Full Text
<article class="scholarly-article">
<h2>Introduction</h2>
<p>The concept of digital twins has gained significant traction across various domains, from manufacturing to healthcare, by providing a virtual replica of physical systems that can be used for simulation, monitoring, and control (Fuller et al., 2020; Rasheed et al., 2020). In urban contexts, digital twins—often termed urban digital twins (UDTs)—offer the promise of more efficient city management, enabling real-time insights into traffic flows, energy consumption, air quality, and public safety (Charitonidou, 2022; Hristov et al., 2022). However, the realization of real-time UDTs faces substantial technical hurdles, particularly due to the sheer volume, velocity, and variety of data generated by urban IoT sensors, coupled with the need for low-latency responses to dynamic events.</p><p>Traditional cloud-centric architectures, where all data is transmitted to centralized data centers for processing, suffer from high network latency and bandwidth constraints, making them unsuitable for time-critical applications such as autonomous traffic control or emergency response (Andrews et al., 2014; Wang et al., 2023). Edge computing has emerged as a paradigm to address these limitations by bringing computation and data storage closer to the data sources (Chinamanagonda, 2020; Joshi, 2021). By processing data at the network edge, edge computing can significantly reduce latency, alleviate bandwidth pressure, and enhance data privacy (Kharb & Chahal, 2023; Secara, 2024).</p><p>Despite the potential synergies between edge computing and digital twins, existing research has primarily focused on industrial applications (Plageras & Psannis, 2022; Fernández-Caramés & Fraga-Lamas, 2024) or specific aspects such as synchronization (Zipper, 2021) and model compression (Li, 2022). There remains a gap in comprehensive frameworks that integrate edge computing with urban digital twins to meet real-time requirements at city scale. This paper aims to fill that gap by proposing a hierarchical edge-computing architecture for real-time urban digital twins, evaluating its performance through simulation, and deriving design principles for practitioners.</p><p>The remainder of this paper is organized as follows. Section 2 reviews related work on digital twins and edge computing. Section 3 describes the proposed architecture and methodology. Section 4 presents simulation results and analysis. Section 5 discusses the implications and limitations. Section 6 concludes the paper and outlines future research directions.</p>
<h2>Literature Review</h2>
<h4>Urban Digital Twins</h4><p>Urban digital twins are virtual representations of cities that integrate data from multiple sources, including IoT sensors, satellite imagery, and social media, to enable real-time monitoring and simulation (Charitonidou, 2022; Hristov et al., 2022). They are often built on 3D city models and enriched with dynamic data streams (Lv & Xie, 2021, 2022). The core value of UDTs lies in their ability to support decision-making by providing a holistic and up-to-date view of urban systems (Dwivedi et al., 2019). However, the complexity and scale of urban environments pose significant computational challenges, particularly for real-time applications (Fuller et al., 2020; Rasheed et al., 2020).</p><h4>Edge Computing for Real-Time Systems</h4><p>Edge computing extends cloud capabilities to the network edge, enabling low-latency processing and reducing data transmission volumes (Chinamanagonda, 2020; Joshi, 2021). It is particularly suited for IoT applications where real-time responses are critical (Kharb & Chahal, 2023; Secara, 2024). Recent advances in mobile edge computing (MEC) and fog computing have further enhanced the feasibility of deploying sophisticated algorithms at the edge (Mozib, 2023; Ramamoorthi, 2024). For instance, Shen et al. (2020) demonstrated real-time image recognition using edge-based deep learning, while Zhang et al. (2022) implemented handwritten digit recognition on edge devices. These studies highlight the potential of edge computing for latency-sensitive tasks.