Artificial Intelligence and Optimization Techniques for Intelligent Power Systems: Fault Detection, by IRJET Journal

Artificial Intelligence and Optimization Techniques for Intelligent Power Systems: Fault Detection, Energy Management, and Grid Stability

1,2Assistant Professor, Baddi University of Emerging Sciences & Technology Makhnumajra, Baddi, Distt. Solan, H.P.173205, India.

ABSTRACT

There is an increasing need for intelligent, adaptive and resilient control strategies under the new paradigm of modern power systems decentralization, sustainability and digitalization. Simple rules and static optimization of traditional, centralizedmodelsarenotsufficienttomanagegrowing amountsofrenewableenergysources(RES),distributedenergy resources (DERs) and bi directional energy flows. To this challenge, the application of Artificial Intelligence (AI) and advancedoptimizationtechniqueshavebeenusedaspowerfultoolsforincreasingthelevelofrealtimemonitoring,fault detection,energyforecastingandsystemstability.Inthisreview,wesynthesizetheapplicationofthisrangeofAI-driven models (machine learning, deep learning, reinforcement learning and hybrids) to key power system functions. For example, fault detection speeds of under 5 ms with 98.5% accuracy have been demonstrated, using wavelet transform integratedwithsupportvectormachines,formultiterminalDC(MTDC)grids.Applyingartificialneuralnetworks(ANNs), theclassificationaccuracywas97.8%andlocalizationerrorwasbelow5%.CNNs,usingdeeplearningbasedonCNNsalso went beyond 99% accuracy in the fault diagnosis of underground cables. Recurrent neural networks (RNNs) and long shorttermmemory(LSTM)modelsimprovedtheshorttermloadforecastingaccuracyupto15%andANN-PSOalgorithms improved the V2G scheduling with minimum battery wear and peak shaving. The paper also recognizes critical research gaps(e.g.,explainability,dataprivacy,edgedeployment,cybersecurity)thatneedtobeaddressedforrobustandimpactful deployment of ML. Integrating these intelligent systems into power networks allows utilities to move closer to that of resilient,efficientandsustainableenergyinfrastructures.Withthisreview,researchersandengineerscanhaveabasisto leadfuturedevelopmentsinsmartgridintelligenceandenergyoptimization.

Keywords: Smart Grids, Artificial Intelligence, Fault Detection, Optimization, Renewable Energy, Energy Management, MachineLearning,DeepLearning,V2G,LoadForecasting.

Because of the move to sustainable and resilient energy systems, designing, operating and managing today’s power networkshasfacedvariousnewdifficultiesandbenefits.Conventionalenergysystemswhichworkedbyasingle planand at a central place, are now moving towards being complex, digital and more flexible smart grids. They have significant sharesofRES,involvepowergenerationthatisfarfromcentralized,two-wayenergyflowandahighuseofEVsanddata technology. Because of their importance, AI and optimization techniques help these systems better manage control, run efficientlyandadapt,dependonandgivebacktotheenvironment.

1.1 Background: The Evolution of Power SystemsHistorically, power was generated at few central locations and moved onlyonewaytouserswhoconsumedit.Mostsystemsusedrulestomanagepowerwhichgavepredictableresultsunder regularloadconditions.Therehasbeenariseindemandforcleanenergyandreducingcarbonintheenergysegment,so solar PV, wind turbines and small-scale hydroelectric energy are now being integrated. Now, the grid works as an interactive system, linking several stakeholders, solar and wind generators, storage systems and any entities that create energy.Onthetechnicalside,complexsensors,devicesusingtheInternetofThings(IoT),advancedmeters(AMI)andnew communicationmethodshavebroughtustheSmartGridwhichcanmonitoritself,collectdataandmakeitsowndecisions. But these developments bring serious problems to the management of the grid, mostly related to stability, coordination, securityandefficiency.HandlingtheunpredictableanderraticqualitiesofmodernpowersystemsnowdependsonAIand optimizationtools..

