Search

CN-121978732-A - Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching

CN121978732ACN 121978732 ACN121978732 ACN 121978732ACN-121978732-A

Abstract

The invention discloses a Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching, and belongs to the technical field of satellite navigation and wireless communication fusion positioning. The intelligent indoor and outdoor seamless switching system comprises a main control module, a Beidou positioning module, a 5G communication positioning module, an IMU inertial measurement module, a storage module, a power module and a positioning result output module, wherein the main control module comprises a data preprocessing unit, a three-level vertical feature extraction unit, a fusion positioning decision model unit and an indoor and outdoor seamless switching execution unit. Through a depth feature extraction system of a three-level vertical processing architecture, progressive excavation from shallow single-source features to middle-layer scene association features and then to deep switching decision features is realized, and the vertical processing rules of the second-level features based on the first-level and the third-level features based on the second-level are strictly followed, so that cross-level feature interference is avoided, the depth and accuracy of multi-source data feature excavation are greatly improved, and high-reliability feature support is provided for scene identification and switching decision.

Inventors

  • WU BAINAN

Assignees

  • 江苏北斗信创检验检测有限公司

Dates

Publication Date
20260505
Application Date
20260403

Claims (10)

  1. 1. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching is characterized by comprising a main control module, a Beidou positioning module, a 5G communication positioning module, an IMU inertial measurement module, a storage module, a power module and a positioning result output module; the main control module comprises a data preprocessing unit, a three-level vertical feature extraction unit, a fusion positioning decision model unit and an indoor and outdoor seamless switching execution unit; The Beidou positioning module, the 5G communication positioning module and the IMU inertial measurement module are used for acquiring multi-source positioning original data in real time and transmitting the multi-source positioning original data to the main control module; the storage module is used for storing a historical positioning data set, model training parameters and an inference model file; the power module is used for providing stable power supply for the terminal; The positioning result output module is used for outputting the final fusion positioning result and positioning state information; The terminal realizes Beidou 5G fusion positioning of indoor and outdoor seamless switching through the following steps: S1, multi-source historical data acquisition and preprocessing, namely acquiring historical positioning original data, constructing a historical positioning data set, and carrying out standardized preprocessing on the data set to obtain a standardized characteristic data set; S2, deep feature extraction of a three-level vertical processing architecture, namely constructing the three-level vertical processing architecture, carrying out deep feature extraction on a standardized feature data set, generating second-level features based on the first-level feature extraction, and generating third-level features based on the second-level feature extraction; s3, training a fusion positioning and switching decision model, namely constructing the fusion positioning and switching decision model based on the extracted depth feature set, and designing a joint loss function to complete training, verification and optimization of the model; s4, real-time positioning and seamless switching are carried out, namely, a terminal acquires multi-source positioning original data in real time, inputs a solidified reasoning model after preprocessing, extracts real-time depth features through three-level vertical features, outputs a positioning mode decision result and a fused positioning solution value, and completes switching of a positioning mode and outputting of a positioning result through a seamless switching execution unit.
  2. 2. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching of claim 1 is characterized in that in the step S1, historical positioning original data comprises Beidou observation data, 5G positioning observation data and IMU inertial measurement data; The Beidou observation data comprises a pseudo range, a carrier phase, a carrier-to-noise ratio, a visible satellite number, a satellite elevation angle, a satellite azimuth angle and a PDOP value, wherein the 5G positioning observation data comprises reference signal receiving power RSRP, reference signal receiving quality RSRQ, a time advance TA, an arrival angle AoA, an departure angle DoA, a visible base station number and channel state information CSI; the standardized pretreatment specifically comprises the following steps: S11, time synchronization alignment, namely performing time stamp alignment on Beidou and 5G, IMU data with different sampling frequencies by taking Pulse Per Second (PPS) of a Beidou system as a time reference, and unifying all the data to a time sampling reference of 100Hz by adopting a linear interpolation method to realize time-space synchronization of multi-source data; S12, outlier elimination and restoration, namely eliminating outliers in multi-source data by adopting a3 sigma criterion, detecting cycle slip by adopting an M-W combination method aiming at Beidou observation data, completing cycle slip restoration by adopting second-order polynomial fitting, and eliminating signal mutation values caused by multipath effects by adopting sliding window median filtering aiming at 5G observation data; S13, carrying out data normalization processing, namely linearly mapping feature data of all dimensions to a [0,1] interval by adopting a min-max normalization method, and eliminating dimension differences of different features to obtain a standardized feature data set; And S14, dividing the standardized characteristic data set into a training set, a verification set and a test set according to the proportion of 7:2:1, and marking each group of data with a corresponding real scene label, a real positioning error label and a real positioning mode label, wherein the scene labels comprise an outdoor scene, an indoor scene and an indoor and outdoor transition scene, and the positioning mode labels comprise a Beidou dominant outdoor mode, a 5G dominant indoor mode and a mixed fusion mode.
