无线传感器网络中数据流异常数据的联合估计算法

A United Estimation Algorithm for the Outlier Value of Data Stream in Wireless Sensor Network

为保证无线传感器网络数据的完整性,针对数据流中存在异常数据的问题,提出了一种基于BP神经网络和多元线性回归的联合估计算法。首先,将存在异常数据的数据流作为样本输入,利用神经网络的非线性拟合能力对异常数据进行估计。然后,通过相邻的传感器节点数据建立多元线性回归模型,对异常数据进行估计。最后,根据两种算法在不同情况下的误差大小,调整它们各自在异常数据估计中的权重,计算出最接近真实值的估计值。以Berkeley Intel实验室的传感器数据为实验数据,通过Matlab软件对本文方法进行测试并分析仿真结果,实验结果表明文中提出的方法能对异常数据进行有效估计,并且具有较高的可靠性和稳定性。

Aiming at the problem of outlier value in data stream, a united estimation algorithm based on neural network and multiple linear regression is proposed to confirm the integrity and accuracy of data in wireless sensor networks. Firstly ,the nonlinear fitting ability of neural network is utilized to estimate outlier value by taking the data stream including outlier value as sample inputting. Then,the outlier value is estimated by building multiple linear regression model according to adjacent sensors＇ data. At last,the usage weights of both algorithms in the united estimation algorithm for estimation of outlier value are adjusted according to errors in different conditions, and the estimation values which ace nearest to real ones are worked out. The Berkeley Intel lab＇ s sensor data streams are adopted as sample data to test and analyze the simulation results by software Matlab. Experimental results show the method proposed can estimate outlier value effectively, which is of great reliability and stability.