2. AI Applications in Power System Monitoring and Fault Detection

Previously,electricityflowedfromcentralpowerplantsdowntoconsumersalongaonedirectionalprocess.Mostcontrol systems, working under continuous load, operated according to set rules to direct actions. Now, however, people want clean energy and want to reduce the carbon footprint of the energy industry and so more solar PV, wind turbines and

Er. Aman Choudhary1* , Er Ankush Pathania 2

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

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

small hydroelectric systems are being added. So the grid is much more interactive now with many stakeholders, distributed generators, energy storage systems and prosumers among others. Sensors, IoT devices, advanced metering infrastructure (AMI) and communication protocols have proliferated over the years and birthed the Smart Grid which makesitselfawareofwhat’sgoingonandmakesdecisionsitselfbasedonthedataitcollects.Althoughthesechangesbring majorchallengesfortheadministrationofthegrid,forstability,coordination,securityandefficiencyatthesametime.The behaviorofpresentday powersystemsiscomplex,uncertainandinconsistent,thususeofAIandoptimizationmodelsis thecalloftheday.

2.1 AI-Based Fault Detection Techniques

Let’s delve into specific AI models and how they are applied in fault detection across different systems, including MTDC, overhead/undergroundlines,andsmartsubstations.

2.1.1 Wavelet Transform & Machine Learning for MTDC Fault Detection

Sabug et al. (2020) introduced a Boundary Wavelet Transform (BWT)-based method for real-time fault detection in Multi-Terminal DC (MTDC) grids. The wavelet transform decomposes transient signals into frequency components, isolatingfaultsfromnormalswitchingevents.

Themathematicalrepresentationofawavelettransformis:

Where:

 :signalunderanalysis

 :motherwavelet

 :scalingparameter(frequency)

 :translationparameter(time)

Sabug'sstudyshowedthatwhenintegratedwitha Support Vector Machine (SVM) classifier,thesystemachieved:

 Detectiontime: <5 ms

 Accuracy: >98% across10testcases

 Resiliencetonoiseandloadfluctuations

2.1.2 Artificial Neural Networks in Overhead Line Protection

Elmitwallyetal.(2020)proposedamodelusing feedforward artificial neural networks (ANNs) trainedoncurrentand voltagewaveformdata.TheANNoutputisusedtodeterminethe type and location offaults.

ThebasicoperationofanANNforfaultdetectionisdescribedby:

Where:

 :inputfeatures(e.g.,current,voltagesamples)

 :connectionweights

 :biasterm

 :activationfunction(e.g.,sigmoid,ReLU)

 :output(faultclass)

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

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

TheANNwastrainedusingabackpropagationalgorithm,achieving:

 Faultclassificationaccuracy: 97.8%

 Locationestimationerror: <5%

 Robustnessagainstvariationinfaultresistanceandinceptionangle

2.1.3

Deep Learning for Underground Cables

For underground cable networks, which are difficult to inspect manually, Deep Learning models like Convolutional Neural Networks (CNNs) canclassifytime-serieswaveformdatawithoutmanualfeatureextraction.

CNNsaregovernedbytheconvolutionoperation:

Where:

 :inputsignal(voltage/currenttime-series)

 :convolutionkernel

 :filteredoutput

Thesesystemsdemonstratedaccuracyof >99% onlabeledtrainingdatasetsindetecting:

 Singleline-to-groundfaults

 Line-to-linefaults

 Cabledegradationevents

2.2

Performance Comparison of AI Techniques

Tobetterunderstandthetrade-offs,thefollowingtablecomparesmajorAIapproachesusedinfaultdetection.

Table 1. Comparison of AI Methods for Fault Detection in Power Systems

Overheadlines,radialfeeders

Gridoptimizationandfaultclassification

3. AI and Optimization for Energy Management and Stability

Theriseofdistributedenergyresources(DERs),renewablepowersources,two-waypowerflowandresponsiveloadshas made managing energy and stability in power grids much more difficult, requiring intelligent solutions. The old ways based on linear programming or heuristics do not work well enough now to cope with the demands of real-time energy needs and the unpredictability of renewables. Instead, using Artificial Intelligence (AI) and hybrid optimization techniques, it is now easy to develop responsive, leading and distributed methods for controlling energy systems.AIpoweredenergysystemshelptopredictenergyusage,arrangeallresources,runandmaintainbatteries,unitemicrogrids andregulatedynamismofenergynetworksatalllevels.ThissectionreviewshowAIandoptimizationplayvariousrolesin makingsureenergyisusedandmanagedproperly.

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

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

3.1 Smart Grids and Stability Enhancement Using AI

Smart grids incorporate advanced metering infrastructure (AMI), communication layers, and automated control devices thatmonitor,collect,andactonenergy-relateddatainrealtime.Acorerequirementinsuchsystemsisthe maintenance of voltage and frequency stability,especiallywhenoperatingwithalargeshareofrenewables.

