A Real-Time Big Data-Driven Cloud-Edge-End Fault Localization Method for Converter Stations

With the increasing complexity of high voltage direct current (HVDC) transmission system, the traditional fault location methods for converter stations face problems such as high heterogeneity of data, poor real-time performance and strong dependence on single source data. This paper proposes a cloud side collaborative fault location method based on real-time big data drive, which realizes efficient fusion of multi-source data and distributed feature extraction by combining Dynamic Spatio Temporal Graph Network (DSTG) and lightweight Federated Learning (FL). Cloud side collaborative federated learning framework in which edge nodes perform local model training, cloud side performs global aggregation and optimization, and gave consideration to data privacy and computing efficiency. The fragmentation transmission strategy based on data priority reduces the communication delay by more than 30%. The simulation experiment based on ± 800kV converter station shows that the positioning accuracy and response time of this method are 23.7% and 41.2% higher than those of existing methods, respectively, which verifies its effectiveness.