面向电力物联网调度任务的分布式知识图谱

Union of Distributed Knowledge Maps for Power Internet of Things

能够充分利用各类型信息的知识图谱技术是实现电力物联网网络智能调度的关键一环。然而由于电力网络分层调度的特点,造成了知识图谱分布式的问题。梳理了知识图谱的概念及构建的一般方法,提出了一种基于联邦学习思想的分布式知识图谱架构,边缘调度中心对知识图谱进行本地更新,并上传至高层云平台。高层云平台对知识图谱进行补全及融合,调度中心之间可相互迁移知识图谱的学习表示方法。在知识图谱融合方面的初步结果表明,基于GCN的方法远优于基于TransE的方法。

The knowledge map technology that can make full use of various types of information is a key link to realize the intelligent scheduling of the power Internet of Things network.However,due to the characteristics of power network hierarchical scheduling,the problem of knowledge map distribution is caused.This paper combs the concept of knowledge map and the general method of its construction,and proposes a distributed knowledge map architecture based on the idea of federated learning.The edge scheduling center updates the knowledge map locally and uploads it to the high-level cloud platform.The high-level cloud platform complements and merges the knowledge map,and the learning and representation methods of the knowledge map can be transferred between dispatching centers.The preliminary results of knowledge map fusion show that the method based on GCN is far superior to the method based on TransE.