A Dispatch Data-Driven Fusion Method with Long Short-Term Memory for Single-Phase-to-Ground Fault Feeder Selection in Small Current Grounding Systems
Haina Rong, Gexiang Zhang, Maojun Ran, Xiantai Gou and Qiang Yang
Single-phase-to-ground fault detection in small current grounding systems has long been a challenging problem due to the small fault current. Although various methods have been developed to select faulty feeders via equipment deployed at each sub-station/bus. A more reliable faulty feeder selection method is desired to match the highly developed dispatch control systems. This paper proposes a novel dispatch data driven fusion method containing long short-term memory for single phase-to-ground fault feeder selection in small current grounding systems. By using this method, only one faulty feeder selection device is deployed in the dispatch center to detect and select faulty feeders from a hybrid grounded distribution network. A time window continuously intercepts the collected dispatch data and extracts the operation information of each sub-station. Two sub-models are combined to handle different grounding situations. A feature-based sub-module extracts features from zero-current and reactive power in order to efficiently detect faults in ungrounded systems when all needed information is available. A data-driven long short-term memory sub-module is designed to select faulty feeders when the previous submodule is unable to effectively detect or cannot be used due to information loss. By jointly using these sub-models, the proposed method achieves high detection accuracy and ensures effective detection in various scenarios. The precision of the method was verified through simulation experiments, validated by case studies, and confirmed over a year of online testing.
Keywords: Fault feeder selection, small current system, single phase to ground fault, deep neural network
DOI: 10.32908/ijuc.v21.zhang08