基于极限学习机模型的粒子滤波无设备定位方法研究

在无线通信环境中，无线射频信号易受到干扰，强度波动较为明显。为定量描述和分析目标位置与无线信号强度之间复杂、多变的关系，更加精准地估计目标位置，提出了基于极限学习机（ELM）模型的粒子滤波（PF）无设备定位算法。该算法包括ELM模型构建（离线阶段）和目标位置估计（在线阶段）。在ELM模型构建阶段，建立目标在不同位置与链路（发射节点与接收节点之间的通信链路）信号强度变化的离线数据库，利用ELM构建目标位置与无线射频信号强度的映射关系。在目标位置估计阶段，通过映射关系结合PF实现目标位置的跟踪。实验结果表明，所提算法不仅有效地解决了目标位置与无线射频信号强度的映射关系，而且比高斯过程模型-PF、支持向量机-PF等现有算法显著提高了目标跟踪精度。

Research on Extreme Learning Machine Model-based Particle Filter Tracking Method for Device-free Localization

Device-free localization (DFL) is an emerging wireless technique for estimating the location of target. The radio frequency signal is seriously polluted due to the uncertainty of wireless channel. An extreme learning machine model-based particle filter (ELM-PF) algorithm for DFL is proposed. The proposed algorithm is used for ELM building （offline stage）and target position estimation (online stage). In the ELM building process, a radio frequency signal propagation model built by ELM is used to describe the mapping relationship between target position and radio signal strength indicator (RSSI). In the process of target position estimation, a target is tracked via particle filter with the radio frequency signal propagation model. Experimental results show that the proposed ELM-PF algorithm can eliminate the fluctuation of the wireless signals and be robust to the tracking accuracy compared with the existing Gaussian process model-particle filter (GPM-PF) and support vector machine-particle filter (SVM-PF).