AI-Based Digital Twin Architecture for Smart City Infrastructure Management Using Deep Learning Models
Abstract
Rapid urbanization in metropolitan areas of developing countries demands a transformative approach to infrastructure management. This study designs, implements, and evaluates an Artificial Intelligence-based Digital Twin (AIDT) architecture for Smart City infrastructure management in the Jakarta Metropolitan Area through a 12-month field study. The proposed four-layer architecture integrates multi-protocol IoT data acquisition, an Eclipse Ditto-based real-time synchronization engine, an ensemble AI analytics module (LSTM, Gradient Boosting, and Federated Learning), and a React.js/CesiumJS decision-support dashboard. The system was evaluated across 47 sensor nodes spanning three infrastructure domains: transportation, energy, and water management. Results show that the AI-DT system achieved a weighted anomaly-detection F1-score of 0.891, a 28.8 percentage-point improvement over a baseline SCADA system evaluated under identical sensor coverage and observation conditions (p < 0.001). The predictive ensemble models achieved an energy-consumption MAPE of 4.23%, a traffic-flow RMSE of 187.4 vehicles/hour, and a water-demand MAPE of 4.88% over a 24-hour forecasting horizon. In addition, the system recorded 99.4% uptime with a 97.3% data-completeness rate, and fully satisfied privacy-compliance requirements under Indonesian Government Regulation PP 71/2019 through Federated Learning implementation with differential privacy (ε = 1.0). A qualitative evaluation of 24 stakeholders confirmed an improved capability to anticipate infrastructure failures, while also identifying the need for human-resource capacity building and stronger inter-system interoperability as prerequisites for full adoption. Reported performance figures characterize instrumented locations rather than the entire metropolitan area, and scalability beyond the 47-node pilot is discussed as an explicit direction for future validation. This research contributes one of the first empirically validated AI-DT frameworks in the context of a developing-country megacity, offering a replicable blueprint for other cities across Southeast Asia