Modern power grids face growing challenges from aging infrastructure, renewable integration, and unstable network. However, traditional monitoring systems struggle with complex data anomalies such as missing values, noise, and static thresholds. To address this, we propose a hybrid deep learning framework that combines the Qwen2 large language model with a novel TimeMixer++ architecture for intelligent anomaly detection and data repair. Our method fuses multi-modal inputs—including voltage, weather, and sensor data—and uses context-aware imputation and multi-scale temporal modeling to reconstruct missing or corrupted time series segments. A generative pipeline further enhances robustness in noisy or incomplete settings. Evaluated on real-world datasets from the Yunnan Power Grid and the IEEE 39-bus system, our approach achieves significantly lower mean absolute error (MAE) and mean squared error (MSE) than conventional baselines (e.g., ARIMA, GANs), especially under high data loss. The framework enables proactive maintenance by producing accurate, interpretable, and physically plausible reconstructions. This work demonstrates a scalable, data-driven path toward resilient grid operations, with potential applicability to diverse smart infrastructure systems.
Tang et al. (Fri,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: