High-precision PPP positioning method based on SSR correction prediction

Abstract

The invention belongs to the technical field of satellite positioning, and particularly discloses a high-precision PPP (Point-to-Point Protocol) positioning method based on SSR (Simple Sequence Repeat) correction prediction, which comprises the following steps that: a receiver receives and stores an SSR correction in real time to obtain an SSR sequence; inputting the stored SSR sequence into a pre-trained prediction model, and outputting an SSR correction number prediction value in a future time period; the broadcast ephemeris is corrected through the predicted SSR correction number, and a precise satellite orbit and clock correction are obtained; and constructing an ionosphere-free combination model based on the double-frequency observation value, and resolving the position of the receiver through Kalman filtering in combination with the corrected satellite orbit and clock correction. According to the method, long-time-sequence prediction is carried out on the SSR correction number through the prediction model, the problem of positioning precision jump caused by satellite correction data interruption is effectively solved, it is guaranteed that PPP positioning continuously outputs centimeter-level precision results under the scenes of signal shielding, network delay and the like, and the limitation that traditional PPP strongly depends on real-time SSR data is broken through.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of satellite positioning, and particularly relates to a high-precision PPP positioning method based on SSR correction number prediction. BACKGROUND

[0002] Precise Point Positioning (PPP) is an absolute positioning technology, and a user can achieve high-precision positioning of centimeter to decimeter level in a global range. PPP considers various errors in the signal transmission process and performs accurate correction according to precise satellite orbits and clock errors, and solves the problem by using a suitable parameter estimation method, so that high-precision position information can be obtained by a single receiver. PPP has the advantages of high precision, simple operation and low cost, and is currently widely used in fields such as geodetic surveying, weather monitoring and intelligent driving.

[0003] PPP is developing towards real-time, but its implementation depends on precise satellite orbit and clock error data, and at present, precise satellite orbit and clock error data are mainly obtained through SSR correction numbers provided by the International GNSS Service (IGS). The SSR correction number includes correction information of the orbit and clock error, and the user uses the SSR correction number to correct the orbit and clock error data calculated from the broadcast ephemeris to restore the precise orbit and clock error, thereby realizing high-precision positioning. The IGS relies on the RTCM network transmission protocol (The Networked Transport of RTCM via Internet Protocol, NTRIP) protocol to broadcast real-time data to global user terminals, but sometimes the continuity and stability of the data stream are relatively insufficient, mainly manifested as service-side instantaneous abnormalities or network fluctuations leading to data interruption. In this case, the user usually switches to a single-point positioning mode with a precision of several meters, or continues to use asynchronous correction for positioning, which is difficult to meet the application requirements of high timeliness and high precision. Therefore, how to maintain the positioning precision during the interruption of the SSR correction number has become a problem to be solved. SUMMARY

[0004] The purpose of the present application is to overcome the defects in the prior art and provide a high-precision PPP positioning method based on SSR correction number prediction.

[0005] The present application provides a high-precision PPP positioning method based on SSR correction number prediction, comprising the following steps:

[0006] (1) The user terminal receiver receives and saves the SSR correction number in real time to obtain an SSR sequence;

[0007] (2) When the SSR data is interrupted, the saved SSR sequence is input into the pre-trained prediction model, and a predicted value of the SSR correction number in the future period is output;

[0008] (3) The predicted SSR correction number is used to correct the broadcast ephemeris, and precise satellite orbit and clock error are obtained;

[0009] (4) A ionosphere-free combination model is constructed based on dual-frequency observations, combined with the corrected satellite orbit and clock error, and the receiver position is solved by Kalman filtering.

[0010] Further, the SSR sequence includes time information, orbit radial correction number, orbit tangential correction number, orbit normal correction number and clock error correction number.

[0011] Further, the prediction model adopts an improved iTransformer model, and the structure of the improved iTransformer model includes:

[0012] An embedding layer is configured to independently map the time series data of each variable in the input sequence into a feature vector;

[0013] A residual module is configured to perform deep feature extraction on the clock error correction number, and includes at least one residual block including a one-dimensional convolution layer;

[0014] A self-attention layer is configured to identify the correlation between variables;

[0015] A feedforward network layer is configured to generate a predicted sequence.

[0016] Further, the specific process of correcting the broadcast ephemeris includes:

[0017] The broadcast ephemeris is matched according to the time information of the SSR orbit correction number and the ephemeris data period number parameter;

[0018] The orbit correction number in the star-fixed system is converted into a ground-fixed system correction through a rotation matrix;

[0019] The broadcast ephemeris coordinates and the ground-fixed system correction are combined to obtain the precise position of the satellite.

[0020] Further, the process of obtaining the precise satellite orbit and clock error includes:

[0021] The predicted orbit correction number in the star-fixed coordinate system is converted into the geocentric and ground-fixed coordinate system through a rotation matrix, and is combined with the broadcast ephemeris to obtain the precise position of the satellite;

[0022] Based on the quadratic polynomial model ΔC=a0+a1(t-t0)+a2(t-t0) 2The satellite clock error is corrected, and a precise clock error sequence is generated by combining the broadcast ephemeris clock error data, wherein a0, a1 and a2 are clock error correction parameters of the SSR, are determined by fitting historical clock error data through a least square method, and are updated in real time; t is a current epoch time, t0 is a reference epoch, and the satellite clock error at the time t after correction is:

[0023]

[0024] In the formula, dt brdc represents the satellite clock error calculated according to the broadcast ephemeris, and c represents the speed of light.

[0025] Further, the ionosphere-free combination model is represented as:

[0026]

[0027] , wherein, represents the ionosphere-free combination pseudo-range observation value from the receiver r to the satellite s, represents the geometric distance from the receiver r to the satellite s, and c is the speed of light; dt r represents the clock error accumulation error of the receiver local oscillator caused by temperature drift, crystal aging and other factors, and dt s,Q represents the clock frequency deviation of the satellite atomic clock caused by the relativistic effect and the influence of the space environment during on-orbit operation; and respectively represent the uncorrected pseudo-range hardware delay at the receiver end and the satellite end; and represent the uncorrected phase hardware delay at the receiver and satellite ends; T r s,Q represents the delay of the troposphere; is and combined wavelength; and respectively represent the pseudo-range and carrier phase observation noise of the model.

[0028] Further, the receiver position calculation process is:

[0029] The receiver position calculation process is:

[0030] (a) linearizing the observation equation of the ionosphere-free combination model to obtain the error equation of the residual vector, the coefficient matrix, the to-be-estimated parameter vector and the observation vector;

[0031] (b) when the SSR data is interrupted, the error equation is solved by using a least square method to obtain the initial coordinates and the initial clock error of the receiver;

[0032] (c) taking the initial coordinates and initial clock difference as the initial value of the state vector of Kalman filtering, combining the double-frequency observation value sequence to perform dynamic parameter estimation;

[0033] (d) iteratively updating the state vector by Kalman filtering, and outputting the final precise coordinates of the receiver.

[0034] Compared with the prior art, the beneficial effects of the present application are that:

[0035] According to the prediction model, the present application predicts the SSR data in a future period of time, corrects the broadcast ephemeris by using the predicted orbit and clock correction, obtains accurate satellite orbit and clock information, constructs a double-frequency ionosphere-free combination based on double-frequency observation values to eliminate ionospheric delay errors, and estimates the position error of the receiver by using Kalman filtering to obtain accurate position coordinates. By fusing the iTransformer model of the residual network and the multi-head self-attention, the long-time sequence prediction of the SSR correction number is performed, the positioning accuracy jump problem caused by the interruption of satellite correction data is effectively solved, the continuous output of centimeter-level precision results of PPP positioning in the scenes such as signal shielding and network delay is ensured, and the strong dependence of traditional PPP on real-time SSR data is broken through. BRIEF DESCRIPTION OF DRAWINGS

[0036] The following drawings only schematically illustrate and explain the present application, and do not limit the scope of the present application, in which:

[0037] Figure 1 : schematic diagram of the positioning method of the present application;

[0038] Figure 2 : iTransformer prediction model based on residual network;

[0039] Figure 3 : positioning error when the SSR data of the CUSV station is interrupted. DETAILED DESCRIPTION

[0040] In order to make the purpose, technical scheme, design method and advantages of the present application more clear and explicit, the present application is further described in detail below by combining the drawings with specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application, and do not limit the present application.

