Improvementshould be in the range of 0 < δpfi < pfi . If a systemoperator decides to upgrade m RTUs, the problem is equivalentto minimize probability of system becoming unobservable.min(

∑Unk (pf1 −∆1δ

pf1 )(pf2 −∆2δ

pf2 )...(pfn−∆nδ

pfn )) where∑n

i=1 ∆i = m, ∆i = 0 or 1, m is the number of RTUupgrade.

III. POWER SYSTEM CONTROL

Another important impact of communication failure is sys-tem control. One advantage of the SCADA system is the re-placement of passive automatic load shedding with intelligentload shedding (ILS). Passive load shedding relies on undervoltage [8] and under frequency [9] relay for load shedding.ILS [10]–[12] uses information gathered from RTUs to makedecisions. Based on this information, the control center canprovide more accurate and optimal load shedding strategy. Ifthe communication channel is congested, the RTUs will not beable to receive commands of load shedding. This could leadto cascading effect and loss of load.

A. Assumption

In this analysis, we assume that if the communicationbetween the control center and RTU at bus i failed, then load

on bus i cannot be shed, otherwise, any amount of load canbe shed.

B. Flow chart

Note that load shedding action is only needed when systemstate changes. In this analysis, we calculate the loss of loadwhen N-1 contingencies happen. Contingency can be eithergenerator or transmission line failure. The process to find theoptimal loss of load is shown in Fig.2. It is a combination ofrelaxed DCOPF from research [13] and cascading analysis inblackout research [14]. The flow chart is capable of simulatingthe cascading effect by using fast transit part of OPA method[15], [16], which is widely used in blackout research.

Fig. 2. Flow Chart to Find Optimal Loss of Load

In step 2, the object function is to minimize loss of load.

min(∑i

Di − Pi) (1)

where Di and Pi stands for demands and injected power at busi. The DCOP subject to generator constraints and transmissionline constraints.

In step 3, the relax the constraints of the generator andtransmission line. Add penalty to objective function. The newobjective function is

min(∑i

Di − Pi + wp ×∑i,k

Efij + wp ×

∑i

EGi ) (2)

where Efij and EG

i stands for excessive flow and generationrespectively. wp is the penalty for exceed the limit.

IV. CASE STUDY

IEEE 14 and 30 bus systems are used as the test system forboth analyses in this case study.

A. System Observability Test and Upgrade Strategy

The probability of a system becoming unobservable dependson the probability of communication failure at each substation.In this case study, we assume that each substation has the sameinitial probability of communication failure. When the prob-ability of communication failure at substations increases, theprobability of system becoming unobservable also increases.When pfi = 1 for all i, the system is surely unobservable.

To improve the probability of system being observable, asystem operator needs to upgrade the RTUs at substations. Weassume that each substation has an initial failure probabilityof pfi = 0.1 . After the RTU upgrade, the failure probabilitydrops to 0.01 . We apply enumeration method to determinethe optimal upgrade strategy. Fig.3 shows that for the samenumber of RTUs upgrade, different combination of RTUs havedifferent results. The gap between optimal upgrade and leasteffective upgrade is quite big. The optimal upgrade strategy isshown in Table I.

Fig. 3. Upgrade IEEE 14 bus system

TABLE IIEEE14 BUS OPTIMAL UPGRADE n RTU

n Punob Optimal Strategy

0 0.0167 01 0.007 72 0.0052 2 73 0.0034 2 6 74 0.0017 2 6 7 95 0.0007 2 6 7 8 96 0.0005 2 6 7 8 9 137 0.00034 2 6 7 8 9 11 138 0.00025 2 4 6 7 8 9 11 139 0.00016 2 4 5 6 7 8 9 11 1310 0.00014 2 4 8 6 7 8 9 10 11 13

For the IEEE 30 bus system, it is unrealistic to list allpossible communication failure scenarios. We then generate anunobservable list resulting from up to 7-RTU failures. If eachsubstation has 0.1 failure probability, we can obtain an initialprobability of being unobservable using different unobservablelists. Test shows that Punb = 0.0287 for U30

5 , Punb = 0.0372for U30

6 , Punb = 0.0420 for U307 . The Punb difference between

U307 and U30

6 is much smaller than the difference between upto U30

6 and U305 . This means that it is acceptable to use U30

7 .The optimal upgrade strategy for 30 bus system using U30

7 isshown in Table II.