  3. 3. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching according to claim 1, wherein in the step S2, a three-level vertical processing architecture comprises a first-level shallow global feature extraction layer, a second-level middle-level scene association feature extraction layer and a third-level deep switching decision feature extraction layer, a three-level structure adopts a serial cascade mode, and the only input of the latter level is the output feature of the former level, so that vertical progressive feature extraction is realized.
  4. 4. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching according to claim 3, wherein the first-stage shallow global feature extraction layer is a multi-source data primary characterization layer, and the input of the first-stage shallow global feature extraction layer is a preprocessed standardized feature data set, and the specific extraction process is as follows: S211, constructing single-source characteristic independent coding branches, respectively setting Beidou characteristic branches, 5G characteristic branches and IMU characteristic branches, wherein the three branches adopt a parallel one-dimensional convolutional neural network 1D-CNN structure, and perform independent primary characteristic coding on corresponding single-source data; S212, each single-source characteristic branch adopts 3 layers of serial 1D-CNN units, each layer of 1D-CNN unit consists of a convolution layer, a batch normalization BN layer and a ReLU activation function layer in sequence, wherein the convolution kernel of the first layer is 7, the step size is 1, the filling is 3 and the output channel number is 32, the convolution kernel of the second layer is 5, the step size is 1, the filling is 2 and the output channel number is 64, and the convolution kernel of the third layer is 3, the step size is 1, the filling is 1 and the output channel number is 128; S213, after the third layer 1D-CNN unit of each single-source characteristic branch is output, the GAP layer is accessed into the global average pooling GAP layer, and the time sequence characteristics of convolution output are subjected to dimensional compression to obtain 128-dimensional single-source primary characteristics of each branch; s214, performing channel dimension splicing on single-source primary features output by the Beidou feature branch, the 5G feature branch and the IMU feature branch to obtain 384-dimensional first-stage shallow global features, namely F1 features, and completing first-stage feature extraction.
  5. 5. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching according to claim 4, wherein the second-stage middle-layer scene association feature extraction layer is a scene perception feature enhancement layer, and the unique input of the second-stage middle-layer scene association feature extraction layer is an F1 feature output by a first-stage, and the specific extraction process is as follows: s221, time sequence dependency modeling, namely inputting F1 features into a Bi-LSTM (two-way long-short-term memory network), capturing the front-rear association relation of the features on a continuous time sequence, and fitting the feature change trend in the indoor and outdoor scene switching process, wherein the Bi-LSTM is provided with a 2-layer serial structure, each hidden layer has a dimension of 256, and a tanh activation function is adopted to output 512-dimensional time sequence association features; S222, attention weighting enhancement, namely inputting time sequence related features output by Bi-LSTM into a multi-head self-attention module to perform self-adaptive weight distribution on feature dimensions, wherein the multi-head self-attention module is provided with 8 parallel attention heads, the dimension of each head is 64, feature dimension weights which are strongly related to indoor and outdoor scene recognition and positioning accuracy are enhanced through scaling dot product attention calculation, noise feature and irrelevant feature weights are restrained, and 512-dimensional attention weighted features are output; S223, residual fusion optimization, namely inputting the attention weighted feature into a 1X 1 convolution dimension mapping layer, wherein the convolution kernel of the layer is 1, the step length is 1, the number of output channels is 384, linearly mapping the attention weighted feature of 512 dimensions into 384 dimensions which are consistent with the dimension of the F1 feature which is originally input; S224, inputting residual characteristics into a full-connection layer, mapping characteristic dimensions to 256 dimensions, and carrying out standardization processing on the characteristics through a layer normalization LN layer to finally obtain second-stage middle-layer scene association characteristics, namely F2 characteristics, and finishing second-stage characteristic extraction.
