基于神经网络的不可靠数据恢复研究

Research on Trustless Data Restoring Based on Neural Networks

智能系统中单点或少数传感器采集的数据在某一段时间出现不可靠问题,在装备有许多传感器的智能系统中普遍存在,即使在由先进的传感器构成的桥梁结构健康监测系统中,80%以上的虚假报警也是由于测量数据的不可靠性造成的。传统上对于不可靠数据的处理主要应用线性回归法、平均法等方法进行恢复,然而,大多数测量数据在时域上表现为非线性特征,传统方法恢复的数据在精度上是很难达到要求。以桥梁挠度数据作为研究对象,利用原始数据对挠度测量点进行了关联分析,并依据RBF神经网络强大的函数逼近能力,提出了一种基于神经网络模型来恢复不可靠测量数据的方法,并在仿真实验中,通过对比实验(该方法的均方误差为2E-9,线性回归法均方误差为0.6974)证实了该方法在理论和实践上的精确性和可行性。

It is a general problem of the Intelligent System that the trustless data are obtained by the single or a few sensors some time. Despite the acquisition system of the bridge structural health monitoring system （BSHMS） is designed by advanced sensors, the abnormity deflection occurred in the acquisition system is the main reason for the illusive alarm（above 80% ）. A novel method based on the correlation analysis of bridge＇s checking points and the RBF neural networks is proposed for restoring nonlinear deflection abnormity data. Compared with conventional methods（ its MSE is 2e -9）, the proposed approach （its MSE is 0. 6974） is more accurate and accords with practice. Simulation results verify the effectiveness of the designed method.