TABLE IIIEEE30 BUS OPTIMAL UPGRADE USING U30

7

Upgrade n buses Punob Optimal Upgrade

0 0.042 01 0.0326 252 0.023 12 253 0.0139 9 12 254 0.0099 9 10 12 25

Results from IEEE 14 and 30 bus systems show that anupgrade on substation communication equipment will decreaseprobability of system being unobservable. Note that cost ofupgrade on different substations is not considered in theanalysis.

B. Loss of Load Analysis

The loss of load analysis has been tested on IEEE 14 busand IEEE 30 bus system. An N-1 contingency is applied toboth systems. The contingency includes generator failures andtransmission line failures.

Table III shows loss of load for IEEE 14 bus N-1 con-tingency analysis when communication failure happens ondifferent buses. The first and second columns show the typesof contingency, either generator failure or transmission linefailure. G stands for generator, T stands for transmissionline. Columns 3 to 15 show different communication failurescenarios and resulted loss of load. Column 3 means allcommunications are working. Column 15 means all commu-nications failed. Columns 4 to 14 represent only one buscommunication failed.

The first test case in the first row shows that when there is nocontingency, the system will have no loss of load, regardlessof communication system status. This result is reasonablebecause when there is no contingency, the system does notneed to change its configuration. The second and the thirdtest cases show that when generator one or two fails, lossof load will inevitably occur. Since generator one is the maingenerator in the system, loss of this generator will incur higherloss of load. Compared with generator one, generator two has asmaller capacity, thus its failure will incur less loss of load. Fora transmission line failure contingency, only line one and twofailures will cause loss of load. This is because transmissionlines one and two are the two major transmission lines, when

TABLE IIIN-1 CONTINGENCY ON IEEE 14 BUS SYSTEM

G T full control B2 B3 B4 B5 B6 B9 B10 B11 B12 B13 B14 All Lost

0 0 0 0 0 0 0 0 0 0 0 0 0 0 01 0 2.0930 2.0930 2.5900 2.0977 2.0930 2.1011 2.1148 2.0995 2.0956 2.0974 2.1028 2.1038 2.592 0 0.1641 0.1819 0.1641 0.1641 0.1641 0.1641 0.1641 0.1641 0.1641 0.1641 0.1641 0.1641 2.590 1 0.5313 0.5313 0.5313 0.5313 0.5313 0.5313 0.5320 0.5314 0.5313 0.5313 0.5313 0.5315 2.590 2 0.4473 0.4473 0.4473 0.4473 0.4473 0.4473 0.4473 0.4473 0.4473 0.4473 0.4473 0.4473 2.590 3-20 0 0 0 0 0 0 0 0 0 0 0 0 0

other transmission line fails, energy can flow through theneighboring lines.

The test case also shows that for the same contingency,different communication failure scenarios will cause differentamount of loss of load. If a communication failure happenson the bus that requires load shedding for optimal operationstrategy, an increase of loss of load will happen. For example,in the generator two failure case, only when communicationon bus two failure will incur a higher loss of load. This isbecause when generator two fails, the optimal load sheddingis to shed load on bus two. Sometimes optimal load sheddingrequires shedding load on several buses, like generator onefailure case, then all these buses’ communication need to beprotected for optimal load shedding.

For test case two to five, when all communications arelost, the system will lose all its loads. This is caused bythe cascading effects. When all communications are lost, thesystem cannot shed any load which leads to an overload oftransmission lines or generators. If a generator is overloaded,because no load can be shed, the generator will automaticallydisconnect from the grid. If transmission line are overloaded,overload relay will open the transmission line, thus causingmore lines to be overloaded. The same analysis was appliedon IEEE 30 bus where results show same conclusion.

V. CONCLUSION

In this paper, we studied at how communication failureaffects power system in monitoring and control. For moni-toring function, we found that a communication failure couldcause the system to be unobservable. Additionally, certainsubstations are more important than others for system observ-ability. To improve system observability, this paper providessuggestions on how to optimally upgrade communicationequipment on substations. For power system control, we foundthat a communication failure could cause extra loss of load.Communication failures cause cascading effects which lead tounnecessary loss of load. By protecting certain substations, thesystem will experience less loss of load.