AIplaysacriticalroleingridstabilityby:

 Predictingpotentialinstabilityeventsbasedonhistoricalpatterns.

 Executingcorrectiveactionsusingadaptivelearningmodels.

 Reducingrelianceontraditionalcontrolcentersviadecentralization.

In their study, Alaerjan et al. (2024) merged various AI tools such as fuzzy inference systems, neural networks and Bayesian techniques to make smart grid operations more stable. The application deals with fresh data from operations, knowtheweathertheyareunderinthefutureandseesuserstoidentifyrisksandprovidetipsonhowtotakeaction.Shi et al. (2020) therefore described in detail how AI could be used in stability studies of the grid. In his paper he looked at howreinforcementlearning(RL)isimplementedinsecondaryandtertiaryvoltagecontrol.TheAIsystemsweredesigned toreactdynamically(andautomatically)tochangesinthegridbyadjustingtheirreactivepowersupportfromcapacitors and inverters and, as a result of that, the number of under and over voltage events in the tested cases was dramatically reduced.Unliketraditionalsystems,AIsystemsusedata(whichoftenhasflaws,errors,delaysetc)andcantoleratethem.

3.2 Adaptive and Real-Time Energy Management

Energymanagementdealswithoptimizationofgeneration,storageandloadinordertomatchthedemand economically and in an environmentally friendly way. Adaptive energy management system using big data analytics and machine learningalgorithmsforsmartgridisproposedbyGupta&Chaturvedi(2023)forthispurpose,includingloadforecasting, generationscheduling,storagemanagementanddemandresponse(DR).Therealtimedatastreamsgeneratedfromsmart metersareprocessedthroughtheirsystemwhereitdetectsloadpatternsandclustersconsumersforefficientdistributed generation..

ThekeycapabilitiesofMLinenergymanagementinclude:

 Short-term load forecasting (STLF) using Recurrent Neural Networks (RNNs) or Long Short-Term Memory (LSTM)networks.

 Energy price prediction tofacilitatemarketparticipation.

 Consumer classification and profiling tomanagedemandresponseprograms.

The mathematical basis for predictive energy modeling in neural networks often revolves around minimizing the mean squarederrorbetweenpredicted ̂ andactualvalues :

oreover, tecyk & iciuła (2023) introduced a collaborative energy optimization platform, where AI agents work togetherina multi-agent system (MAS) tocoordinate DERsacross regions.These agentsautonomouslymanageenergy flows based on local objectives (e.g., battery charge state, solar availability) while contributing to global grid objectives (e.g.,peakshaving,frequencyregulation).

3.3 AI and Optimization for Renewable Integration

Renewableenergyintegrationposesamajorchallengeforgridoperatorsduetointermittencyandlimiteddispatchability. AImodels,especiallydeeplearningones,canpredict:

 Solar irradiance

 Wind speed

 Cloud coverage

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

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

 Energy yield variability

Danish (2023) demonstrated that hybrid AI approaches significantly improved renewable forecasting accuracy, reducingcurtailmentandreserverequirements.Forexample,CNNmodelsweretrainedonhistoricalsatelliteimages and weathertimeseriestopredictsolargeneration30minutesahead,outperformingphysicalmodelsby12–15%inRMSE.

In addition to forecasting, optimization is used for dispatch planning. Nouri et al. (2024) proposed an ANN-PSO algorithm tomanage Vehicle-to-Grid (V2G) interactions.Thesystempredictsgridneedsandbatterystates,optimizing whenelectricvehiclesshouldchargeordischarge.

Theirobjectivefunctionminimizesthecostofpowerwhileensuringbatteryconstraints:

Where:

 buy sell:Powerbought/soldattime

 :Marketpricesforbuying/selling

 :Stateofchargeofthebattery

ThisoptimizationensuredthatV2Goperationsdidnotdegrade batterylifewhilesupportingpeakshavingandfrequency regulation.

3.4 Energy Forecasting and Demand Response Optimization

ForecastingdemandandimplementingeffectiveDRmechanismsarekeytobalancinggenerationandloadinrealtime.AI enables:

 Customer-specificdemandmodeling.

 Probabilisticpeakloadprediction.

 Incentive-basedresponsetriggering.

Fengetal.(2021)describedapplicationsof clustering algorithms (likeK-means)and support vector regression (SVR) to segment consumers and forecast load. This allowed operators to target the right customers for DR signals, increasing participationratesbyover20%.Additionally,AImodelswereusedtodynamicallycompute Time-of-Use (TOU) tariffsand automateDRprogramadjustmentsbasedonbehaviorandpreferences.