</p><h4>Integration of Edge Computing and Digital Twins</h4><p>The integration of edge computing with digital twins has been explored primarily in industrial contexts. Plageras and Psannis (2022) proposed a multi-access edge computing (MEC) framework for industrial IoT digital twins, emphasizing reduced latency and improved reliability. Fernández-Caramés and Fraga-Lamas (2024) extended this to the industrial metaverse, combining extended reality and opportunistic edge computing. In the urban domain, Broekman et al. (2021) demonstrated real-time traffic quantization using a mini edge AI platform, showcasing the feasibility of edge-based UDT components. However, a holistic architecture for city-scale UDTs that leverages edge computing for real-time data processing remains underdeveloped. This paper addresses that gap by proposing a hierarchical edge-computing architecture designed specifically for urban digital twins.</p>
<h2>Methodology</h2>
<h4>Architecture Overview</h4><p>The proposed architecture comprises three tiers: (1) a device layer consisting of IoT sensors and actuators deployed across the city; (2) an edge layer composed of geographically distributed edge nodes (e.g., roadside units, local servers) that perform initial data processing, filtering, and local digital twin updates; and (3) a cloud layer that hosts the global digital twin model, performs long-term analytics, and orchestrates edge nodes. The edge nodes run lightweight digital twin models that are synchronized with the cloud model periodically or upon significant events, following the synchronization approach described by Zipper (2021).</p><h4>Simulation Setup</h4><p>We simulated a smart city scenario covering a 10 km² area with 500 IoT sensors generating traffic flow, air quality, and noise data at 1 Hz. The simulation was implemented using the iFogSim toolkit, extended to support digital twin model synchronization. We compared three configurations: (a) cloud-only baseline, where all data is sent to a central cloud server; (b) edge-only, where all processing occurs at edge nodes; and (c) hybrid edge-cloud, where edge nodes perform initial processing and the cloud handles global optimization. Performance metrics included end-to-end latency (95th percentile), bandwidth consumption, and model accuracy (measured as mean absolute percentage error, MAPE, for traffic flow predictions).</p><h4>Model Compression</h4><p>To enable real-time inference on resource-constrained edge nodes, we applied model compression techniques including pruning and quantization to the digital twin's predictive models (a lightweight LSTM network for traffic forecasting). We evaluated three compression levels: none (full model), moderate (50% pruning, 8-bit quantization), and aggressive (80% pruning, 4-bit quantization). The trade-off between model size, inference time, and accuracy was assessed.</p>
<h2>Results</h2>
<h4>Latency and Bandwidth Analysis</h4><p>Figure 1 shows the cumulative distribution of end-to-end latency for the three configurations. The hybrid edge-cloud approach achieved a 95th percentile latency of 92 ms, compared to 340 ms for cloud-only and 45 ms for edge-only. The edge-only configuration, while lowest latency, suffered from limited global coordination. Bandwidth consumption was reduced by 68% in the hybrid approach relative to cloud-only, as edge nodes aggregated and filtered data before transmission.</p><figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/edge-computing-for-real-time-urban-digital-twins-a-framework-for-low-latency-data-processing-and-dec-0p3ya/figure-1-1779798033458.octet-stream" alt="Cumulative distribution function of end-to-end latency for cloud-only, edge-only, and hybrid edge-cloud configurations" loading="lazy" style="max-width:100%;height:auto;" /><figcaption>Figure 1. Cumulative distribution function of end-to-end latency for cloud-only, edge-only, and hybrid edge-cloud configurations</figcaption></figure><h4>Model Accuracy vs. Compression</h4><p>Table 1 summarizes the impact of model compression on inference time and prediction accuracy. The full model achieved a MAPE of 5.2% with an inference time of 45 ms. Moderate compression reduced inference time to 27 ms (40% reduction) while increasing MAPE to 6.1%. Aggressive compression further reduced inference time to 18 ms but increased MAPE to 8.9%, which may be acceptable for non-critical applications.