REFERENCES

[1] S. W. Blume, “SYSTEM CONTROL CENTERS AND TELECOMMU-NICATIONS,” in Electric Power System Basics. Hoboken, NJ, USA:John Wiley & Sons, Inc., Sep. 2007.

[2] C. Singh and A. Sprintson, “Reliability assurance of cyber-physicalpower systems,” in 2010 IEEE Power and Energy Society GeneralMeeting. IEEE, Jul. 2010, pp. 1–6.

[3] F. Cleveland, “Enhancing the reliability and security of the informationinfrastructure used to manage the power system,” in IEEE PowerEngineering Society General Meeting, 2007. IEEE, Jun. 2007, pp.1–8.

[4] Z. Xie, G. Manimaran, V. Vittal, A. G. Phadke, and V. Centeno, “Aninformation architecture for future power systems and its reliabilityanalysis,” IEEE Transactions on Power Systems, vol. 17, no. 3, pp. 857–863, Aug. 2002.

[5] “Final report on the august 14th blackout in the united states andcanada,” U.S.-Canada Power System Outage Task Force, Tech. Rep.,Apr. 2004.

[6] G. N. Ericsson, “Information security for electric power utilities(EPUs)CIGR07 developments on frameworks, risk assessment, andtechnology,” IEEE Transactions on Power Delivery, vol. 24, no. 3, pp.1174–1181, Jul. 2009.

[7] P. Ivan L.G., “Smart grid cyber security for europe,” Energy Policy,vol. 39, no. 9, pp. 5211–5218, Sep. 2011.

[8] J. Deuse, J. Dubois, R. Fanna, and I. Hamza, “EWR under voltage loadshedding scheme,” Power Systems, IEEE Transactions on, vol. 12, no. 4,pp. 1446 –1454, Nov. 1997.

[9] H. Lokay and V. Burtnyk, “Application of underfrequency relays forautomatic load shedding,” Power Apparatus and Systems, IEEE Trans-actions on, vol. PAS-87, no. 3, pp. 776 –783, Mar. 1968.

[10] F. Shokooh, J. Dai, S. Shokooh, J. Tastet, H. Castro, T. Khandelwal, andG. Donner, “Intelligent load shedding,” Industry Applications Magazine,IEEE, vol. 17, no. 2, pp. 44 –53, Apr. 2011.

[11] S. Majumdar, D. Chattopadhyay, and J. Parikh, “Interruptible loadmanagement using optimal power flow analysis,” Power Systems, IEEETransactions on, vol. 11, no. 2, pp. 715 –720, May 1996.

[12] R. Faranda, A. Pievatolo, and E. Tironi, “Load shedding: A newproposal,” Power Systems, IEEE Transactions on, vol. 22, no. 4, pp.2086 –2093, Nov. 2007.

[13] T. S. Fernandes, J. R. Lenzi, and M. A. Mikilita, “Load sheddingstrategies using optimal load flow with relaxation of restrictions,” IEEETransactions on Power Systems, vol. 23, no. 2, pp. 712–718, May 2008.

[14] R. Baldick, B. Chowdhury, I. Dobson, Z. Dong, B. Gou, D. Hawkins,H. Huang, M. Joung, D. Kirschen, F. Li, J. Li, Z. Li, C.-C. Liu, L. Mili,S. Miller, R. Podmore, K. Schneider, K. Sun, D. Wang, Z. Wu, P. Zhang,W. Zhang, and X. Zhang, “Initial review of methods for cascadingfailure analysis in electric power transmission systems IEEE PES CAMStask force on understanding, prediction, mitigation and restoration ofcascading failures,” in 2008 IEEE Power and Energy Society GeneralMeeting - Conversion and Delivery of Electrical Energy in the 21stCentury. IEEE, Jul. 2008, pp. 1–8.

[15] I. Dobson, B. A. Carreras, V. E. Lynch, and D. E. Newman, “An initialmodel fo complex dynamics in electric power system blackouts,” inProceedings of the 34th Annual Hawaii International Conference onSystem Sciences, 2001. IEEE, Jan. 2001, pp. 710– 718.

[16] I. Dobson, K. R. Wierzbicki, J. Kim, and H. Ren, “Towards quantifyingcascading blackout risk,” in Bulk Power System Dynamics and Control -VII. Revitalizing Operational Reliability, 2007 iREP Symposium. IEEE,Aug. 2007, pp. 1–12.