|
Title
|
Detection and Classification of Transmission Line Faults Using Multi-Source Transfer Learning Enhanced by Active Learning
|
|
Type
|
Refereeing
|
|
Keywords
|
Faults, Machine learning, Transfer learning, Active Learning, Phasor Measurement Units
|
|
Abstract
|
This paper presents a novel application of transfer learning to facilitate the detection and classification of transmission line faults captured by phasor measurement unit (PMU) datasets with limited to no labels. The proposed algorithm is based on Multi-Source Transfer Learning enhanced by Active Learning (MSTL-AL). It guarantees the ability to utilize combined PMU data collected from multiple electric grid sources across various geographical locations and topologies to develop a model that can be applied to a new (target) dataset. The suggested algorithm also takes the multi-source transfer learning process a step further by using active learning, coupled with the knowledge of a domain expert, to enhance the model’s performance for the new dataset. The MSTL-AL algorithm exceeds the performance of the benchmark deep transfer learning algorithms, as demonstrated. It also speeds up the detection and classification of faults in smaller, less diverse PMU datasets collected from a new application domain target.
|
|
Researchers
|
Seyed Alireza Bashiri Mosavi (Referee)
|