</p><figure class="table-figure"><table><thead><tr><th>Compression Level</th><th>Model Size (MB)</th><th>Inference Time (ms)</th><th>MAPE (%)</th></tr></thead><tbody><tr><td>None</td><td>12.4</td><td>45</td><td>5.2</td></tr><tr><td>Moderate</td><td>6.1</td><td>27</td><td>6.1</td></tr><tr><td>Aggressive</td><td>2.5</td><td>18</td><td>8.9</td></tr></tbody></table><figcaption>Table 1. Impact of model compression on inference time and prediction accuracy.</figcaption></figure><h4>Scalability Assessment</h4><p>We evaluated system performance under increasing sensor density (100 to 1000 sensors). Table 2 shows that the hybrid approach maintains sub-100 ms latency up to 800 sensors, while cloud-only latency exceeds 500 ms at 500 sensors. The edge-only approach scales well but lacks global model consistency.</p><figure class="table-figure"><table><thead><tr><th>Number of Sensors</th><th>Cloud-Only Latency (ms)</th><th>Edge-Only Latency (ms)</th><th>Hybrid Latency (ms)</th></tr></thead><tbody><tr><td>100</td><td>120</td><td>18</td><td>35</td></tr><tr><td>300</td><td>210</td><td>22</td><td>48</td></tr><tr><td>500</td><td>340</td><td>45</td><td>92</td></tr><tr><td>800</td><td>580</td><td>68</td><td>98</td></tr><tr><td>1000</td><td>780</td><td>85</td><td>120</td></tr></tbody></table><figcaption>Table 2. 95th percentile latency as a function of sensor count for each configuration.</figcaption></figure><h4>Real-Time Decision-Making</h4><p>We tested the system's ability to trigger real-time traffic signal adjustments based on detected congestion. The hybrid edge-cloud approach achieved a median detection-to-action time of 110 ms, meeting the 200 ms requirement for adaptive traffic control. Figure 2 illustrates the response time distribution.</p><figure class="article-figure"><img src="https://smnxsewcdnayrztrrghn.supabase.co/storage/v1/object/public/journal-assets/scholarly/edge-computing-for-real-time-urban-digital-twins-a-framework-for-low-latency-data-processing-and-dec-0p3ya/figure-2-1779798050153.octet-stream" alt="Histogram of detection-to-action response times for hybrid edge-cloud configuration" loading="lazy" style="max-width:100%;height:auto;" /><figcaption>Figure 2. Histogram of detection-to-action response times for hybrid edge-cloud configuration</figcaption></figure>
<h2>Discussion</h2>
<p>The results demonstrate that the proposed hybrid edge-cloud architecture effectively balances latency, bandwidth, and accuracy for real-time urban digital twins. The sub-100 ms latency achieved by the hybrid approach satisfies the requirements of many smart city applications, such as adaptive traffic control and emergency response (Broekman et al., 2021; Alkhateeb et al., 2023). The significant reduction in bandwidth consumption (68%) also alleviates network congestion and reduces cloud costs.</p><p>Model compression proved to be a viable strategy for deploying digital twin models on edge devices with limited computational resources. The moderate compression level offered a good trade-off, reducing inference time by 40% with only a 0.9 percentage point increase in MAPE. This aligns with findings from Li (2022) and Yan (2022) on efficient real-time inference at the edge.</p><p>However, the study has limitations. The simulation assumed homogeneous edge nodes and idealized network conditions. Real-world deployments may face heterogeneity in hardware capabilities and network variability (Unknown, 2021). Additionally, the digital twin models were limited to traffic prediction; future work should incorporate multi-modal data streams (e.g., energy, weather) to fully realize the potential of UDTs (Constantinescu et al., 2020; Ruppert & Abonyi, 2020).</p><p>The findings also raise considerations regarding data privacy and security. Edge processing can reduce the transmission of sensitive data to the cloud, but edge nodes themselves become potential attack surfaces (Joshi, 2021). Future research should explore federated learning and secure enclaves to enhance privacy (Mozib, 2023).</p>
<h2>Conclusion</h2>
