(1) Use Ray Tracing (RT) technology to simulate the UWB indoor 3D propagation model, and analyze the line of Sight (LOS) and Non Line of Sight (NLOS)scenarios to characterize the channel characteristics Compared with the traditional model parameters, it proves the feasibility of this simulation method; analyzes theinfluence of the NLOS environment on the ranging value, reveals the necessity of ranging error correction, and lays a foundation for the subsequent research on rangingerror correction Theory and data basis.(2) Based on the research of the UWB 3D propagation model, it can be seen that in the NLOS scene, the measurement error of the time difference of arrival (TDOA) isrelatively large. Aiming at this problem, a 3D TDOA positioning algorithm based on Back Propagation Neural Network (BPNN) error correction is proposed. BPNNregression algorithm is used to correct the TDOA measurement error, and then the three-dimensional TDOA classic positioning algorithm is analyzed through simulation,and the CHAN-TALOR algorithm with the highest accuracy is combined with BPNN to estimate the position coordinates of the target point to be located. Finally, simulationdata is used to verify the performance of the algorithm.(3) In the complex and dynamic indoor NLOS scene, the old fingerprint database cannot match the newly collected fingerprints due to environmental changes, and thetraditional fingerprint positioning algorithm has poor positioning accuracy. A SAE-RF-based three-dimensional fingerprint positioning method is proposed. First, a sparse autoencoder (SAE) network is used to obtain the estimated position coordinates of the target point to be located, and then a random forest (RF) regression model isadded. After slight changes in the environment, a few reference points are taken for positioning error correction. This algorithm uses UWB's more accurate ranging valueto replace the traditional signal strength value as the fingerprint amount, and reduces the maintenance and update cost of the fingerprint database, and weakens the impact ofenvironmental changes on fingerprint positioning. Finally, the performance of the algorithm is verified by experimental data.There are 35 Figures,7 tables, and 84 references in this paper.