Fault Prediction and Interdependencies Identification in Smart Grids Using Deep Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 09 | Sep 2022 www.irjet.net p-ISSN: 2395-0072

Fault Prediction and Interdependencies Identification in Smart Grids Using Deep Learning

***

Abstract - Smart Grid is an modernisation of current Electrical Grid which uses both communication and Information Technology to make electricity transfer more efficient. The smart grid is an interconnection of many communication devices and electrical components hence it makes an interdependent network . As the smart grid is interconnected, small changes or attacks on the network can lead to a cascade of failures across the networks which can lead to grid failures or power blockouts. Hence it is necessary to identify the faults and interdependencies among communication systems and electrical networks. This type of networks can be referred to as cyber-physical systems. This paper proposes an Artificial Neural Network based model that can detect the grid failures and interdependencies of grid components. This predictive model can help systems operators in Smart grid to make necessary preventive actions and mitigate the attacks or failures from time to time. As a case study to perform analysis we have usedIEEEpower bus system, namelyIEEE57 which represents AmericanPower system. Andthe model achieves an accuracy of 99.92% on simple data and also 99.19% on complex data in predicting grid failure sequences.

Key Words: failure prediction, Artificial Neural Network, deep learning, Multi-class classification, Smart grid, Interdependencies Issues.

1.INTRODUCTION

Smart grids are cyber-physical systems which means that it consists of sensors, actuators, communication networks and protocols. The smart grids are different from conventionalgridduetoitstwowaycommunicationandit isaresilientsystemthatcanhealitselfduringanyfailures in the network. The communication systems are responsibleforthis.Thecommunicationnetworksconnect each component to achieve communication between variousphysicalsystemsofthegrid.[1]Thecyber-physical systems are found everywhere like Industrial control process, Power systems, etc. The communication systems in the smart grid have enabled high dependability standards to achieve high accuracy. Nowadays, due to population growth, the electricity demand is also increased, hence it is extremely vulnerable to any attacks and can cause cascading failure attacks [1], [2], [3].Cascade failures is the failure of one or more

components which in turn can cause failure to all other components in network. This vulnerability can result in devastatingconsequencesifanyinitialcomponentfailures arecriticalones.

Considering the smart grid as graphical representation. In smart grid, each component (node) has load and capacity [3], [5] . Due to cascading failure in smart grid, evensingle failureatsingle node[6]canlead to failure of itsneighborcomponents(nodes)duetochangeinbalance powerflow.Thiscanleadtoglobalredistributionofloads in entire system components causing overloads and powershutdownofgrids.

There are few large scale networks that are affected due to damaged components in the network or due to connected systems. In 2011 Arizona-southern blackout is an example for cascaded failures which caused an 11 minutes power outage in transformer causing 2.7 million people to be left out with power outage nearly up to 12 hours [7]. Italy blackout in 2003 due to connection issues withtheinternalarchitecturesofthenetwork[8].Thereis an most recent cyber attacks on power systems is, attack on the Ukraine power grid in December 2015, which is a synchronized and coordinated cyberattack, causing a 6-h blackout and affecting hundreds of thousands of customers[ 9 ].So, it is evident that these are very dangerous issues that can affect the economy and also causes physical damage to the systems. So, it is necessary todetectthefailuresandissuesduetointerdependency.

Smart Grids bring better capabilities and improvements to the present Power system. This makes systemsmorecomplexandvulnerabletodifferenttypesof attacks and even unintentional failures due to increased complexity of systems and levels of cyber fragility. Hence, there are more security issues that need to be addressed. In smart grid these happen because of communication networks.

This paper provides a better understanding of interdependencies of network components and also fault detection in smart grids. To illustrate this work we used the IEEE-57 bus system and it is simulated on various faults. Thedata iscollectedforvariouscomponentsofthe grid.Thedatacollectionanddescriptionwillbeexplained in detail below. To predict the faults and identify interdependencies we used the Artificial Neural Network

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 09 | Sep 2022 www.irjet.net p-ISSN: 2395-0072

to observe the underlying pattern of the data and analyze interdependencies so that it can be used to find the cascading components and also detect the failures in the grid due to communication networks . The dependency canbefoundwiththehelpofcorrelationfactors.Thereare variouscorrelationmethodsthatcanhelptofindrelations between the components in the grid. And also this predictivetoolcanbeusedbysmartgridoperators,Smart gridsecuritytoolsprovider,Smartgridmanufacturersand vendors to analyze the attacks and provide counter measuresfrom timetotime. TheExperiments and results shows that this method can effectively detect failure sequencesingrid.

