Research On Mobile Phone Positioning In Urban Environment Integrating ANFIS Classifier And 3DMA

In recent years,the construction of GNSS(Global Navigation Satellite System)and the surge in the number of smartphone users have promoted the vigorous development of mobile-based location services.With the continuous progress of urbanization,the height and density of buildings in urban areas are increasing day by day,resulting in the decline of satellite positioning performance in high-density building environments.In the process of satellite positioning in urban areas,the satellite signals are affected by the occlusion of buildings,resulting in multi-path effects and the phenomenon of receiving NLOS(None-Line-Of-Sight),resulting in the inability of GNSS positioning accuracy in cities.At this stage,the emergence of urban 3D maps has brought new ideas for urban environment positioning,but there are still positioning limitations in practical applications.In order to improve the positioning performance of the current 3D map-assisted positioning technology in urban areas,this paper further optimizes the 3DMA(3D Mapping Aided)positioning algorithm and proposes PF-ML-LBR algorithm(Partial Filter and Machine Learning likelihood-based 3DMA ranging).this paper mainly focuses on NLOS judgment,scoring model optimization and algorithm efficiency optimization in the algorithm,and the research work in the following aspects is carried out:(1)Aiming at the problem of insufficient accuracy of NLOS judgment in 3DMA positioning technology,the NLOS discrimination model based on supervised learning is studied,and the satellite observations related to the NLOS signal features are extracted,combined with principal component analysis to reduce the dimension of the feature vector,and finally The NLOS classification models based CART(Classification And Regression Tree),SVM(Support Vector Machines)and ANFIS(Adaptive Network-based Fuzzy Inference System)were established respectively,and compared the NLOS discriminant effect of three different models is shown.The results show that In the experiment under the same environment as the training set,the NLOS classification accuracy based on ANFIS classifier is 95.66%,which is better than the 64.56% accuracy of CART and 88.03% accuracy of SVM.In experiments in different environments from the training set,the NLOS classification accuracy based on the ANFIS classifier is 82.56%,which is still better than the 60.41%correct rate of the CART decision tree and the 78.93% correct rate of the SVM.(2)In order to solve the problems of unreliable score value of 3DMA location technology in candidate area matching and multiple high score areas in urban environment,MLE-LBR(Machine Learning Enhanced likelihood-based 3DMA ranging)is proposed.For the problem of unreliable scoring value due to inaccurate NLOS judgment in matching scoring,the ANFIS-based NLOS classifier is used to further enhance the accuracy of NLOS judgment and enhance the reliability of the scoring value of candidate regions.For the inaccurate location of high score in the scoring map,by comparing the scoring maps of likelihood based LBR(Likelihoodbased 3DMA ranging)and SDM(shadow matching)in different occlusion environments,and using the complementarity of the scoring values of the two algorithms in the candidate areas,a joint scoring optimization scheme is proposed.Combining the above two solutions,a new algorithm framework MLE-LBR is proposed.Finally,by testing the MLE-LBR algorithm in different scenarios,the positioning accuracy is improved by 33.11% in the horizontal direction compared with LBR.(3)Aiming at the low computational efficiency of candidate point-by-point matching in 3DMA positioning technology,PF-ML-LBR algorithm is proposed.PFML-LBR uses importance sampling particles to simulate candidate positions in the initialization stage,which reduces the amount of calculation in the positioning process.In terms of positioning accuracy,in order to further avoid the problem of unreliable particle score values,a joint particle transfer weight calculation system fused with Mahalanobis distance is constructed,and the positioning solution is finally determined by Monte Carlo sampling.Static and dynamic experiments show that the longer the observation time is,the more the operation efficiency of PF-ML-LBR will be improved.In the two-hour static observation data,the computing load of PF-MLLBR is reduced by 82.54% and the computing time is reduced by 53.48% compared with LBR.Compared with the positioning accuracy of LBR algorithm,the positioning accuracy in the East and north directions is improved by 61.19% and 72.98%.PFML-LBR further enhances the positioning stability and computational efficiency of the algorithm in urban environments.