Research on Intelligent Fault Diagnosis Algorithm of Transmission Line Equipment Based on Deep Learning Spatio-Temporal Model and Monitoring Data Fusion

This study proposes a deep learning based spatiotemporal fusion fault diagnosis model (CASN) to address the challenges of multi-source data spatiotemporal heterogeneity and adaptability to complex scenarios in transmission line equipment fault diagnosis. A three-level architecture is constructed for decoupling temporal features, enhancing spatial attention, and cross modal adaptive fusion. The time-frequency domain features are coupled through a bidirectional long short-term memory network combined with Fourier transform. The GhostNet lightweight backbone network and coordinate attention mechanism are used to improve the accuracy of small target detection, and a gate controlled loop unit is designed to solve the problem of spatiotemporal alignment of multimodal data. The experiment was based on 200000 time-series samples from a provincial power grid and 320000 frames of drone image data for verification. The results showed that the joint diagnosis accuracy of the model reached 96.8 %, which was 13.3 % and 5.6 % higher than that of a single time-series and image model, respectively. The small target recall rate reached 88.0 %, the edge device inference delay was controlled at 42 ms, and the computational load was reduced to 1.8 GFLOPs. The research breaks through the single mode information barrier, constructs a collaborative characterization mechanism of “time dynamic evolution spatial structure anomaly”, and realizes the collaborative optimization of diagnostic accuracy and edge deployment efficiency. This model provides a solution for fault diagnosis of transmission line equipment that combines technological innovation and engineering practicality. Its multimodal spatiotemporal fusion strategy has important reference value for status monitoring of smart grid equipment.