Improved XGBoost and GM UWB/MEME IMU Positioning Methods for Non-Line-of-Sight Environments

The non-line-of-sight (NLOS) in ultrawide-band (UWB) wireless positioning systems in complex indoor environments is problematic. Therefore, in this study, a hybrid algorithm based on particle swarm optimization (PSO)–genetic algorithm (GA) is proposed to optimize the NLOS recognition algorithm with extreme gradient boosting (XGBoost), and the gray prediction model (GM) is then used to correct the NLOS data. A classification feature system is constructed based on UWB data for XGBoost decision trees. On this basis, the gray relational analysis (GRA) algorithm is used to calculate the comprehensive importance measurement index of each feature, and these features are subsequently used to construct an effective feature subset and improve the accuracy of the NLOS recognition. The PSO algorithm has memory and tends to produce local optimal solutions, whereas the GA algorithm has good global convergence but no memory for individuals. Therefore, to prevent the PSO algorithm from falling into local optima, a hybrid algorithm, the PSO-GA, is adopted to optimize the parameters of the XGBoost model. The optimized XGBoost model is used for NLOS recognition in UWB systems and further improves the accuracy of NLOS recognition. For UWB data identified as NLOS data, the GM is used for correction to improve the utilization of the UWB measurement values. Then, the corrected UWB data are combined with the microelectromechanical system (MEMS) inertial measurement unit (IMU) using a tight coupling method to achieve combined positioning, which avoids possible multipath interference or signal attenuation problems when UWB positioning is used alone. The experimental results show that in complex indoor environments, the proposed algorithm can effectively recognize and correct the NLOS signals. The root mean square errors of the combined UWB/MEMS IMU positioning system in the X- and Y-directions are 0.107 and 0.076 m, respectively. Compared with the original UWB data and MEMS IMU combination results, the root mean square error in the X-direction decreases by 0.161 m, and the root mean square error in the Y-direction decreases by 0.085 m.