基于改进深度置信网络的UWB无线定位方法

针对传统指纹定位算法中接收信号强度值在室内复杂环境中波动较大，指纹信息不可靠，造成定位精度不足的问题，提出了一种以测距值作为指纹信息的基于深度置信网络和极限学习机的超宽带定位方法。首先在深度置信网络底层采用多个堆叠受限玻尔兹曼机对输入数据做无监督学习，来提取深层次特征，然后在顶层选用极限学习机对输入数据及位置标签进行有监督学习。建立指纹库阶段，为优化指纹采集过程并减少人工勘测成本，提出一种基于高斯过程回归的超宽带指纹库扩充方法。真实场景下实验结果显示，视距环境和非视距环境中，该定位方法均能够达到厘米级定位精度。

UWB Wireless Positioning Method Based on Improved Deep Belief Network

The received signal strength in the traditional fingerprint positioning algorithm fluctuates greatly in the complex indoors environment, which generates unreliable fingerprint information and results in insufficient positioning accuracy.We propose an ultra-wideband (UWB) fingerprint positioning method based on deep belief network (DBN) combined with extreme learning machine (ELM) using the range value as fingerprint information. Firstly, the multiple stacked restricted Boltzmann machines are used at the bottom of DBN to do unsupervised learning on the input data to extract deep features, and the ELM is used at the top layer to do supervised learning on the input data and location labels. In the offline fingerprint databse stage, a UWB fingerprint database expansion method based on Gaussian process regression is proposed to optimize the fingerprint acquisition process and reduce the cost of manual surveying. The experimental results show that the algorithm can achieve centimeter-level positioning accuracy in both line of sight (LOS) and non line of sight (NLOS) environments.