  6. 6. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching of claim 5 is characterized in that the third-stage deep switching decision feature extraction layer is a seamless switching decision feature mapping layer, the only input of the third-stage deep switching decision feature extraction layer is the F2 feature output by the second stage, and the specific extraction process is as follows: S231, constructing positioning mode matching branches, namely setting three parallel fully-connected neural network MLP branches which respectively correspond to a Beidou dominant outdoor mode, a 5G dominant indoor mode and a mixed fusion mode, wherein the three branches are completely consistent in structure and are all an input layer, a first hidden layer, a second hidden layer, a Dropout layer and an output layer, wherein the input layer is 256 and is matched with F2 characteristic dimensions, the first hidden layer is 128, a ReLU activation function is adopted, the second hidden layer is 64, a ReLU activation function is adopted, the Dropout layer deactivation rate is set to 0.3, the output layer is 1, and the matching degree original score corresponding to the positioning mode is output; S232, normalizing the mode probability, namely inputting the original scores output by the three mode matching branches into a Softmax function, and carrying out probability normalization processing to obtain probability distribution corresponding to the three positioning modes, wherein the probability distribution is respectively a Beidou dominant outdoor mode probability, a 5G dominant indoor mode probability and a mixed fusion mode probability, and the sum of the three probabilities is 1; S233, constructing a positioning accuracy prediction branch, namely setting an independent positioning accuracy prediction MLP branch, inputting the branch as an F2 characteristic, wherein the branch structure is an input layer, a hidden layer 1, a hidden layer 2, a Dropout layer and a multi-output layer, wherein the input layer dimension is 256 and is matched with the F2 characteristic dimension, the first hidden layer dimension is 128, a ReLU activation function is adopted, the second hidden layer dimension is 64, a ReLU activation function is adopted, the Dropout layer deactivation rate is set to 0.3, the multi-output layer dimension is 2, and the positioning accuracy prediction values under the dominant mode are respectively output Positioning accuracy prediction value in 5G dominant mode The units are rice; S234, fusing and mapping decision features, namely performing channel splicing on the mode probability distribution obtained in the step S232 and the two positioning precision predicted values obtained in the step S233 to obtain a 5-dimensional fused decision vector, inputting the fused decision vector into a full-connection layer to perform high-dimensional feature mapping, finally obtaining third-level deep switching decision features, marking the third-level deep switching decision features as F3 features, and finishing third-level feature extraction.
  7. 7. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching of claim 1 is characterized in that in the step S3, a fusion positioning and switching decision model is an end-to-end deep learning model, the input is preprocessed standardized multi-source characteristic data, and the output is a positioning mode decision result, a multi-source data fusion weight and a final fusion positioning result; the model main body is a three-level vertical processing architecture of the step S2, the rear end of the model main body is connected with a loose combined type federal Kalman filtering resolving unit, the federal Kalman filtering resolving unit comprises 1 main filter and 3 parallel sub-filters (Beidou sub-filter, 5G sub-filter and IMU sub-filter), the sub-filters independently complete Kalman filtering resolving of each single source positioning data, the main filter performs federal fusion on resolving results of the sub-filters based on self-adaptive fusion weights output by the model, and a globally optimal fusion positioning result is output; in the model training process, a joint loss function is adopted as an objective function of model optimization, and the joint loss function Cross entropy loss by scene classification Regression loss of positioning accuracy Smooth loss of handover Three-part weighted composition, scene classification cross entropy loss Model-output-based mode probability distribution and real scene tag calculation for optimizing scene recognition and mode matching accuracy, wherein the positioning accuracy is lost by regression The multi-output mean square error MSE is adopted, and the model-output-based calculation of the sub-mode positioning precision predicted value and the real positioning error label is used for optimizing the accuracy of the positioning precision prediction, wherein the switching smoothing loss Based on the mode probability distribution calculation of the mode probability change gradient and the continuous time sequence, only nonsensical frequent jumps are punished, and the probability gradual change with trend is not punished; and (3) model training adopts a AdamW optimizer, adopts a cosine annealing attenuation strategy to adjust the learning rate, adopts an early stop system based on the accuracy of the verification set in the training process, stops training when the loss of the verification set is continuously 10 rounds without reduction, and stores optimal model parameters to obtain a solidified reasoning model.
