基于最小二乘支持向量回归机的无线传感器网络目标定位法

摘要

针对RSSI测距误差直接影响无线传感器网络(WSN)目标定位准确度的问题,从目标位置与目标到传感器节点测距矢量的双射关系入手,建立最小二乘支持向量回归机(LSSVR)目标定位的数学模型,提出了一种基于LSSVR的WSN目标定位方法TL-LSSVR.根据虚拟目标坐标和虚拟目标到传感器节点距离矢量构造出训练样本,通过确定学习区域及网格化采样获得训练样本集,采用LSSVR训练得到定位模型,将测量得到的距离矢量输入定位模型实现目标定位.对不同传感器节点数量以及不同节点分布情况下的WSN目标进行了定位实验.结果显示,对于节点随机分布的情况,TL-LSSVR方法的定位误差比最小二乘法减小21.0%～43.1%;对于节点均匀分布的情况,TL-LSSVR方法的定位误差则减小26.5%～48.7%,表明TL-LSSVR方法能有效减小测距误差对定位结果的影响,提高目标定位准确度.%In consideration of the effect of ranging errors of the RSSI method on the target localization accuracy in a Wireless Sensor Networks (WSN), a mathematical model of target localization based on Least Square Support Vector Regression (LSSVR) is established according to the double mapping between the target's coordinate and the distance vector measured from the target to sensor nodes. Furthermore, the target localization method based on LSSVR in the WSN,TL-LSSVR,is proposed. According to TL-LSSVR, the training samples are formed in accordance with the virtual target coordinate and the distance vector between the virtual target and the sensor nodes, and then the training sample sets are obtained by selecting learning areas and grid sampling. Moreover, the localization model can be trained using LSSVR and the target can be located by inputting the distance vector between the target and the sensor nodes into a localization model. The experiments of target localization in the WSN under different numbers and distributions of sensor nodes are performed. Experimental results show that when sensor node distributes randomly, the target localization errors using the TL-LSSVR are reduced by 21.0%-43.1% compared with that of a least square estimation,and when sensor node distributes uniformly, the target localization errors are reduced by 26.5%-48.7%,which indicates that the target localization errors are reduced evidently, and the accuracy of target localization is improved.