Deep learning empowering gas monitoring in GIL pipeline corridor: super-resolution reconstruction of gas leakage distribution data and optimization of sensor placement

In Gas Insulated Metal Enclosed Transmission Lines (GIL), which are a unique environment with complex spatial structures, extensive spanning distances and enclosed spaces, the prevailing monitoring practices for Sulphur Hexafluoride (SF₆) concentration depend on a limited number of sensors. The gas concentration distribution is not readily visualized, and the monitoring lacks intuitive visibility. Thus, we employed computational fluid dynamics (CFD) to simulate typical gas leakage scenarios and constructed a digital twin model that accurately reflects the physical state of GIL systems, generating high-precision datasets. We propose a U-net-based super-resolution reconstruction (SRR) method capable of reconstruct high-resolution 3D gas concentration distribution from limited low-resolution sensor data. The investigation encompasses the impact of varying sensor numbers and distributions on the reconstruction outcomes. The findings indicate that enhancing the number of sensors can improve the reconstruction accuracy, with uniform sensor distribution yielding better results than non-uniform. The proposed method facilitates the reconstruction of sparsely distributed monitoring data of gas concentration into a high-density distribution and an intuitively visible monitoring image. It provides data support for optimizing the distribution of sensors in the pipeline corridor and the safe evacuation paths of the personnel in case of emergencies.