  8. 8. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching of claim 7 is characterized in that core resolving details of the federal Kalman filtering resolving unit are as follows: state vector definition the state vectors of the sub-filters and the main filter are set uniformly Wherein the method comprises the steps of For the position coordinates of the terminal in three-dimensional space, The motion speed of the terminal in the three-dimensional space is given; the sub-filter observation equation: The Beidou sub-filter observation equation: wherein The Beidou positioning observation matrix is constructed by Beidou pseudo-range and carrier phase observation values, For Beidou observation noise, obeying Gaussian distribution with mean value of 0 and variance; The 5G sub-filter observation equation: wherein For a 5G positioning observation matrix, constructed from 5GAoA+TA observations, Observing noise, subject to mean 0 and variance 0 Is a gaussian distribution of (c); IMU sub-filter observation equation: wherein An IMU inertial measurement observation matrix is constructed by triaxial acceleration and angular velocity integral values, For IMU observation noise, the obeying mean value is 0 and the variance is Is a gaussian distribution of (c); Main filter fusion calculation formula, self-adaptive fusion weight output by main filter based on model (Satisfy the following ) Optimal estimation value for each sub-filter Weighted fusion is carried out to obtain a global optimal fusion positioning result: 、 The main filter updates the covariance matrix of the fusion result: 、 Wherein the method comprises the steps of The covariance matrix of each sub-filter is obtained.
  9. 9. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching of claim 8 is characterized in that a calculation formula of the joint loss function is as follows: 、 Wherein, the Is a weight coefficient; Regression loss of positioning accuracy The calculation formula of (2) is as follows: a true positioning error label corresponding to the mode; handoff smoothing loss The calculation formula of (2) is as follows: ; Is the mode probability distribution vector at time t, A mode probability distribution vector at the time t-1; for the mode stability factor, the calculation formula is: 、 is the gradient of the variation of the probability distribution of the mode at the moment t, , Is a preset gradient threshold.
  10. 10. The Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching according to claim 1, wherein in the step S4, real-time positioning and seamless switching execution specifically comprises the following steps: S41, real-time data acquisition and preprocessing, namely acquiring multisource positioning original data at the current moment in real time by a terminal through a Beidou positioning module, a 5G communication positioning module and an IMU inertial measurement module, and completing the standardized processing of the real-time data according to the preprocessing flow of the step S1 to obtain real-time standardized characteristic data; s42, real-time depth feature extraction, namely inputting real-time standardized feature data into a solidified reasoning model, and completing progressive extraction from F1 features and F2 features to F3 features sequentially through a three-level vertical processing architecture to obtain real-time depth features; S43 model reasoning and decision output, namely outputting probability distribution of three positioning modes at the current moment and positioning precision predicted values of the Beidou dominant mode based on real-time depth characteristics 5G dominant mode positioning accuracy predictor And the fusion weight of Beidou and 5G, IMU multisource data and the fusion positioning result based on federal Kalman filtering solution; S44, positioning mode switching decision is carried out by adopting a composite rule of probability threshold value, gradual change trend and mode dividing precision based on a model output result, wherein the specific decision rule is that a preset basic probability threshold value is as follows The gradual trend trigger threshold is A continuous frame threshold value N and a first positioning accuracy threshold value of The second positioning accuracy threshold value is The weight threshold value is C; the outdoor scene-Beidou dominant mode can be judged by meeting any one of the following conditions, (1) the probability of the Beidou dominant outdoor mode is more than or equal to And (2) and (2) The probability of the Beidou dominant outdoor mode is continuous N frames or more And has monotonous rising trend ≤ After judgment, taking the Beidou pseudo-range differential positioning result as a main part, carrying out auxiliary correction on 5G and IMU data, wherein the Beidou data fusion weight is not lower than C; the indoor scene-5G dominant mode can be judged by meeting any one of the following conditions, (1) the probability of the 5G dominant indoor mode is more than or equal to And (2) and (2) Probability of continuous N frames of 5G dominant indoor mode And has monotonous rising trend After the determination, using a 5GAoA+TA positioning result as a main part, carrying out IMU data auxiliary correction, wherein the 5G data fusion weight is not lower than C; An indoor and outdoor transition scene-mixed fusion mode, wherein when the outdoor and indoor scene judging conditions are not met, the indoor and outdoor transition scenes are judged, and based on the self-adaptive fusion weights output by the model, the positioning resolving results of Beidou and 5G, IMU are optimally fused through federal Kalman filtering; S45, seamless switching smooth transition, namely in the process of switching the positioning modes, taking a final fusion positioning result at the previous moment as a priori state value of Kalman filtering at the current moment, carrying out smooth constraint on a positioning resolving result at the current moment, and simultaneously, linearly adjusting fusion weights of multi-source data in a transition window period of a preset duration, avoiding jump of the positioning result and realizing seamless switching of indoor and outdoor scenes; and S46, outputting a positioning result, namely outputting a final fusion positioning result, a current positioning mode, a positioning precision estimated value and positioning state information in real time through a positioning result output module.

