基于自监督图神经网络的传感器数据监测
Sensor Data Monitoring Based on Self-Supervised Graph Neural Networks
武琼1
作者信息
- 1. 北京冉腾语云科技有限公司,北京 100080
- 折叠
摘要
物联网环境监测是当今世界最为重要且核心的技术之一,而物联网感知世界的能力是通过传感器获得的,传感器数据预警是智能物联网的基础.文章利用端到端图神经网络技术,为传感器数据建立传感器知识图谱,在编码端,通过传感器属性注意力图神经网络和传感器关系注意力图神经网络,构建传感器属性图谱和传感器关系图谱;在解码端,用图卷积神经网络对传感器数据解码,通过余铉交叉损失函数对编码端与解码端数据训练.依据训练好的模型,以传感器属性分类作为下游任务,实时监测传感器数据.在真实数据下实验,以多种基线为目标,该方法均表现优良.
Abstract
IoT environmental monitoring is one of the best important and core technologies today,and IoT perception capability is obtained through sensors,and sensor data monitoring is the basis of intelligent internet of Things.In the paper,the sensor knowledge graph is established by using the end-to-end graph neural network technology.At Encoding end sensor-attribute knowledge graph and sensor-relationship knowledge graph is constructed by GAT.At Decoding end sensor data is decoded by GCN.The encoding end and decoding end data are trained by the residual cross loss function.According to the trained model,the sensor attribute data is classified as a downstream task,and the sensor data is monitored in real time.In experiments with real data and multiple baselines,our approach has performed well.
关键词
自监督/图神经网络/图注意力/图卷积/物联网/传感器Key words
self-supervised/GNN/GAT/GCN/IoT/sensor引用本文复制引用
出版年
2024
NETL
NSTL
万方数据