摘要
通常的数据融合算法都是对传感器数据在同一时间不同空间的特征进行融合。其中,深度卷积神经网络(DCNN)融合性能突出,通过多个卷积层、池化层、全连接层实现特征提取、信息关联、决策判断一体化。在DCNN的基础上,提出基于卷积神经网络的数据融合算法(DAECNN_attention),引入降噪自编码器,用以实现数据去噪重建;引入自注意力机制模型,用以提高关键信息的提取能力。试验结果表明,与DCNN、BPNN数据融合算法相比,本文算法在UCI数据集上融合性能更优。
The common data fusion algorithms are to fuse the characteristics of sensor data in different spaces at the same time.Among them,the fusion performance of deep convolution neural network(DCNN)is outstanding.This method realizes the integration of feature extraction,information association and decision-making through multiple convolution layers,pooling layers and full connection layers.Based on DCNN,a data fusion algorithm based on convolutional neural network(DAECNN_attention)is proposed in this paper.In this method,noise reduction self encoder is introduced to realize data denoising and reconstruction;Self attention mechanism model is introduced to improve the ability of extracting key information.Experimental results show that compared with DCNN and BPNN data fusion algorithms,the proposed algorithm has better fusion performance on UCI data sets.
作者
孙淑娥
姚柳
赵怡
SUN Shue;YAO Liu;ZHAO Yi(School of Sciences,Xi’ an Shiyou University,Xi’ an,Shaanxi 710065,China;Xi’ an Thermal Power Research Institute Co.Ltd.,Xi’ an,Shaanxi 710054,China;School of Mathematics and Statistics,Xidian University,Xi’ an,Shaanxi 710126,China)
出处
《西安石油大学学报（自然科学版）》
北大核心
2021年第5期136-142,共7页
Journal of Xi’an Shiyou University（Natural Science Edition）
基金
国家自然科学基金(No.91338115)。
关键词
时间序列数据
数据融合
特征提取
一维卷积神经网络
time series data
data fusion
feature extraction
one dimensional convolutional neural network
