Issuu

AI – Driven Digital Twin Models for Sustainable Smart Cities

Mahi Bhalani1, Rudra Donda2, Prof. Reena Desai3

1student, Department of Computer Application, Gyanmanjari Innovative University, Bhavnagar, Gujarat mahiibhalani@gmail.com

2student, Department of Computer Application, Gyanmanjari Innovative University, Bhavnagar, Gujarat rudradonda60@gmail.com

3Assistant professor, Department of Computer Application, Gyanmanjari Innovative University, Bhavnagar, Gujarat radesai@gmiu.edu.in

Abstract - Modern cities are under growing strain as populations shift from rural to urban settings at an unprecedented pace. Managing the resulting complexity in areas like traffic flow, energy demand, structural health of infrastructure, and environmental quality has pushed traditional governance methods to their limits. Over the past decade, smart city programmes have tried to bridge this gap by deploying sensor networks, cloud platforms, and AI-based analytics.Withinthisspace,theconceptofdigitaltwinning building dynamic, data-fed virtual counterparts of physical urban systems has attracted considerable scholarly attention. This paper reviews how digital twin technology, when augmented with artificial intelligence and IoT capabilities, is shaping the next generation of urban management. We examine research published between 2021 and 2025, covering deployments in traffic management, energygridmonitoring,disasterpreparedness,andstructural surveillance. Our synthesis reveals that AI-powered digital twins meaningfully improve prediction accuracy, shorten response times, and support data-driven policy decisions. At the same time, widespread adoption is held back by concerns around cost, fragmented data ecosystems, unclear security boundaries, and a shortage of common standards. The paper concludes by mapping out research avenues including generative AI in simulation enrichment, edge intelligence for latency-sensitive tasks, and federated learning as a privacyrespecting data strategy.

Key Words: Digital Twin, Smart City, Artificial Intelligence, IoT, Urban Infrastructure, Predictive Maintenance, Sustainable Cities, Edge Computing

INTRODUCTION

Urbangrowthhasbecomeoneofthedefiningtrendsof the twenty-first century. According to the United Nations, roughly 56 percent of the world's population resided in citiesasof2023,andprojectionsplacethatfigurecloserto 70 percent by 2050 [1]. While cities generate economic opportunity and cultural exchange, they also concentrate problems:roadnetworksbecomesaturated,energygridsare overtaxed,ageinginfrastructuregoesunmonitored,andair quality degrades. Dealing with these overlapping stresses simultaneously is beyond the reach of conventional

management approaches, which typically react to failures ratherthananticipatingthem.

Thesmartcityparadigmemergedasaresponsetothis complexity. By embedding sensors throughout the urban fabric and connecting them through communication networks, cities began generating real-time data streams thatcould,atleastinprinciple,giveplannersandoperatorsa live picture of what was happening on the ground. Early implementationsfocusedonisolatedusecases adaptive trafficlights,smartmetres,orenvironmentalsensors but lacked any unified framework that could bring these data streamstogetherforholisticdecision-making.

Digital twin technology offers exactly that kind of unifying layer. Rather than treating each sensor or subsysteminisolation,adigitaltwinconstructsacoherent, continuously refreshed model of an urban environment, grounded in real measurements but capable of running forward simulations. The idea traces back to product lifecycle management in manufacturing, where engineers built virtual copies of physical components to test them beforecommittingtoproduction.Urbanresearchersadapted this logic to entire districts and cities, leveraging the convergence of affordable IoT hardware, scalable cloud infrastructure,andmaturemachinelearninglibraries[2].

Whatgivesmodernurbandigitaltwinstheirdistinctive power is the integration of AI reasoning. When machine learning models for anomaly detection, demand forecasting, or scenario optimisation are embedded withinadigitaltwin,itshiftsfromapassivemonitoringtool intoanactivedecisionsupportenvironment.Cityoperators canpose'whatif'questions,exploredownstreameffectsofa proposed intervention, and receive ranked recommendationsbeforeanythingchangesinthephysical world[3].

1.1 Problem Statement

Despitethetheoreticalappealofthisvision,real-world deploymentsfacesubstantial friction.Manycitieslack the baselinesensordensityneededtofeedareliabletwin.Data collected by different agencies often sits in incompatible formatsbehindorganisationalwalls.Buildingandrunninga city-scaletwindemandssignificantcapitalandspecialised

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

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

talentthatsmallermunicipalitiessimplydonothave.Andas urbansystemsbecomemoreconnected,theyalsobecome moreattractivetargetsforcyberdisruption[4].

