基于Attention机制的GRU交换机故障研判方法  

A Fault Diagnosis Method for GRU Switch Based on Attention Mechanism

作　　者：李宁 张建功 刘学 LI Ning;ZHANG Jiangong;LIU Xue(State Grid Hebei Electric Power Co.,Ltd.,Cangzhou Power Supply Branch,Cangzhou 061000,China)

机构地区：[1]国网河北省电力有限公司沧州供电分公司,河北沧州061000

出　　处：《通信电源技术》2023年第11期213-215,共3页Telecom Power Technology

摘　　要：提出了一种基于Attention机制的门控循环单元(Gated Recurrent Unit,GRU)交换机故障研判方法,该方法将交换机故障前10 min的内存占用率、中央处理器(Central Processing Unit,CPU)使用率以及风扇转速等数据作为输入,用于搭建GRU网络架构学习交换机动态变化特征,将时间序列输入GRU模型,建模学习特征向量的变化规律;同时,引入Attention机制将隐藏状态设定不同的权重,加强了重要方面数据的影响。所设计实验使用了该交换机数据集,实验证明该方法可提高交换机故障定位与识别的效率和精度,对样本误差具有良好的鲁棒性。This paper proposes a fault detection method for Gated Recurrent Neural Network(GRU)switches based on Attention mechanism.In this method,data such as the memory usage,Central Processing Unit(CPU)usage,and fan speed within the first 10 minutes of a switch failure are used as inputs to build a GRU network architecture and learn the dynamic change characteristics of switches.The time series is input into the GRU model.The change rule of modeling learning feature vector.At the same time,the Attention mechanism is introduced to set different weights for hidden states,which strengthens the influence of important aspects of data.In this paper,the switch data set is used in the design experiments.Experiments show that this method can improve the efficiency and accuracy of switch fault location and identification,and has good robustness to sample errors.

关 键 词：门控循环单元(GRU) Attention机制 交换机 故障研判 

分 类 号：TN915.05[电子电信—通信与信息系统]