Weak Feature Fault Identification and Location of Distribution Network Based on Multi-Task Learning

Abstract

The field of distribution network has entered the era of “big data”, and deep learning with strong adaptive feature extraction, classification and prediction ability has also achieved fruitful results in the distribution network data processing. However, most of these studies were all performed under a single label system to achieve a single task objective. In the context of big data, single label system not only cuts off the connection between multi-tasks in fault identification of distribution network, but also can't completely describe various state information such as fault type and segment location of distribution network from fault data. Aiming at the above problems, a method of weak feature fault identification of distribution network based on multi-task learning is proposed. Its advantage lies in that it adaptively extracted the features of different target tasks from the same distribution network fault data and discriminated the types through the shared network with global feature pooling. The experimental results show that the proposed method can't only realize the classification of weak feature faults in distribution network and locate the faults in and out of sections, but also has high accuracy and calculation efficiency.