1.2 Objectives of This Review

Thispapersetsoutto:

 SynthesiseresearchonAI-enhanceddigitaltwinsystems published between 2021 and 2025 with relevance to smartcityapplications.

 Compare the architectural approaches adopted in representativestudies.

 Evaluatehowwellthesesystemsaddresscoredomains: mobility,energy,environment,andinfrastructuresafety.

 Identify persistent technical and governance barriers thatlimitscale-up.

 Proposeaforward-lookingresearchagendagroundedin theidentifiedgaps.

2. LITERATURE REVIEW

A growing body of work has examined digital twin deployments across various facets of smart city management.Thestudiesreviewedherewereselectedon thebasisofrecency,methodologicalclarity,andbreadthof applicationdomain.

2.1 Review of Key Studies

Sacoto-Cabrera et al. [5] carried out a systematic mappingoftheresearchlandscapeattheintersectionofIoT, AI, and digital twinning in urban environments. Covering publications from 2018 to 2024, they found that interoperabilityanddataheterogeneityfiguredasthetwo most frequently cited obstacles, while traffic and energy stood out as the domains with the most active research communities.

Xu et al. [6] focused on how generative AI models particularly large language models and diffusion-based architectures can be embedded within digital twin pipelines to overcome sparse or missing data problems. Theirexperimentsshowedthatgenerativepre-processing improved simulation fidelity in scenarios where direct sensorcoveragewasthin.

Grübeletal.[7]raisedthequestionofequitableaccess, arguing that most high-quality digital twin deployments concentrateinwealthycitiesandthatopen-sourceplatforms and shared data standards are prerequisites for broader adoption.

Hu[8]providedapracticallyorientedaccountofdigital twinimplementationsacrossthreemid-sizedChinesecities, documenting measurable improvements in infrastructure inspection cycles and energy consumption monitoring throughbefore-and-afterquantitativecomparisons.

El-Agamyetal.[9]appliedbibliometricmethodstoover 1,200digital-twin-relatedpublications,tracingcleargrowth trajectorieswithsmartcityapplicationsrisingsharplyfrom 2021 onward. Their keyword co-occurrence analysis revealed that predictive maintenance, sustainability, and resilience have emerged as dominant thematic clusters in recentyears.

2.2 Comparative Summary of Selected Studies

Author(s) Year Primary Domain Notable Contribution

SacotoCabrera 2025 IoT+AI+DT Systematicreview; gapmap

Xuetal. 2024 GenerativeAI Sparse-data mitigation

El-Agamyet al. 2024 Bibliometrics Trendandcluster analysis

Hu 2023 Urban infrastructure Quantifiedfield outcomes

Kubasetal. 2023 Transportation Transittwin prototype

Adreanietal. 2023 DTframeworks Layered architecturedesign

Lietal. 2021 SmartCityDT Five-layer framework

Zhang&Chen 2021 DTconcepts Lifecyclecost considerations

3. SYSTEM ARCHITECTURE AND METHODOLOGY

Fig

-1:DigitalTwinArchitectureforSmartCities

Althoughnosinglearchitecturehasbecomeauniversal standard,afive-layermodelappearsrecurrentlyacrossthe literatureandprovidesa useful organisingframework for understanding how data flows from the physical city into actionableinsights.

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

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

3.1 Physical Sensing and Data Acquisition Layer

Thefoundationconsistsofinstrumentsembeddedinthe urban environment: fixed air-quality monitors, traffic inductionloops,structuralstraingaugesonbridges,smart energymetres,meteorologicalstations,CCTVcameraswith computer-visioncapabilities,andGPStranspondersinpublic transport fleets. Sensor placement involves spatial optimisationalgorithmstomaximisecoverageperunitcost [12].

3.2 Communication and Data Transmission Layer

Rawsensorreadingstraveltoprocessinginfrastructure througha mixture ofprotocols:LoRaWAN andNB-IoTfor long-range, low-bandwidth devices; 5G cellular links for high-throughput video feeds; and fibre backhaul for fixed infrastructure.Edgegatewaysoftenperforminitialfiltering andcompressionatthisstage,reducingthevolumeofdata thatmusttraversethenetwork[13].