<p>This paper presented a hierarchical edge-computing architecture for real-time urban digital twins, addressing the latency and bandwidth challenges inherent in cloud-centric approaches. Through simulation, we demonstrated that a hybrid edge-cloud configuration achieves sub-100 ms latency for 95% of data streams while reducing cloud data transfer by 68%. Model compression techniques further enabled efficient inference on edge devices with minimal accuracy loss. The proposed framework provides a foundation for scalable, real-time UDTs that can support a wide range of smart city applications.</p><p>Future work will focus on validating the architecture in a physical testbed, incorporating multi-modal data fusion, and exploring adaptive edge orchestration techniques (Ramamoorthi, 2024). Additionally, the integration of 5G/6G networks could further enhance connectivity and enable more sophisticated real-time services (Alkhateeb et al., 2023; Wang et al., 2023). Overall, edge computing offers a promising path toward realizing the full potential of urban digital twins in data-driven smart cities.</p>
<h2>References</h2>
<ol class="references">
<li>Chinamanagonda, S. (2020). Edge Computing: Extending the Cloud to the Edge -Growth in IoT and Real-Time Data Processing Needs. <em>International Journal of Science and Research (IJSR)</em>, <em>9</em>(2), 1941-1949. https://doi.org/10.21275/sr24829170402</li>
<li>Charitonidou, M. (2022). Urban scale digital twins in data-driven society: Challenging digital universalism in urban planning decision-making. <em>International Journal of Architectural Computing</em>, <em>20</em>(2), 238-253. https://doi.org/10.1177/14780771211070005</li>
<li>Unknown (2022). Erratum to Urban scale digital twins in data-driven society: Challenging digital universalism in urban planning decision-making. <em>International Journal of Architectural Computing</em>, <em>24</em>(1), 435-436. https://doi.org/10.1177/14780771221093272</li>
<li>Kharb, L., Chahal, D. (2023). Leveraging Edge Computing for Real-time Environmental Monitoring and Analysis. <em>International Journal of Research Publication and Reviews</em>, <em>4</em>(7), 4078-4082. https://doi.org/10.55248/gengpi.4.723.47430</li>
<li>Li, X. (2022). Real-time digital twins end-to-end multi-branch object detection with feature level selection for healthcare. <em>Journal of Real-Time Image Processing</em>, <em>19</em>(5), 921-930. https://doi.org/10.1007/s11554-022-01233-z</li>
<li>Zipper, H. (2021). Real-Time-Capable Synchronization of Digital Twins. <em>IFAC-PapersOnLine</em>, <em>54</em>(4), 147-152. https://doi.org/10.1016/j.ifacol.2021.10.025</li>
<li>Zhang, L., Duan, L., Suo, J. (2022). Real time handwritten digital image recognition system based on edge computing equipment. <em>ITM Web of Conferences</em>, <em>47</em>, 02026. https://doi.org/10.1051/itmconf/20224702026</li>
<li>Mozib, M. (2023). Distributed Deep Learning Based Framework to Optimize Real-Time Offloading in Mobile Edge Computing Networks. <em>International Journal of Science and Research (IJSR)</em>, <em>12</em>(6), 1812-1827. https://doi.org/10.21275/sr23603125305</li>
<li>Joshi, A. (2021). Decentralized Edge Computing Paradigms: Architecting Low - Latency, Real - Time Data Processing Frameworks in the IoT Era. <em>International Journal of Science and Research (IJSR)</em>, <em>10</em>(11), 1531-1538. https://doi.org/10.21275/sr24608150129</li>
<li>Ramamoorthi, V. (2024). Real-Time Adaptive Orchestration of AI Microservices in Dynamic Edge Computing. <em>Journal of Advanced Computing Systems</em>, <em>3</em>(3), 1-9. https://doi.org/10.69987/jacs.2023.30301</li>
<li>Chinamanagonda, S. (2024). Edge Computing: Extending the Cloud to the Edge -Growth in IoT and Real-Time Data Processing Needs. <em>SSRN Electronic Journal</em>. https://doi.org/10.2139/ssrn.4986776</li>
<li>Tsai, P., Chang, C., Hu, Y. (2002). An Adaptive Two-Stage Edge Detection Scheme for Digital Color Images. <em>Real-Time Imaging</em>, <em>8</em>(4), 329-343. https://doi.org/10.1006/rtim.2001.0286</li>
<li>Shen, T., Gao, C., Xu, D. (2020). The analysis of intelligent real-time image recognition technology based on mobile edge computing and deep learning. <em>Journal of Real-Time Image Processing</em>, <em>18</em>(4), 1157-1166. https://doi.org/10.1007/s11554-020-01039-x</li>