Table 2. AI and Optimization Applications in Energy Management and Stability

Domain Technique Used Goal

Key References

GridStability Fusion AI, RL, Fuzzy Systems Maintainvoltage/frequency Alaerjanetal.(2024);Shietal.(2020)

LoadForecasting LSTM,RNN,SVR Predictshort-termdemand Gupta & Chaturvedi (2023); Feng et al. (2021)

Renewable Forecasting CNN,BayesianModels Predictsolar/windgeneration Danish(2023);Mortier(2020)

Energy Trading & Pricing DeepQ-Learning,MAS Real-time pricing and bidding optimization tecyk& iciuła(2023)

V2GManagement ANN+PSO OptimizeEVcharge/discharge Nourietal.(2024)

DemandResponse Clustering+Regression Targeted load shedding and incentive design Fengetal.(2021)

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

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

4. Challenges, Research Gaps, and Future Directions

Table 3. Summary of Challenges, Gaps, and Future Directions

Category Current Challenge

Data&Training Lackoflabeled,diversedata

Model Transferability

Explainability

Poor generalization across regions

Black-boxmodelsinhibittrust

Cybersecurity Vulnerability to data poisoning, adversarialattacks

SystemIntegration Difficulty embedding AI in SCADA/EMS

EdgeIntelligence Highcomputationaloverhead

Human-AI Collaboration

4. Conclusion

Research Gap

Underdeveloped unsupervised/selfsupervisedlearning

Inflexible and topology-specific models

LackofexplainablepowersystemAI

FewAI-specificgridthreatmodels

Limitedindustrialcompatibility

Lack of lightweight algorithms for edgedevices

Operatordistrustandskillgaps No interfaces for human–AI decision co-analysis

Future Direction

Federated learning, synthetic data generation

Transfer learning with domain adaptation

XAItailoredforenergyoperators

Resilient, secure AI + adversarial training

Development of API-friendly, lightweightmodels

TinyML and real-time edge-AI models

Operator-centric dashboards and interactiveXAItools

With modern power systems progressing towards greater sustainability, flexibility and intelligence, this integration of Artificial Intelligence (AI) and optimization has become not only advantageous, but necessary. High levelsofalgorithmicstochasticity oruncertaintymakeitimpossibletomanageinrealtimewithtraditionaldeterministic approaches such as those that are deployed and still in use today the increasing complexity resulting from distributed generation, renewable integration, two‐way energy flow and dynamic consumer participation. Through this reviewithasbecomeapparentthatAIisplayingamajorroleinfaultdetection,increasingstabilityandallowingrealtime adaptiveenergymanagement.Neuralnetworks,deeplearning,reinforcementlearningandhybridalgorithmsareblending withoptimizationtechniqueslikePSOtoenablegridstobemorepredictive,responsiveandefficient.Nevertheless,some issuescontinuetoexistsuchasdatapaucity,modelinterpretability,cybersecuritysecurityrisksandinteroperabilitywith antiquated systems. To remedy these problems we need to dedicate ourselves to research areas like edge intelligence, federatedlearning,explainableAIandsecurecontrolarchitectures.Thefutureenergyinfrastructuresmustbeintelligent, resilient,interpretableandcollaborative.MultidisciplinaryinnovationcanbridgethesegapsandensurerobustnessofAI powered energy systems seeking to meet tomorrow's demands but retain reliability, transparency and trust. While the vision of a decentralized, sustainable and intelligent energy ecosystem has historically been theoretical, the historic inroadsmadewithAIatitscoreandcontinuingtodaybringitwithintheproximityofreality.

References

1. Sabug, L., Jr., Musa, A., Costa, F., & Monti, A. (2020). Real-Time Boundary Wavelet Transform-Based DC Fault Protection System for MTDC Grids. International Journal of Electrical Power & Energy Systems, 115, 105475. https://doi.org/10.1016/j.ijepes.2019.105475

2. Elmitwally, A., Mahmoud, S., & Abdel-Rahman, M. (2020). Fault Detection and Identification of Three-Phase OverheadTransmissionLines. Mansoura Engineering Journal, 35,129–137.[CrossRef]