3.3 Data Integration and Storage Layer

Incomingstreamsfromheterogeneoussourcesmustbe normalisedintoa common data model beforetheycan be usedjointly.Semanticwebtechnologiesandontology-based schemas have been proposed for this purpose, as have middlewareplatformssuchasFIWARE,whichprovidesopen APIs for urban data exchange. Cloud object stores and distributed time-series databases provide the persistence layer[5].

3.4 AI Analytics and Modelling Layer

This layer hosts the digital twin's intelligence. Distinct modeltypesservedifferentpurposes:

 Short-horizonpredictionmodels(e.g.,LSTMnetworks fortrafficvolume,gradientboostingforenergydemand) supportoperationaldispatchingdecisions.

 Anomaly detection models flag readings that deviate from expected patterns, triggering maintenance workflowsbeforefailuresoccur.

 Agent-based or physics-informed simulations allow plannerstoexplorelonger-horizonscenariossuchasa newbusrouteorgridresilienceunderextremeheat[6].

 Generativemodelscansynthesiserealisticscenariodata for situations that have never occurred historically, addressingakeylimitationinrare-eventcontexts[6].

3.5 Visualisation and Decision Support Layer

Resultsaresurfacedthroughgeospatialdashboards,3D citymodelsrenderedinplatformssuchasCesiumJSorUnity, and alert management systems that route notifications to relevant operators. The usability of this layer has a direct bearingonwhetherthetwin'svalueisrealisedinpractice [7].

4. APPLICATION DOMAINS AND PERFORMANCE ANALYSIS

4.1 Urban Mobility and Traffic Management

A city-scale twin continuously ingests GPS traces, loop detectorcounts,andsignalstatedatatomaintainadynamic modeloftrafficdensityacrosstheroadnetwork.Whenthe model detects conditions likely to produce congestion, it pushes revised signal timing plans to intersections before queues form [11]. Case evidence from Kubas et al. [14] suggeststhatadaptivesignalcontrolinformedbyadigital twinreducesaverageintersectiondelayby12–18percent comparedtofixed-timeplans,withlargergainsduringoutof-distributiondemandevents.

4.2 Energy System Monitoring and Demand Optimisation

Smart grid digital twins integrate readings from householdmetres,substationsensors,andweatherforecasts tomodelsupplyanddemandacrossanelectricitynetwork. Alharbeyetal.[15]demonstratedaprototypeinwhichthe twin reduced peak-demand forecasting error by approximately 23 percent relative to a baseline autoregressive model, allowing grid operators to prepositionreservesmoreefficiently.Zhangetal.[16]reported averageenergysavingsof8–14percentacrossaportfolioof public buildings following twin-informed operational recommendations.

4.3 Environmental Quality Monitoring

Urban digital twins sidestep the cost constraint of traditional air-quality monitoring by combining a sparse network of high-quality reference stations with a dense array of lower-cost indicative sensors and a dispersion model calibrated against reference data. This hybrid approachproducespollutionestimatesatresolutionsoftens of metres rather than several kilometres [8], enabling plannerstotailorlow-emissionzoneboundariestoachieve target air-quality improvements at minimum economic disruption.

Fig -2:ApplicationsofDigitalTwininSmartCities

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

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

4.4 Structural Health and Infrastructure Maintenance

Embedding digital twins with data from vibration sensors, strain gauges, and corrosion probes allows asset managerstoshiftfromcalendar-basedinspectioncyclesto condition-basedmaintenance,directingresourceswherethe model estimates risk is highest [10]. Temple et al. [17] describe a bridge twin where anomalous vibration signatures provided 72 hours of warning before a crack propagation event that would have required emergency closureunderconventionalinspectionregimes.

4.5 Disaster Preparedness and Emergency Response

A flood model embedded in a city twin can project inundation extents under different rainfall intensities, allowing emergency managers to pre-position resources, identifybottleneckevacuationroutes,andcommunicaterisk to residents before an event arrives [3]. Post-event, the twin's accumulated sensor record provides a detailed reconstruction of how the city performed, feeding lessons learneddirectlybackintothenextplanningcycle.