[0041] As shown in Figure 1 , the present application provides a high-precision PPP positioning method based on SSR correction number prediction, which comprises the following steps:

[0042] (1) The user terminal receiver receives and saves the SSR correction number in real time to obtain the SSR sequence;

[0043] (2) When the SSR data is interrupted, the saved SSR sequence is input into the pre-trained prediction model, and the predicted value of the SSR correction number in the future period is output;

[0044] (3) The predicted SSR correction number is used to correct the broadcast ephemeris to obtain the precise satellite orbit and clock error;

[0045] (4) A ionosphere-free combination model is constructed based on dual-frequency observations, and the receiver position is solved by Kalman filtering combined with the corrected satellite orbit and clock error.

[0046] In step (1), the receiver is composed of a receiver board module, a GNSS antenna, a power supply module, and various connection lines. The receiver receives the signals transmitted by the navigation satellite through the antenna, and encodes the carrier phase value and pseudorange observation value of the observed satellite into RTCM format data and sends it to the computer end. The computer end receives the RTCM format data from the receiver and decodes it using the RTKLIB software. In this embodiment, for the reception of SSR data, the computer end needs to connect to the Internet and log in to the NTRIP account using the BNC software, select one of the mount points (such as WHU) as the data source to receive the RTCM format SSR data, and the software can effectively process the RTCM format data, and then save the SSR data in the computer. Among them, the RTKLIB software is an open source software for real-time processing of GNSS observation data, integrating SSR correction numbers and realizing high-precision PPP positioning. The SSR sequence includes the time information of the SSR correction number, the radial correction number, the tangential correction number, the normal correction number, and the clock correction number of the orbit.

[0047] In step (2), when the server network is abnormal and the user cannot receive the SSR data, the saved SSR sequence is input into the pre-trained prediction model, and the predicted value of the SSR correction number in the future period is output; Figure 2The deep learning model shown makes predictions. The model discards the complex encoder-decoder architecture in traditional Transformer prediction models and only retains the encoder component, which includes an embedding layer, a projector, and stackable Transformer modules, thereby making the model structure more concise and efficient. The feature embedding layer independently maps the input variable historical time series data, specifically: for the time information, the radial correction number of the orbit, the tangential correction number of the orbit, the normal correction number of the orbit, and the clock correction number, the time series of the five variables are respectively mapped to high-dimensional feature vectors through embedding matrices to generate a unique feature representation for each variable; for the feature representation of the clock correction number, because its change rule is complex and difficult to accurately model, a residual module is used for deep feature extraction, the residual module enhances the ability to capture the nonlinear change characteristics of the clock correction number through stacked residual blocks; after the clock feature processed by the residual module is spliced with the feature representation of the other variables, it is input into the multi-head self-attention mechanism to identify the global correlation between the feature representations of the variables; the normalized layer processes the spliced feature vector to normalize the feature channels of all variables to a similar distribution state, thereby reducing the influence of the difference in the value range of the variables on the model; finally, the feedforward neural network processes the normalized features, fuses the historical time series information and global correlation features, and generates prediction results of the SSR correction number in the future period of time through nonlinear mapping extrapolation.

[0048] When predicting the SSR correction number, the saved historical SRR data is preprocessed to obtain the SSR sequence as described in step (1), and the time information of the SSR correction number, the radial, tangential, and normal correction numbers of the orbit, and the clock correction number are taken as the input of the model. The input sequence length and the prediction sequence length of the model directly affect the prediction performance of the model. In this embodiment, the input SSR sequence length of the model is set to 336, and the output predicted SSR sequence length is 96. The time interval between adjacent SSR data in the sequence is 5 minutes. The input real-time orbit correction number and clock correction number sequence are sampled by a sliding time window strategy, and the continuous time series data is divided into continuous overlapping subsequences, each time step corresponds to a feature vector, realizing the conversion from the original data to the feature matrix. The sequence data sampled by the sliding window is input into the reverse dimension embedding module to independently embed the time series of a single variable into a token. Then the clock difference and the corresponding time features processed by the embedding layer are input into the residual processing module. The convolution in the residual block uses one-dimensional convolution to process time series data, allowing each time step to consider adjacent time points, which can capture short-term dependencies. After layer normalization, residual connection is performed, and three residual blocks are stacked to enable the network to learn longer dependencies. The processed orbit and clock difference feature sequence is spliced and input into the subsequent network for processing.

