Edge-Cloud Collaborative Fault Detection for Distribution Networks Using Attention Mechanism

Abstract

As the penetration rate of renewables-based distributed generators soars in the power distribution network, the trustworthiness and timeliness of fault detection become a critical challenge. This paper embeds the concept of edge-cloud collaboration in the prevalent deep learning techniques so as to perform the data-driven fault detection in a fast and accurate way. First, an attention mechanism is proposed to associate the line fault probability with bus voltage profiles. On this basis, an attention-based deep neural network is designed for edge computing purposes, which manages to estimate the fault status of a certain line only by analyzing the voltage magnitude of adjacent buses. Besides, an efficient deep neural network is deployed in the cloud computing platform which work collaboratively with edge devices to detect any fault. Moreover, a batch training method is proposed to improve the training speed and the model accuracy while retaining the topology information. The validity of the proposed method is finally validated by numerical experiments based on several IEEE test feeder systems.