基于时空相关性的连续缺失值填补算法  

Imputation Algorithm of Continuous Missing Values Based on Temporal and Spatial Correlation

作　　者：蒋晨阳[1] 张云飞[1] 李鑫[1] 

机构地区：[1]河海大学计算机与信息学院,江苏南京211100

出　　处：《计算机与现代化》2016年第9期6-9,共4页Computer and Modernization

基　　金：国家科技支撑计划资助项目(2013BAB05B01)

摘　　要：无线传感器网络中的缺失数据对后续的数据分析带来很多不利影响,在数据分析之前,预处理工作必不可少。传感器网络数据在时间和空间方面均存在一定的变化规律,现有的缺失值填补算法往往只从单一角度分析解决问题,为了充分利用时空2个维度的特性,本文提出一种基于时空相关性的缺失值填补方法。该方法运用回归拟合、改进的BP神经网络等方法,对缺失数据进行填补。实验结果表明,该方法可以有效地提升缺失值填补的精度。On wireless sensor network, the missing data causes many difficulties in data analysis. Prior to data analysis, data pre- processing is necessary. Sensor network data has some change rules both in time and space. Existing imputation algorithms of missing values solve the problems only from a single point of view. In this paper, a imputation algorithm of missing values based on temporal and spatial correlation is proposed in order to make full use of the characteristics of space and time. It adopts regres- sion model and improves BP neural network to estimate the missing values. Experiments show that this method can improve the imputation accuracy of missing values effectively.

关 键 词：时空相关性 缺失值 传感器网络 BP神经网络 

分 类 号：TP309.2[自动化与计算机技术—计算机系统结构]