IoT-driven anomaly detection in smart grids using multimodal deep learning models

The increasing complexity of modern power systems, driven by the integration of Internet of Things (IoT) devices and distributed energy resources, has amplified the need for robust anomaly detection mechanisms in smart grids. This study proposes an IoT-driven multimodal deep learning framework that integrates time series data from SCADA/PMU systems, environmental sensor readings, and thermal/visual images to enhance anomaly detection performance. The proposed architecture combines Long Short Term Memory (LSTM) networks for temporal modeling, Convolutional Neural Networks (CNNs) for spatial feature extraction, and a hybrid feature fusion strategy to exploit complementary information across modalities. Experiments conducted on benchmark datasets demonstrate that the framework outperforms traditional machine learning and single modality deep learning methods, achieving an F1 score of 0.88 and a ROC AUC of 0.94. These results confirm the potential of multimodal deep learning to improve the reliability, resilience, and situational awareness of smart grids.