[0049] In step (3), since the predicted SSR orbit correction number is defined as a three-dimensional correction component in the satellite body coordinate system, coordinate system conversion is needed when correcting the broadcast ephemeris. The specific process is as follows: first, according to the time information of the SSR orbit correction number and the Issue Of Data Ephemeris (IODE) parameter value, the corresponding broadcast ephemeris is matched; then, the orbit correction number in the star-fixed coordinate system is converted into a correction vector in the earth-fixed coordinate system; finally, the coordinates of the matched broadcast ephemeris are combined with the orbit correction quantity in the earth-fixed coordinate system to obtain the accurate position of the satellite. If the satellite orbit correction number and the change rate at the reference time t0 are known, the correction number vector δO at any time t is expressed as:

[0050]

[0051] Then, the rotation matrix R = (e R e A e C ) T is calculated, and the star-fixed coordinate system correction number is converted to the earth-centered earth-fixed coordinate system:

[0052]

[0053] In the formula, the parameters r sat and v sat respectively represent the satellite position and velocity vectors in the earth-fixed coordinate system solved by the corresponding broadcast ephemeris at epoch t. Then, the broadcast ephemeris is corrected by using the correction number to obtain the accurate satellite position (X post , Y post , Z post ):

[0054]

[0055] In the formula, (X brdc , Y brdc , Z brdc ) is the satellite position calculated according to the broadcast ephemeris at t. It should be particularly pointed out that the reference datum defined by the SSR includes two types of satellite antenna phase center (APC) and satellite center of mass (COM). When the satellite center of mass is used as the reference datum, the antenna phase center deviation correction quantity needs to be applied to obtain the accurate coordinates of the APC.

[0056] The real-time precise clock correction data realizes real-time correction of the satellite clock difference through a quadratic polynomial model, and the mathematical expression is:

[0057] ΔC = a0 + a1 (t-t0) + a2 (t-t0) 2

[0058] In the formula, a0, a1, a2 are clock correction parameters of SSR, t is the current epoch time, t0 is the reference epoch, and the satellite clock error at the time t after correction is:

[0059]

[0060] In the formula, dt brdc represents the satellite clock error calculated according to the broadcast ephemeris, and c represents the speed of light.

[0061] In step (4), a dual-frequency observation value is used as the basis, such as the B1I and B3I observation frequencies of the Beidou system, to establish an ionosphere-free (IF) function model for solving, and the IF model can be expressed as:

[0062]

[0063] wherein, represents the ionosphere-free combined pseudo-range observation value from the receiver r to the satellite s, represents the geometric distance from the receiver r to the satellite s, and c is the speed of light; dt r represents the clock error accumulation error of the receiver local oscillator caused by temperature drift, crystal aging and other factors, dt s,Q represents the clock frequency deviation of the satellite atomic clock during on-orbit operation caused by the relativistic effect and the influence of the space environment; and respectively represent the uncorrected pseudo-range hardware delay at the receiver end and the satellite end; and represent the uncorrected phase hardware delay at the receiver and satellite end; T r s,Q represents the delay of the troposphere; is and combined wavelength; and respectively represent the pseudo-range and carrier phase observation noise of the model.

[0064] When the data is interrupted, according to Figure 2As shown in the flow, the user end establishes the above-mentioned IF model according to the observation value obtained by the receiver, the broadcast ephemeris and the predicted SSR data, first uses the least square method to obtain the initial coordinates of the receiver, and then uses the initial value of Kalman filtering for parameter estimation, so as to obtain the accurate position coordinates of the receiver of the user end. In order to evaluate the performance of the SSR data predicted by the iTransformer model based on the residual network in precise point positioning, in this embodiment, the model is used to predict the SSR data at future time when the data is interrupted, so as to ensure the continuity of positioning. RTKLIB software is used for positioning solution, extended Kalman filter is used as a parameter estimator, ambiguity parameter remains in floating solution in positioning solution, receiver clock difference is modeled as a white noise process, receiver coordinates are constant estimated, and coordinates in the coordinate file provided by IGS are used as reference values. As shown in the figure, Figure 3 As shown in the figure, the observation data of the CUSV station is selected, the experimental time is 22:00 to 02:00 for 4 hours, the SSR data interruption is simulated at 00:00, and the positioning is switched to use the predicted SSR data of the deep learning model. The orange and yellow parts represent the normal data receiving condition from 22:00 to 00:00, and when the data interruption occurs at 00:00, the positioning is switched to use the predicted SSR data. It can be seen that the positioning error of the predicted data of the method changes smoothly and maintains in a small range, and the error in E / N / U direction within 60 min is less than 10 cm.

[0065] The above has described the embodiments of the present application, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes are obvious to those skilled in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles, practical application or technical improvement in the market of the embodiments, or to enable other ordinary skilled persons in the art to understand the embodiments disclosed herein.

Claims

1. A high-precision PPP positioning method based on SSR correction number prediction, characterized in that, The method comprises the following steps: (1) a user terminal receiver receives and saves SSR corrections in real time to obtain an SSR sequence; (2) when SSR data is interrupted, the saved SSR sequence is input into a pre-trained prediction model to output a predicted value of SSR corrections in a future period; (3) the predicted SSR corrections are used to correct broadcast ephemeris to obtain precise satellite orbits and clock errors; (4) a combination model without ionosphere is constructed based on dual-frequency observations, combined with the corrected satellite orbits and clock errors, and a receiver position is solved by Kalman filtering.

2. The high-precision PPP positioning method based on SSR correction number prediction according to claim 1, characterized in that, The SSR sequence comprises time information, orbit radial corrections, orbit tangential corrections, orbit normal corrections and clock error corrections.

3. The high-precision PPP positioning method based on SSR correction number prediction according to claim 1, characterized in that, The prediction model adopts an improved iTransformer model, and the structure of the improved iTransformer model comprises: an embedding layer for independently mapping time series data of each variable in an input sequence into a feature vector; a residual module for deep feature extraction of clock error corrections, comprising at least one residual block comprising a one-dimensional convolution layer; a self-attention layer for identifying the correlation between variables; a feedforward network layer for generating a predicted sequence.

4. The high-precision PPP positioning method based on SSR correction number prediction of claim 1, wherein, The specific process of correcting the broadcast ephemeris comprises: matching broadcast ephemeris according to the time information of SSR orbit corrections and ephemeris data period parameters; converting orbit corrections in a star-fixed system into geostationary system corrections through a rotation matrix; combining broadcast ephemeris coordinates and geostationary system corrections to obtain a satellite precise position.

5. The high-precision PPP positioning method based on SSR correction number prediction according to claim 1, characterized in that, The process of obtaining the precise satellite orbits and clock errors comprises: converting predicted orbit corrections in a star-fixed coordinate system into a geocentric geostationary coordinate system through a rotation matrix, and combining the broadcast ephemeris to obtain a satellite precise position; Based on the quadratic polynomial model ΔC=a0+a1(t-t0)+a2(t-t0) 2 The satellite clock error is corrected, and the precise clock error sequence is generated combined with the broadcast ephemeris clock error data, wherein a0, a1, and a2 are clock error correction parameters of the SSR, are determined by fitting historical clock error data through the least square method, and are updated in real time; t is a current epoch time, t0 is a reference epoch, and the satellite clock error at the time t after correction is: In the formula, dt brdc denotes the satellite clock error calculated from the broadcast ephemeris, and c denotes the speed of light.

6. The high-precision PPP positioning method based on SSR correction number prediction of claim 1, wherein, the combination model without ionosphere is expressed as: wherein, denotes the ionosphere-free combined pseudorange observation from receiver r to satellite s, denotes the geometric distance from receiver r to satellite s, c is the speed of light; dt r denotes the clock bias accumulated error of receiver local oscillator caused by temperature drift, crystal aging and other factors, dt s,Q denotes the clock frequency deviation of satellite atomic clock caused by relativistic effects and space environment during on-orbit operation; and denote the uncorrected pseudorange hardware delay at the receiver end and the satellite end, respectively; and denote the uncorrected phase hardware delay at the receiver and satellite ends; denotes the delay of the troposphere; is and the combined wavelength; and denote the pseudorange and carrier phase observation noise of the model, respectively.

7. The high-precision PPP positioning method based on SSR correction number prediction according to claim 6, characterized in that, the process of calculating the receiver position comprises: (a) linearizing the observation equation of the combination model without ionosphere to obtain error equations of a residual vector, a coefficient matrix, a to-be-estimated parameter vector and an observation vector; (b) when SSR data is interrupted, the error equations are solved by a least square method to obtain initial coordinates and initial clock errors of the receiver; (c) the initial coordinates and the initial clock errors are taken as initial values of a state vector of Kalman filtering, and dynamic parameter estimation is performed in combination with a dual-frequency observation sequence; (d) the state vector is updated through Kalman filtering iteration, and a final precise coordinate of the receiver is output.

Images