Cloud-Native Digital Twin Architecture for Intelligent Microgrids: Real-Time Synchronization and Data-Driven Predictive Management

Current electrical networks must be restructured to meet decarbonization targets and modern energy requirements, particularly due to the high integration of distributed resources, which complicates their management. In this context, this paper proposes a cloud-native digital twin architecture for real-time monitoring, predictive management, and decision support in a microgrid. The proposed model is based on an intelligent modular architecture implemented in a cloud environment. Each entity in the microgrid is modeled using the Digital Twins Definition Language, where updates are synchronized in real time. The architecture also integrates three machine learning models dedicated to prediction and diagnosis: (1) a multilayer perceptron (MLP) for predicting and real-time simulation of the microgrid behavior when a digital twin parameter is modified; (2) a temporal fusion transformer (TFT) model for sequential forecasting of power generation and demand; and (3) a Random Forest classifier for fault identification. The architecture was evaluated in a simulation-in-the-loop testbed comprising a MATLAB-Simulink microgrid model, a physical ESP32 communication interface, and a Node-RED dashboard. The results show consistent replication of the Simulink source telemetry in the cloud-based twin under the evaluated conditions. The MLP achieves mean absolute errors between 0.1 and 0.36 kW, while the TFT achieves an error of approximately 2 kW for Ppv. This integrated digital twin architecture, combining prediction and diagnosis in a cloud environment, provides a solid foundation for more efficient management of electrical microgrids.