Deep spatiotemporal feature fusion using CNN-LSTM autoencoder for real-time, phase-aware anomaly identification and localization in high-voltage power transmission networks

• Hybrid CNN–LSTM–Autoencoder enables phase-aware fault detection and localization. • Achieves 99.5% detection and 98.5% phase-level localization accuracy. • Reconstruction errors and thresholds provide interpretable fault analysis. • Outperforms state-of-the-art models in accuracy and robustness for smart grids. Reliable fault detection and phase-level fault localization in three-phase power transmission systems remain challenging due to increasing grid nonlinearity and dynamic operating conditions in modern smart grids. Conventional protection schemes and data-driven approaches often exhibit limited adaptability, poor generalization to unseen operating conditions, and insufficient capability for precise phase-wise fault identification. Recent deep learning-based methods primarily focus on fault classification and rely on supervised learning and heuristic decision thresholds, limiting their interpretability and practical deployment. To address these limitations, this paper proposes a novel spatiotemporal deep learning framework based on a hybrid CNN–LSTM–Autoencoder (CNN–LSTM–AE) architecture that integrates spatial feature extraction, temporal dependency modeling, and unsupervised reconstruction-based anomaly analysis within a unified system. A key innovation is a phase-wise reconstruction error profiling mechanism that enables identification of faulty phases and characterization of faults, with an adaptive thresholding strategy to enhance interpretability. The proposed framework is trained and validated using real-world transmission voltage and current measurements and evaluated under diverse fault scenarios and operating conditions. Comparative results demonstrate superior fault detection and localization performance compared with conventional CNN, LSTM, and hybrid deep learning models, with improved robustness and reduced false-positive rates. This unified, interpretable fault intelligence framework provides a scalable solution for real-time protection and self-healing operations in next-generation smart grid transmission systems.