物联网环境下异步多传感器数据深度融合算法研究

Research on Asynchronous Multi sensor Data Deep Fusion Algorithm in IoT Environment

在线阅读下载PDF

导出

摘要在物联网环境中,现有方法未考虑异步多传感器数据融合过程中权重和偏置的计算,从而导致信息出现缺失,降低融合结果的质量。为了改善这个问题,提出了一种考虑引入权重和偏置计算的异步多传感器数据深度融合算法。首先采用经验小波变换方法对异步多传感器数据展开重构处理,提高数据质量;其次利用逐步回归特征选择方法选取出最有信息量的特征,以减少冗余信息降低维度;最后,通过计算选择特征在深度融合过程中的权重与偏置,并结合深度自动编码器网络(DAEN网络),完成对异步多传感器数据的深度融合。结果表明,所提算法均方误差可维持在1.0 dB以下,平均绝对百分比误差在3.5%以下,拟合度为0.96,融合耗时在8.5s以下,具有较好的融合效果和效率。 In the context of the Internet of Things,existing methods do not consider the calculation of weights and biases in asynchronous multi-sensor data fusion,resulting in missing information and reducing the quality of fusion results.To improve this issue,an asynchronous multi-sensor data deep fusion algorithm considering the introduction of weight and bias calculation is proposed.Firstly,the empirical wavelet transform method is used to reconstruct asynchronous multi-sensor data and improve data quality.Secondly,the stepwise regression feature selection method is used to select the most informative features to reduce redundant information and dimensionality.Finally,by calculating the weights and biases of selected features in the deep fusion process,and combining them with the deep automatic encoder network(DAEN network),the deep fusion of asynchronous multi-sensor data is completed.The results show that the proposed algorithm can maintain a mean square error of less than 1.0 dB,an average absolute percentage error of less than 3.5%,a fit degree of 0.96,and a fusion time of less than 8.5 seconds.It has good fusion performance and efficiency.

作者殷存举张薇 YIN Cunju;ZHANG Wei(Changzhou Liu Guojun Branch,Jiangsu Union Technical Institute,Changzhou Jiangsu 213025,China;School of Information and Software Engineering,East China Jiaotong University,Nanchang Jiangxi 330013,China)

机构地区江苏联合职业技术学院常州刘国钧分院华东交通大学信息与软件工程学院

出处《传感技术学报》北大核心 2025年第7期1321-1326,共6页 Chinese Journal of Sensors and Actuators

基金江西省教育厅科学技术研究项目(GJJ2200642)。

关键词异步多传感器数据融合经验小波变换方法逐步回归特征选择 DAEN网络 asynchronous multi-sensor data fusion empirical wavelet transform method stepwise regression feature selection DAEN network

分类号 TP273 [自动化与计算机技术—检测技术与自动化装置]

作者简介殷存举(1980-),男,汉族,山东费县人,中共党员,硕士,副教授,常州市学科带头人,常州市优秀教育工作者。主要研究方向:计算机软件技术、物联网应用技术,Yincj198003@163.com。