2.EFFECT OF CYBER ATTACKS ON SMART GRID

Due to advancements in smart grid technology there is a chance for cyber attacks, which deeply affect power industries. However the security in industries is a new topic and still experts are learning and trying to make securitytoolsandfindthevulnerabilities.

Vulnerabilities occur when hackers try to change the original values of devices in grids .There are many vulnerabilitieswhichcanaffecttheconfigurationofsmart grids.[ ] There are few vulnerabilities that possess a seriousthreattosmartgrids.

1. Smart grid has many communication devices that can help in controlling and monitoring the smart grid remotely. And Intelligent Electronic devices (IEDs) which handle many devices to manage both electricity supply and demand of load. So, a small update or modification of configuration can lead to great economic losses and also the operators need to travel to remote locations to identifythecauseoftheproblem.

2. IEDsinsmartgridsprocesslargeamountsofdata and hence preserving the information is one of the important tasks. Disclosure or unauthorized access to data or information can lead to major problems that can’t be controlled after their occurrence.

3. As we use communication networks, small outages in the network can lead to power blockoutandcausemanyeconomicalissues.

4. FDI attacks can be possible due to vulnerable usersatthecustomersendwhointendtostealor enter the network to access back-doors and hack information and also modify the devices configurationwhichinturncancauseacascadeof failuresinSmartgrid.

Hence we need to provide more security measures compared to conventional power systems. In this paper

we will use a computational method to resolve this problem.

3.INTERDEPENDENCIES IN SMART GRID

This section deals with the interdependencies analysis of components in cyber-physical systems. We have various communication protocols in smart grid to interconnect various physical devices like substations, distribution grids and transmission grids. And some of these communications like Local Area Network (LAN), Home Area Network (HAN),etc are provided by Internet service providers (ISP’s) and these service providers might be shared with multiple companies which can lead to unethical information access, as this data will be shared across multiple communication devices which makes accessingofthisdatafeasibleandhenceanycyberattacks can take place. Where, a simple attack on transmission linesorfalsedatainjectionorfailureofasmartmeter,etc canaffect entireload stabilityof grid,and resulting a grid failure. As the smart grid is an interdependent network failure of one component in grid results in cascading failures.Hencepropersecuritymeasuresmustbetaken.

Interdependence of components is the Influence of one state component on the other state components and vice versa.Theidentificationofinterdependenciesdepends on systemsbehavior duringfailureofsystem.Thedisruption insystemduringfailurecancausefailuresequencesinthe grid from which we can find the dependencies between components. As, the interdependency of the communication network and power network shows the linkage of these two infrastructures. Figure 1 shows the power and information flow between the Power system network and communication network. The interdependencies in any cyber-physical systems can be classified into mainly four types : logical, Physical, cyber, geographic [10 ]. As, we are discussing about the interdependency relationship between Power system network and communication network they can described asbelow,

I. Communication networks have physical dependence on power system networks as it needs power supply to perform its data transmissionfunctionalities.

II. The power system has cyber dependency on communication because, as a type of cyberphysical system (CPS), control of the power grid relies on the latter to deliver the monitoring data and control messages between the control entity andthefielddevices.

III. Geographical interdependence between the two infrastructures, since the transmission line and optical fiber are usually located close to each other in the power transmission network, while

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 09 | Sep 2022 www.irjet.net p-ISSN: 2395-0072

theutilitypolesoftencarrybothdistributionlines andcommunicationequipmentonthem.

Identification of Interdependencies and Prediction of Fault Propagation for Cyber-Physical Systems

The study authors Koosha Marashia , Sahra Sedigh Sarvestanib,AliR.Hursonbuseddeeplearningtopredict the failure of components of a grid using an Artificial Neural Network and also their work provided an analysis of interdependencies in smart grid that causes a cascade offailures.

5.PROPOSED WORK

In this section we will discuss the proposed work to identifyfaultsanddetectinterdependenciesinsmartgrid.

5.1 Simulationstocollectdata

Figure 1 InterconnectionofCommunicationNetworkand PowersystemNetwork

4.RELATED WORK

Anomaly Detection in Smart Grids using Machine Learning Techniques

The study paper authors Manikant Patni examined how the anomalies occur in smart grids. He discussed the detection of faults, cyber-attacks , etc. His work aims to detectcyberattacksinsmartgridusingMachineLearning techniques and for evaluating and experimenting he used IEEE-3bussystem.

Evaluating Anomaly Detection Algorithms through different Grid scenarios using k-Nearest Neighbor, iforest and Local Outlier Factor

Thestudypaperauthors NilsJakobJohannesen;Mohan Lal Kolhe; Morten Goodwin used K-Nearest Neighbors to study the anomalies in smart grid and predict the cyberattacksongrid.

