基于BP神经网络的厘米级超宽带测距误差改正模型设计与实验  

Design and experiment of centimeter UWB ranging error correction model based on BP neural network

作　　者：肖庭杨 章迪[1] XIAO Tingyang;ZHANG Di(School of Geodsy and Geomatics,Wuhan University,Wuhan 430079,China)

机构地区：[1]武汉大学测绘学院,武汉430079

出　　处：《全球定位系统》2021年第4期52-58,共7页Gnss World of China

基　　金：国家自然科学基金项目(41604019);武汉大学实验技术项目(WHU-2021-SYJS-15);国家级大学生创新创业训练计划项目(202010486084)。

摘　　要：在室内复杂环境下,超宽带(UWB)测距误差难以通过常规方法进行有效补偿,严重制约了其定位精度.在分析室内环境下UWB测距误差分布特点的基础上,设计了两种不同结构的BP神经网络误差改正模型.模型BP1输入单个标签与4个基站的测距值,输出对应的4个测距误差;模型BP2输入一对标签、基站的三维坐标,输出对应的一个测距误差.以高精度全站仪测量结果作为参考值,对网络进行训练,并对模型改正前后的测距和定位精度进行了对比分析.结果表明:两种模型均能有效改正测距误差,提升定位精度.其中BP1测距和定位精度分别提高83.0%、75.9%,BP2测距和定位精度平均提高91.7%、93.8%.BP2相比于BP1能够更加有效地提高测距和定位精度,使定位精度由dm级提升至cm级.In complex indoor environment,ultra wide band(UWB)ranging error can not be effectively compensated by conventional methods,which seriously restricts its positioning accuracy.Based on the analysis of the distribution characteristics of UWB ranging error in indoor environment,two different BP network error correction models with different structures are designed.Model BP1 inputs the ranging values of a single label and four base stations,and outputs four corresponding ranging errors;Model BP2 inputs the three-dimensional coordinates of a pair of labels and a base station,and outputs one corresponding ranging error.The network is trained with high precision total station measurement result as reference value,and the ranging and positioning accuracy before and after model correction are compared and analyzed.The results show that both models can effectively correct ranging errors and improve positioning accuracy.BP1 ranging and positioning accuracy are improved by 83.0%,75.9%,and BP2 ranging and positioning accuracy is improved by 91.7%,93.8%on average.BP2 can improve the ranging and positioning accuracy more effectively than BP1,and the positioning accuracy can be improved from decimeter level to centimeter level.

关 键 词：超宽带(UWB) 测距误差改正 BP神经网络 最小二乘定位 

分 类 号：P228.4[天文地球—大地测量学与测量工程]