Research on Intelligent Fault Diagnosis and Self-Healing Control for Smart Grids Based on Digital Twin

As the structure of smart grids becomes increasingly complex, traditional fault handling methods face severe challenges in response speed and decision optimization. To address this issue, this paper proposes and constructs a framework for smart grid fault diagnosis and self-healing control based on digital twin technology. The core of this framework is a co-simulation-based multi-scale hybrid modeling method: an Electromagnetic Transient (EMT) model built with MATLAB/Simulink is capable of generating high-fidelity data, while an Electromechanical Transient (RMS) model developed in OpenDSS serves as a "decision-making sandbox" for the rapid deduction of multi-agent self-healing strategies. This study introduces a novel validation methodology where the digital twin is used to generate data for complex fault scenarios—such as high-impedance faults—that are poorly represented in standard datasets. Our CNN-LSTM fault diagnosis model, when augmented with this twin-generated data, demonstrates significantly improved robustness. In terms of fault diagnosis, the CNN-LSTM model designed in this study achieved an accuracy of 99.5% on a public benchmark dataset, with an average diagnosis time of only 26.1ms, striking a superior balance between performance and efficiency compared to other advanced algorithms. Building on this, the twin-inferred self-healing control strategy can complete fault isolation and restoration decisions within 2 seconds, increasing the system load recovery rate from 70% with traditional methods to 95% in typical scenarios. In summary, the digital twin framework proposed in this study provides an effective technical solution for the rapid and precise handling of grid faults, demonstrating its application potential in building future autonomous power systems.