3. Onwusinkwue,S.,Osasona,F.,Ahmad,I.A.I.,Anyanwu,A.C.,Dawodu,S.O.,Obi,O.C.,&Hamdan,A.(2024).Artificial Intelligence (AI) in Renewable Energy: A Review of Predictive Maintenance and Energy Optimization. World Journal of Advanced Research and Reviews, 21,2487–2499.https://doi.org/10.30574/wjarr.2024.21.1.0934

4. Alaerjan,A.,Jabeur,R.,Chikha,H.B.,Karray,M.,&Ksantini,M.(2024).ImprovementofSmartGridStabilityBased onArtificialIntelligencewithFusionMethods. Symmetry, 16,459.https://doi.org/10.3390/sym16030459

5. Yaseen, A. (2023). AI-Driven Threat Detection and Response: A Paradigm Shift in Cybersecurity. International Journal of Information and Cybersecurity, 7,25–43.[CrossRef]

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

Volume: 12 Issue: 06 | Jun 2025 www.irjet.net p-ISSN:2395-0072

6. Baduge,S.K.,Thilakarathna,S.,Perera,J.S.,Arashpour,M.,Sharafi,P.,Teodosio,B.,Shringi,A.,&Mendis,P.(2022). Artificial Intelligence and Smart Vision for Building and Construction 4.0: Machine and Deep Learning Methods andApplications. Automation in Construction, 141,104440.https://doi.org/10.1016/j.autcon.2022.104440

7. tecyk, A., & iciuła, I. (2023). Harnessing the Power of Artificial Intelligence for Collaborative Energy OptimizationPlatforms. Energies, 16,5210.https://doi.org/10.3390/en16145210

8. Gupta,R.,&Chaturvedi,K.T.(2023).AdaptiveEnergyManagementofBigDataAnalyticsinSmartGrids. Energies, 16,6016.https://doi.org/10.3390/en16166016

9. Nouri, A., Lachheb, A., & El Amraoui, L. (2024). Optimizing Efficiency of Vehicle-to-Grid System with Intelligent ManagementandANN-PSOAlgorithmforBatteryElectricVehicles. Electric Power Systems Research, 226,109936. https://doi.org/10.1016/j.epsr.2023.109936

10. Dobbe, R., Hidalgo-Gonzalez, P., Karagiannopoulos, S., Henriquez-Auba, R., Hug, G., Callaway, D.S., & Tomlin, C. (2020). Learning to Control in Power Systems: Design and Analysis Guidelines for Concrete Safety Problems. Electric Power Systems Research, 189,106615.https://doi.org/10.1016/j.epsr.2020.106615

11. Danish, M.S.S. (2023). AI in Energy: Overcoming Unforeseen Obstacles. AI, 4, 406–425. https://doi.org/10.3390/ai4030024

12. Mortier,T.(2020).WhyArtificialIntelligenceIsaGame-ChangerforRenewableEnergy. Ernst & Young Global Ltd. https://assets.ey.com/content/dam/ey-sites/ey-com/en_gl/topics/technology/ey-ai-in-energy-renewablesreport.pdf

13. Nix, A., Decker, S., & Wolf, C. (2021). Enron and the California Energy Crisis: The Role of Networks in Enabling OrganizationalCorruption. Business History Review, 95,765–802.https://doi.org/10.1017/S0007680521000536

14. Szczepaniuk,H.,&Szczepaniuk,E.K.(2022).ApplicationsofArtificialIntelligenceAlgorithmsintheEnergySector. Energies, 16,347.https://doi.org/10.3390/en16010347

15. Reda, A., Al Kurdi, I., Noun, Z., Koubyssi, A., Arnaout, M., & Rammal, R. (2021). Online Detection of Faults in TransmissionLines.In Proceedings of the 2021 IEEE 3rd International Multidisciplinary Conference on Engineering Technology (IMCET), Beirut, Lebanon, 8–10 December 2021. https://doi.org/10.1109/IMCET51167.2021.9689800

16. Rong, X., Shek, J.K.H., Macpherson, D.E., & Mawby, P. (2021). The Effects of Filter Capacitors on Cable Ripple at Different Sections of the Wind Farm Based Multi-Terminal DC System. Energies, 14, 7000. https://doi.org/10.3390/en14217000

17. Feng, C., Sun, M., Dabbaghjamanesh, M., Liu, Y., & Zhang, J. (2021). Advanced Machine Learning Applications to Modern Power Systems. In New Technologies for Power System Operation and Analysis, Elsevier, pp. 209–257. https://doi.org/10.1016/B978-0-12-820169-7.00016-7