Research On Underground Positioning Technology Based On UWB/PDR

Traditional downhole positioning usually uses a single positioning technology.Due to the effects of underground dust,noise,and Non Line of Sight(NLOS)occlusion,a single positioning technology in the underground does not work well.In this background,downhole fusion positioning came into being.Among many positioning technologies,Ultra-wideband(UWB)positioning technology and Pedestrian Dead Reckoning(PDR)can complement each other's advantages and achieve better underground positioning.In UWB positioning technology and PDR positioning technology,UWB technology has good positioning accuracy under Line of Sight(LOS),can provide accurate initial coordinates for PDR technology.But in the case of NLOS,the positioning accuracy of UWB technology is reduced or even UWB technology is unavailable.PDR technology is not easily affected by the external environment,and the positioning accuracy is high in a short time,but another positioning technology is required to provide the initial coordinates and the cumulative error is serious over a long period of time.This thesis combines UWB technology and PDR technology to achieve downhole positioning.The specific research work is as follows:(1)Aiming at the problem of low accuracy of the fixed threshold peak detection method in different motion states,an adaptive threshold step frequency detection method based on PSO-GRU is proposed.First of all,this thesis selects three common sports states of slow walking,fast walking and running,and extracts the classification feature data from the three-axis acceleration modulus window difference.Particle Swarm Optimization(PSO)is used to optimize the network parameters of the Gated Recurrent Unit(GRU)neural network to further identify the movement state of the personnel and assign the peak threshold and time difference threshold.The experimental results show that the recognition rate of the PSO-GRU algorithm reaches 94.3%,and the accuracy of the adaptive threshold step counting method is much higher than that of the fixed threshold peak detection method.Aiming at the slow convergence speed of the heading angle calculation based on the gradient descent method,this thesis uses the Adam algorithm to dynamically update the step size of the gradient descent method.When the learning rate and initial value of the Adam algorithm are the same as the fixed step method,the dynamic step size method converges faster and the data is smoother.(2)Aiming at the problems of low accuracy of NLOS recognition based on single channel characteristic parameter and slow speed of NLOS recognition based on channel impulse response,a NLOS recognition algorithm based on GA-XGBoost is proposed.First,this thesis extracts multiple channel characteristic parameters from the signal and performs data preprocessing.Genetic Algorithm(GA)is used to optimize the training parameters of e Xtreme Gradient Boosting(XGBoost)and further realize NLOS recognition.The experimental results show that the NLOS recognition rate of GA-XGBoost algorithm reaches 91.7%,and the recognition effect is better than other machine learning algorithms.(3)Aiming at the problem of low accuracy and stability of single underground positioning technology,a UWB/PDR fusion positioning algorithm based on Extended Kalman Filter(EKF)is proposed.First,this thesis uses GA-XGBoost algorithm for NLOS recognition.If it is a LOS scene,set the UWB positioning coordinates to the initial value of the fusion positioning algorithm.If it is a NLOS scenario,remove the abnormal ranging value and perform error compensation for other ranging values.Finally,this thesis uses EKF to fuse UWB ranging and PDR motion information.The experimental results show that the fusion positioning algorithm in this thesis has higher positioning accuracy and stronger stability.There are 35 figures,10 tables,and 97 references in this thesis.