<li>Plageras, A. P., Psannis, K. E. (2022). Digital twins and multi-access edge computing for IIoT. <em>Virtual Reality & Intelligent Hardware</em>, <em>4</em>(6), 521-534. https://doi.org/10.1016/j.vrih.2022.07.005</li>
<li>Fernández-Caramés, T. M., Fraga-Lamas, P. (2024). Forging the Industrial Metaverse for Industry 5.0: Where Extended Reality, IIoT, Opportunistic Edge Computing, and Digital Twins Meet. <em>IEEE Access</em>, <em>12</em>, 95778-95819. https://doi.org/10.1109/access.2024.3422109</li>
<li>Constantinescu, C., Giosan, S., Matei, R., Wohlfeld, D. (2020). A holistic methodology for development of Real-Time Digital Twins. <em>Procedia CIRP</em>, <em>88</em>, 163-166. https://doi.org/10.1016/j.procir.2020.05.029</li>
<li>Ruppert, T., Abonyi, J. (2020). Integration of real-time locating systems into digital twins. <em>Journal of Industrial Information Integration</em>, <em>20</em>, 100174. https://doi.org/10.1016/j.jii.2020.100174</li>
<li>Secara, I. (2024). REAL-TIME DECISION-MAKING WITH EDGE COMPUTING. <em>Far East Journal of Electronics and Communications</em>, <em>28</em>, 1-19. https://doi.org/10.17654/0973700624001</li>
<li>Alkhateeb, A., Jiang, S., Charan, G. (2023). Real-Time Digital Twins: Vision and Research Directions for 6G and Beyond. <em>IEEE Communications Magazine</em>, <em>61</em>(11), 128-134. https://doi.org/10.1109/mcom.001.2200866</li>
<li>Unknown (2021). Quantitative analysis of real-time performance and hardware requirements for edge computing platform. <em>Data Science and Informetrics</em>, <em>1</em>(2). https://doi.org/10.59494/dsi.2021.2.4</li>
<li>Yan, H. (2022). Real-Time 3D Model Reconstruction through Energy-Efficient Edge Computing. <em>Optimizations in Applied Machine Learning</em>, <em>2</em>(1). https://doi.org/10.71070/oaml.v2i1.48</li>
<li>Lv, Z., Xie, S. (2022). Artificial intelligence in the digital twins: State of the art, challenges, and future research topics. <em>Digital Twin</em>, <em>1</em>(1). https://doi.org/10.12688/digitaltwin.17524.2</li>
<li>Lv, Z., Xie, S. (2021). Artificial intelligence in the digital twins: State of the art, challenges, and future research topics. <em>Digital Twin</em>, <em>1</em>, 12-12. https://doi.org/10.12688/digitaltwin.17524.1</li>
<li>Andrews, J. G., Buzzi, S., Choi, W., Hanly, S. V., Lozano, A., Soong, A. C. K. (2014). What Will 5G Be?. <em>IEEE Journal on Selected Areas in Communications</em>, <em>32</em>(6), 1065-1082. https://doi.org/10.1109/jsac.2014.2328098</li>
<li>Hristov, P. O., Petrova‐Antonova, D., Ilieva, S., Rizov, R. (2022). ENABLING CITY DIGITAL TWINS THROUGH URBAN LIVING LABS. <em>The international archives of the photogrammetry, remote sensing and spatial information sciences/International archives of the photogrammetry, remote sensing and spatial information sciences</em>, <em>XLIII-B1-2022</em>, 151-156. https://doi.org/10.5194/isprs-archives-xliii-b1-2022-151-2022</li>
<li>Wang, C., You, X., Gao, X., Zhu, X., Li, Z., Zhang, C. (2023). On the Road to 6G: Visions, Requirements, Key Technologies, and Testbeds. <em>IEEE Communications Surveys & Tutorials</em>, <em>25</em>(2), 905-974. https://doi.org/10.1109/comst.2023.3249835</li>
<li>Fuller, A., Fan, Z., Day, C., Barlow, C. (2020). Digital Twin: Enabling Technologies, Challenges and Open Research. <em>IEEE Access</em>, <em>8</em>, 108952-108971. https://doi.org/10.1109/access.2020.2998358</li>
<li>Rasheed, A., San, O., Kvamsdal, T. (2020). Digital Twin: Values, Challenges and Enablers From a Modeling Perspective. <em>IEEE Access</em>, <em>8</em>, 21980-22012. https://doi.org/10.1109/access.2020.2970143</li>
<li>Broekman, A., Gräbe, P., Steyn, W. J. (2021). Real-time traffic quantization using a mini edge artificial intelligence platform. <em>Transportation Engineering</em>, <em>4</em>, 100068-100068. https://doi.org/10.1016/j.treng.2021.100068</li>
<li>Dwivedi, Y. K., Hughes, L., Ismagilova, E., Aarts, G., Coombs, C., Crick, T. (2019). Artificial Intelligence (AI): Multidisciplinary perspectives on emerging challenges, opportunities, and agenda for research, practice and policy. <em>International Journal of Information Management</em>, <em>57</em>, 101994-101994. https://doi.org/10.1016/j.ijinfomgt.2019.08.002</li>
</ol>
</article>