An Optical Partial Discharge Localization Method Based on Simulation and Machine learning in GIL

As gas-insulated transmission line (GIL) is widely used, the partial discharge (PD) phenomenon that occurs during their operation is one of the main reasons for the deterioration of their insulation status. Therefore, the detection and localization of PD in the GIL plays an important role in ensuring the safe and stable operation of the equipment. At present, the PD localization methods for GIL are mainly ultrahigh frequency (UHF) method and ultrasonic method, while these methods are susceptible to mechanical vibration and electromagnetic interference. Optical detection, as a sensitive and effective PD detection method, is rarely used in the field of GIL PD localization. Accordingly, this paper proposes a GIL PD localization method based on optical PD simulation and machine learning. This method establishes a simulation model that is exactly the same as the actual GIL in terms of structure, size and sensor arrangement, where the PD optical simulation experiment is performed to build a PD simulation fingerprint database. Each fingerprint in the fingerprint database corresponds to a PD source location information. Based on this, the PSO-SVM machine learning algorithm is used to match the actual PD fingerprints with the fingerprints in the simulation fingerprint database to obtain the localization results. This method overcomes the difficulty of collecting a large amount of field data to build a fingerprint database in the existing optical fingerprint localization method through simulation. And the structure of GIL in the simulation can be customized according to the actual equipment, while the workload of obtaining the fingerprint database through actual experiments is very heavy for different types of equipment. Through experimental verification, this paper selects 12 typical locations of PD sources in the GIL experimental tank for testing. The average localization error is 10.58mm.