A Systematic Literature Review of Machine Learning Approaches for Detecting Events and Disturbances in Smart Grid Systems

The study paper authors Ricardo Buettner; Johannes Breitenbach; Jan Gross; Isabell Krueger; Hari Gouromichos; Marvin Listl; Louis Leich;Thorsten Klier proposed an literature of different Machine Learning algorithms to detect abnormal events occurring in grid andalsoanomalieslikecyber-threatsandFDIattacks.

To demonstrate the proposed work we used the IEEE-57 Bus system. Figure shows the single line diagram of the IEEE-57 bus system. As we know that classic IEEE bus systemdon'tuseanycommunicationnetworkbutconsists of generators, Transformers, etc. But our study is on cyber-physical infrastructure hence, we add communication network to the classic IEEE system to make an equivalent smart grid. This communication networkconsistsofSCADA,PMU’s,FACTSdevices. FACTS devices help in power flow in transmission lines and we used methods [11] to find the location of PMU’s and FACTSdevicesingrid.

WehavedesignedtheIEEEbussystemtestcaseasper our needs , now we need to perform the required simulationforfailurecasesofthegrid.Aswearelearning the interdependencies of components it is important to find the number of failure sequences/cases needed to obtain all dependencies.In thecase of predictive models, a large amount of data can give accurate results. Studies present in [12] describe a method to select required failurecasesprovidingmaximumaccuracy.[10]described thefollowingscenarios,

1. twosimultaneoustransmissionlineoutages.

2. atmostoneFACTSdevicefailed.

3. atmostonePMUfailed,failureofdecisionsystem.

The selection of failure cases are taken based on the real worldfailurerateofcomponentsinthegrid.FACTS,PMU’s failure cases are considered because they are prone to cyberattackssoasmallattackonthesecancausefailures in cascaded systems. And transmission lines are also considered as they are a major part in power systems in whichfaultsareusual

5.2 IdentificationofInterdependenciesandanalysis

Interdependencyisthestateofanentitythatinfluencesor is correlated with the state of another, and vice versa[10

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 09 | Sep 2022 www.irjet.net p-ISSN: 2395-0072

].In order to represent interdependencies in any cyber physical systems we use [10 ] dependency weighted directedgraphinwhichedgeexistsfromnodeitonodejif and only if the state of component i impacts, in one time step, the state of component j .There are four types of dependency based on the source and destination of the edge.ThefairtypesareasshowninFigure 2.

WeusedANNtopredictfailures,andthedatacollected is splitted to train and test in the ratio of 80:10:10 and 80% of data is used for training and 10% for testing and another 10% for cross validation. After training and testing we use the created model for evaluating its performance. The ANN architecture implemented in this researchisshownintheFigure3below. Whereithasone Inputlayer,oneOutputlayer,andfivehiddenlayers.

Figure 3 ArtificialNeuralNetwork

5.4 TrainingandTesting

Figure 2 Interdependenciesincyber-physicalsystems

InFigureabove thebottomplaneconsistscomponentsof physical (in our case power system network) and upper plane consists of cyber infrastructure (in our case communication infrastructure).This cyber infrastructure controls and monitors the physical infrastructure. As, this is an weighted graph the values on edges indicates the weightofeachedgerangingfrom0to1,where0indicates no influence of one component on another and close to 1 indicates the maximum interdependence of one component on other. To represent this in mathematical form we introduce a adjacency matrix of this directed graph, D = [d .Where dᵢⱼ represents the influence/dependence of component i on component j andnisthenumberofcomponents.Thisinterdependency analysis helps in understanding the importance of components in the cyber-physical infrastructure. And the weighted graph representation can help us in describing theamountofweighteach edgeorcomponentexhibitson itsneighborhoodcomponents.

5.3 PredictingFaults

We have collected the required amount of failure data to predict failures. Now we should pre-process that failure data into a proper dataset to perform modeling and get insights from that dataset. We have created two dataset samples one is simple test data and the other is complex testdatasothedifferencebetweenthetwois,incomplex test data we have simulated the bus system for more advancedfailurecasessuchasfailureof5to7concurrent transmissionlinefailuresandtwotothreePMUarefailed and at least two FACTS devices are failed. But in simple testdataweuseddataasdiscussedinthesection5.1.

AfterdevelopingtheANNmodelnowweneedtotrainthe data collected and evaluate the performance of the algorithm. As we have discussed above about the test cases considered now we need to see the performance of the model on both these cases i.e simple test case and complextestcases.Table1showstheevaluationresultsof themodelontestdatasets.

