基于核自适应滤波的无线传感网络定位算法研究

针对动态室内环境的变化及时变的接收信号强度(Received signal strength,RSS)对定位精度的影响,提出了一类基于核自适应滤波算法的农业无线传感器网络室内定位方法。核自适应滤波算法具体包括量化核最小均方(Quantized kernel least mean square,QKLMS)算法及固定预算(Fixed-budget,FB)核递推最小二乘(Kernel recursive least-squares,KRLS)算法。QKLMS算法基于一种简单在线矢量量化方法替代稀疏化,抑制核自适应滤波中径向基函数结构的增长。FB-KRLS算法是一种固定内存预算的在线学习方法,与以往的滑窗技术不同,每次时间更新时并不修剪最旧的数据,而是旨在修剪最无用的数据,从而抑制核矩阵的不断增长。通过构建RSS指纹信息与物理位置之间的非线性映射关系,核自适应滤波算法实现WSN的室内定位,将所提出的算法应用于仿真与物理环境下的不同实例中,在同等条件下,还与其他核学习算法、极限学习机(Extreme learning machine,ELM)等定位算法进行比较。仿真实验中2种算法在3种情形下的平均定位误差分别为0.746、0.443 m,物理实验中2种算法在2种情形下的平均定位误差分别为0.547、0.282 m。实验结果表明,所提出的核自适应滤波算法均能提高定位精度,其在线学习能力使得所提出的定位算法能自适应环境动态的变化。

Wireless Sensor Network Location Algorithms Based on Kernel Adaptive Filtering

For the change of dynamic indoor environment and the effect of time-varying received signal strength on positioning accuracy, a class of indoor positioning algorithms for agricultural wireless sensor networks using kernel adaptive filtering was proposed, which included quantized kernel least mean square (QKLMS) as well as fixed-budget kernel recursive least-squares (FB-KRLS) algorithm. The QKLMS algorithm used a simple vector quantization approach as an alternative of sparsification to curb the growth of the radial basis function structure in kernel adaptive filtering. The FB-KRLS algorithm was an online kernel method by fixed memory budget, which was capable of recursively learning nonlinear mapping and tracking change over time. In contrast to a previous sliding-window based technique, the presented algorithm did not prune the oldest data point in every time instant but it was aimed to prune the least significant data point, thus suppressing the growth of kernel matrix. The kernel adaptive filtering algorithms achieved the indoor positioning for WSNs by building the non-linear mapping relations between the RSS fingerprint information and the physical location. The employed algorithms were applied to different indoor positioning instances in the simulation and physical environments for WSNs, under the same circumstances, compared with other kernel-based learning methods and extreme learning machine (ELM) etc. In the simulation experiment, the average localization error of the two algorithms was respectively 0.746m and 0.443m under three scenarios, and the average localization error of the two algorithms in the physical experiments was respectively 0.547m and 0.282m under two scenarios. Experimental results showed that the proposed adaptive filtering algorithms can improve the positioning accuracy, and its online learning ability made the proposed two localization algorithms all adaptable to the dynamic changes of the environments.