Description

Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching Technical Field The invention relates to the technical field of satellite navigation and wireless communication fusion positioning, in particular to a Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching. Background With the comprehensive establishment of the Beidou No. three global satellite navigation system and the large-scale deployment of the 5G communication network, beidou+5G fusion positioning becomes a core technical scheme for realizing indoor and outdoor full-scene seamless positioning. The Beidou satellite navigation system has wide-area and high-precision outdoor positioning capability, but has remarkable satellite signal serious attenuation and multipath effect in indoor, building shielding, underground space and other scenes, and cannot provide stable and reliable positioning service, and the 5G communication network has the characteristics of indoor dense coverage, large bandwidth, low time delay, and can realize high-precision indoor positioning by AoA, TA, PRS and other technologies, but has insufficient base station coverage density in an outdoor wide-area scene, and the positioning precision and stability are far lower than those of the Beidou system. In the prior art, a hard switching mechanism based on a fixed signal threshold is mostly adopted by the Beidou 5G fusion positioning terminal, and the switching of an outdoor Beidou mode and an indoor 5G mode is realized by setting a Beidou carrier-to-noise ratio and a fixed threshold of 5 GRSRP. The scheme has the core defects that firstly, a switching mode of a fixed threshold value is poor in adaptability to complex indoor and outdoor transition scenes (such as building inlets, underground garage entrances and exits, under overhead bridges and urban canyons), signal fluctuation easily causes frequent switching of positioning modes to generate ping-pong effect and cause jump or even interruption of positioning results, secondly, the prior art is insufficient in characteristic mining of Beidou and 5G multi-source data, mostly only adopts basic statistical characteristics to perform scene judgment and fusion positioning, deep association and time sequence change characteristics between the multi-source data cannot be captured, scene recognition accuracy is low, reliable support cannot be provided for switching decisions, thirdly, the existing fusion positioning model cannot simultaneously consider scene recognition accuracy, positioning error prediction and switching smoothness, positioning continuity and stability are insufficient in the indoor and outdoor switching process, real seamless switching is difficult to realize, and the requirements of intelligent inspection, automatic driving, emergency rescue, personnel management and control and other scenes on continuous high-precision positioning of the whole scene cannot be met. Disclosure of Invention The invention aims to provide the Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching, which can solve the problems that in the prior art, ping-pong effect is easy to occur in positioning mode switching, positioning result is jumped, complex transition scene identification accuracy is low, and positioning continuity and accuracy are insufficient. According to one aspect of the invention, the Beidou/5G fusion positioning terminal for indoor and outdoor seamless switching comprises a main control module, a Beidou positioning module, a 5G communication positioning module, an IMU inertial measurement module, a storage module, a power module and a positioning result output module; the main control module comprises a data preprocessing unit, a three-level vertical feature extraction unit, a fusion positioning decision model unit and an indoor and outdoor seamless switching execution unit; The Beidou positioning module, the 5G communication positioning module and the IMU inertial measurement module are used for acquiring multi-source positioning original data in real time and transmitting the multi-source positioning original data to the main control module; the storage module is used for storing a historical positioning data set, model training parameters and an inference model file; the power module is used for providing stable power supply for the terminal; The positioning result output module is used for outputting the final fusion positioning result and positioning state information; The terminal realizes Beidou 5G fusion positioning of indoor and outdoor seamless switching through the following steps: S1, multi-source historical data acquisition and preprocessing, namely acquiring historical positioning original data, constructing a historical positioning data set, and carrying out standardized preprocessing on the data set to obtain a standardized characteristic data set; S2, deep feature extraction of a three-level vertical processing architecture, namely co