Table 1 NeuralNetworkPerformanceonTwodatasets

Test case Accuracy Recall Precision F-score

SimpleTest data 9992% 9747% 9900% 9796%

Complex testdata 9919% 7215% 8527% 7573%

From above table it is proven that the model can detect thefailuresingridforsimpletestdatabutthemodelisnot greatenoughforcomplexdatabutthisisbecausewehave trained our model only for the simple failure data but the model is enough good to predict the complex failures in smart grid with an accuracy of 99.19%. The training and testingperformanceofthemodelisshowninFigure4.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056

Volume: 09 Issue: 09 | Sep 2022 www.irjet.net p-ISSN: 2395-0072

So,theArtificialNeuralNetworkhasachievedanaccuracy of 99.92% on simple test cases and 99.19% on complex testcases.

7.CONCLUSION

We proposedan effective DeepLearningmodel to predict the failures of smart grid and also analyzed the components relationships . This can help to identify the failures in smart grid immediately so that immediate countermeasures can be taken. And also the cyberphysical systems are more vulnerable to cyber attacks hence using these models can be faster than conventional methodsandalsocanhelptointerprettheproblem.

8.REFERENCES

Figure

4 Trainingandtestinglossforlargeepochs

So,wecanseethatasthenumberofepochsincreasesthe error decreases which is enough to say that the model createdisscalable.

6.RESULT AND DISCUSSION

We have used an Artificial Neural Network to predict the failures in components of smart grid and also the architecture of this ANN is shown in the Figure . The architecture is a seven layered architecture in which it contains one input and five hidden layers along with an output layer. In classification tasks we have few performancemetricsthatneedtobestudiedtoknowhow well the algorithm performs in real time. Figure 5 shows the performance of ANN model along with several performancemetrics.

[1] J. Yan, Y. Tang, H. He, and Y. Sun, “Cascading failure analysis with dc power flow model and transient stability analysis,” IEEE Transactions on Power Systems, vol. 30, pp.285–297,2015.

[2] P. Dey, R. Mehra, F. Kazi, S. Wagh, and N. M. Singh, “Impact of topology on the propagation of cascading failure in power grid,” IEEE Transactions on Smart Grid, vol.7,pp.1970–1978,2016.

[3] V. Rampurkar, P. Pentayya, H. A. Mangalvedekar, and F.Kazi,“Cascadingfailureanalysisforindian powergrid,” IEEE Transactions on Smart Grid, vol. 7, no. 4, pp. 1951–1960,July2016.

[4] P. Dey, R. Mehra, F. Kazi, S. Wagh, and N. M. Singh, “Impact of topology on the propagation of cascading failure in power grid,” IEEE Transactions on Smart Grid, vol.7,no.4,pp.1970–1978,July2016.

[5] S. Paul and Z. Ni, “Vulnerability analysis for simultaneous attack in smart grid security,” in 2017 IEEE PowerEnergySocietyInnovativeSmartGridTechnologies Conference(ISGT),April2017,pp.1–5.

[6] P. D. H. Hines, I. Dobson, and P. Rezaei, “Cascading poweroutagespropagatelocallyinaninfluencegraphthat is not the actual grid topology,” IEEE Transactions on PowerSystems,vol.32,pp.958–967,2017.

[7]https://en.wikipedia.org/wiki/2011_Southwest_blacko ut

[8] A. Berizzi, "The Italian 2003 blackout," IEEE Power Engineering Society General Meeting, 2004., 2004, pp. 1673-1679Vol.2,doi:10.1109/PES.2004.1373159.

Figure 5 ModelPerformanceevaluationondifferent metrics

[9]https://www.power-technology.com/analysis/thefive-worst-cyberattacks-against-the-power-industrysince2014/

[10] Koosha Marashi, Sahra Sedigh Sarvestani, Ali R. Hurson, ”Identification of interdependencies and predictionoffaultpropagationforcyber–physicalsystems “,Reliability Engineering & System Safety,Volume 215,2021,107787.

[11]M.Asprou,E.Kyriakides,OptimalPMU placementfor improving hybrid state estimator accuracy, in: IEEE Trondheim PowerTech, 2011, pp. 1–7. doi:10.1109/PTC.2011.6019247.

[12] J. Qi, K. Sun, S. Mei, An interaction model for simulation and mitigation of cascading failures, IEEE Trans. Power Syst. 30 (2) (2015) 804–819. doi:10.1109/TPWRS.2014.2337284.

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056 Volume: 09 Issue: 09 | Sep 2022 www.irjet.net p-ISSN: 2395-0072 © 2022, IRJET | Impact Factor value: 7.529 | ISO 9001:2008 Certified Journal | Page