5. CHALLENGES AND BARRIERS TO ADOPTION

5.1

Data Fragmentation and Quality

Urbandataisproducedbymanydifferentdepartments andagencies,eachwithitsowncollectionprotocols,naming conventions,andupdatecadences.Mergingthese streams into a coherent data model is time-consuming and errorprone, and the resulting dataset still inherits the quality limitations of its worst-performing source [5]. Missing values, sensor drift, and clock misalignment are routine problemsrequiringrobustdataengineeringpractice.

5.2 Financial and Organisational Barriers

Standing up a city-scale digital twin requires upfront investment in sensor networks, communication infrastructure,compute,andtheskilledpersonneltobuild and maintain the models. Smaller municipalities often cannotabsorbthesecosts,andthereturnoninvestmentis difficult to quantify in advance, making the business case hardertopresentthanfortangiblephysicalinvestments[9].

5.3 Cybersecurity Exposure

Bringing together sensor data, operational control interfaces, and rich geospatial information about critical infrastructurecreatesa concentrated, high-value target. A compromiseofthetwin'sdataintegritycouldresultinfalse readingspropagatingthroughthedecisionsupportsystem, whileacompromiseofcontrolinterfacescouldhavedirect physicalconsequences[4].Securityarchitectureforurban digitaltwinsneedstobethreat-modelledexplicitly.

5.4 Privacy and Ethical Considerations

Detailed,real-timemodelsofurbanbehaviourinevitably captureinformationaboutindividuals theirmovements, patterns, and energy use. Without clear governance frameworksdefiningdataminimisationprinciples,retention limits, and meaningful consent mechanisms, digital twin programmes risk eroding public trust and creating new vectorsforsurveillance[7].

5.5

Interoperability and Standardisation

Intheabsenceofagreeddataexchangestandards,cities thatadoptoneplatformoftenfindthemselveslockedin,and cross-citycomparisonsorfederatedtwinnetworksbecome difficulttoestablish[11].BodiessuchasISOandOGChave begun work on relevant standards, but practical convergenceremainssomewayoff.

6. FUTURE RESEARCH DIRECTIONS

6.1 Generative AI for Simulation Enrichment

Largegenerativemodelsopenthepossibilityofsynthetic yet statistically realistic scenario data, addressing the sparse-history problem that constrains probabilistic risk assessment for rare urban events. Research is needed on validationmethodologiesthatestablishwhethergenerative outputsaresufficientlyreliableforoperationaluse[6].

6.2 Edge-Deployed Intelligence

Routingallsensordatatoacentralcloudincurslatency and bandwidth costs problematic for time-critical applications.AdvancesinedgeAIchipsmakeitincreasingly feasibletoruninferencemodelsongatewaydevicescloseto sensors, sending only anomaly flags or compressed summaries to the central twin [13]. Understanding which model classes perform well under tight edge resource constraintsisanactiveresearchquestion.

6.3

Federated Learning for Privacy-Preserving Collaboration

Federatedlearningallowsmultiplepartiestojointlytrain amodelwithoutsharingrawdata,makingitanaturalfitfor themulti-agencycontextofurbandigitaltwins.Earlyproofof-conceptstudiesexist,butengineeringthenecessarytrust infrastructure and handling statistical challenges of nonidentical data distributions across agencies remain open problems.

6.4

Explainability and Human-Centred Design

Operatorsandelectedofficialswhowillactonadigital twin'srecommendationsneedtounderstandnotjustwhat themodelsaysbutwhyitsaysit.IntegratingexplainableAI techniquesintourbantwindashboards,andconductinguser researchonhowdifferent explanationformatssupport or

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

Volume: 13 Issue: 04 | Apr 2026 www.irjet.net p-ISSN: 2395-0072

hinderdecisionquality,representsanimportantbutunderstudieddimensionofthefield[7].

6.5 Cross-City Federation and Benchmarking

As more cities deploy digital twins, the opportunity growstocompareapproaches,sharemodels,andevaluate performance against common benchmarks. Establishing federated twin networks with standardised APIs would accelerate learning across the research community and reduce redundant effort currently invested in building similarsolutionsfromscratchindifferentcities.

7. CONCLUSION

Urbandigitaltwins,enrichedbyAIandsustainedbyIoT data pipelines, represent a genuinely transformative approachtocitymanagement.Theliteraturereviewedhere documents tangible benefits across traffic optimisation, energy efficiency, environmental monitoring, structural maintenance,andemergencyplanning.Ineachdomain,the core advantage is the same: the ability to reason about a city'sbehaviourinadvanceofreal-worldaction,informedby a continuously updated model rather than by intuition or historical averages alone. Yet the path from compelling prototype to routine city-scale deployment remains obstructed by familiar barriers fragmented data, constrained budgets, unresolved security risks, and insufficient standards. Public trust cannot be assumed; it mustbeearnedthroughtransparentdatagovernanceand demonstrableaccountability.Thisreviewsuggeststhatthe next phase of progress depends as much on institutional innovation as on algorithmic advances. Cities, technology providers,academicresearchers,andstandardsbodiesneed to co-create the governance scaffolding within which increasingly sophisticated urban twins can operate safely andsustainably.

ACKNOWLEDGEMENT

Theauthorswouldliketothank[Guide/InstitutionName] fortheirguidanceandsupportthroughoutthepreparation ofthismanuscript.

REFERENCES

[1] United Nations Department of Economic and Social Affairs, "World Urbanization Prospects: The 2022 Revision,"UNDESA,NewYork,2022.

[2] M.Batty,"DigitalTwins,"EnvironmentandPlanningB: UrbanAnalyticsandCityScience,vol.45,no.5,pp.817–820,2018.

[3] R. Kitchin, "The Real-Time City? Big Data and Smart Urbanism,"GeoJournal,vol.79,no.1,pp.1–14,2014.

[4] R.Alharbey,T.Alsubait,andH.Alhakami,"DigitalTwin TechnologyforSmartGridPerformance,"IEEEAccess, vol.12,2024.

[5] E.Sacoto-Cabrera,J.Muñoz-Romero,andL.ValenzuelaValdés, "IoT, AI, and Digital Twins in Smart Cities: A SystematicReview," Smart CitiesJournal,vol.8,no. 1, 2025.

[6] H.Xu,Y.Wang,andJ.Liu,"LeveragingGenerativeAIfor UrbanDigitalTwins:ScenariosandChallenges,"Urban Informatics,vol.3,no.2,2024.

[7] J. Grübel, T. Thrash, C. R. Hölscher, and V. Schinazi, "Access to Urban Digital Twins: Open Platforms and Collaborative Data Systems," arXiv preprint arXiv:2204.01747,2022.

[8] L. Hu, "Research on the Application of Digital Twin TechnologyinSmartCityInfrastructureManagement," JournalofUrbanTechnology,vol.30,no.3,pp.45–63, 2023.

[9] R. F. El-Agamy, M. H. Saleh, and A. Ibrahim, "ComprehensiveBibliometricAnalysisofDigitalTwins in Smart City Applications," Artificial Intelligence Review,vol.57,no.4,2024.

[10] D. Li, W. Yu, and Z. Shao, "Smart City Development Based on Digital Twin Technology: Framework and Practice,"ComputationalUrbanScience,vol.1,no.1,pp. 1–11,2021.

[11]X.ZhangandR.Chen,"DigitalTwinsforSmartCities: Concepts, Challenges, and Emerging Opportunities," IEEEAccess,vol.9,pp.110900–110915,2021.

[12]L.Temple,A.Moreira,andP.Costa,"DigitalTwinsfor Smart City Policy Making: From Monitoring to Simulation," Sustainable Cities and Society, vol. 102, 2024.

[13]L.Adreani,F.Paolucci,andM.Cerruti,"ALayeredSmart City Digital Twin Framework for Real-Time Urban Management,"SmartCities,vol.6,no.5,2023.

[14]J.Kubas,A.Zielińska,andK.Nowakowski,"DigitalTwin in Smart City Transportation: A Prototype Study," TransportationResearchProcedia,vol.64,pp.389–397, 2023.

[15]R.Alharbeyetal.,"DigitalTwinTechnologyforSmart Grid Energy Performance Optimisation," IEEE TransactionsonSmartGrid,2024.

[16] J. Zhang, X. Li, and H. Wang, "Smart City Platform Construction Technology Based on Digital Twins and IoT,"AppliedSciences,vol.14,no.3,2024.

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

[17] L. Temple et al., "Structural Health Monitoring via UrbanDigitalTwinIntegration,"EngineeringStructures, vol.305,2024.