基于神经网络的传感器网络数据融合技术研究

摘要

研究传感器网络数据融合优化问题,由于采集数据过程各节点汇集存在大量的冗余信息,需通过融合,提高采集效率.针对传统的数据融合算法需要获得对象比较精确的数学模型,对于复杂难于建立模型的场合无法适用.为解决上述问题,提出了一种BP神经网络传感器网络数据融合方法,可对对象的先验要求不高,具有较强的自适应能力.首先建立三层网络结构,接着提取数据库中属性数据的特征值并作为网络的输入,然后通过调节输入向量与中心向量的距离及中心向量的值确定网络权值,最后对数据进行有效融合,仿真结果表明,通过对有损数据融合,无损数据融合相比较,得出采用BP神经网络对传感器数据进行融合处理,输出输入稳定简单,是一种有效的数据融合处理方法.%Optimization of sensor network data fusion was studied. Traditional sensor networks for data fusion algorithms need precise object model, and it is difficult to establish a model for complex situations and other issues. A sensor network was presented based on BP neural network data fusion method, which was prior to the object is less demanding. With a strong adaptive ability, the algorithm layer network structure was established, then the characteristics of attribute data in the database were extracted as the network input values, and then, by adjusting the input vector, center vector and center vector of the distance, the value of network weights were determined. Finally, simulation results show that compared the damage through data fusion with non-destructive data fusion, BP neural network can obtain input - output stability , and is an effective data fusion method.