CN-116893432-B - Standard single point positioning method based on big data cross-modal residual error model compensation

CN116893432BCN 116893432 BCN116893432 BCN 116893432BCN-116893432-B

Abstract

The invention provides a standard single point positioning method based on big data cross-modal residual error model compensation, which adopts an undirected graph to establish the relation between time and residual error, space and residual error, satellite elevation angle and residual error and satellite azimuth angle and residual error conditions, then utilizes random walk to establish global relation, and finally carries out cross-modal characterization learning on tuples generated by the random walk to establish a history relation model. In the aspect of prediction, a final prediction result is obtained according to the input space condition, the satellite elevation angle condition and the satellite azimuth angle condition and the combination of an output strategy. This residual is eventually compensated into a standard single point positioning algorithm. The invention has higher precision, does not need to rely on an external network at any time, and can ensure that the localization calculation can be carried out with optimal precision within 60 days every time the model is updated.

Inventors

YU FENGZHENG
YUAN HONG
LUO RUIDAN
CHEN SI
YANG GUANG
LI YAPING
WAN HONGXIA

Assignees

中国科学院空天信息创新研究院

Dates

Publication Date: 20260512
Application Date: 20230727

Claims (4)

1. A standard single point positioning method based on big data cross-modal residual model compensation is characterized by comprising the following steps: step 1, obtaining a pseudo-range residual error, wherein the pseudo-range residual error is calculated by adopting a precise ephemeris and an observation file: (1) In the middle of Is an observation station And satellite Pseudo-range observations at frequency i, For the true distance of which, Is the speed of light, which is the speed of light, , Receiver clock skew and satellite clock skew, Is the ionospheric delay and is a function of the ionospheric delay, Is the tropospheric delay and the delay of the layer, Is residual; The acquired pseudo-range residual error, the corresponding satellite number, elevation angle, azimuth angle, observation station coordinates and observation station time form a group of data sets, and each element in the data sets is regarded as a node; step 2, establishing an undirected graph, wherein the undirected graph shows whether the nodes are related and the degree of the related, and the weight among the nodes is set as the occurrence frequency of the undirected graph; step3, establishing a random walk tuple, which comprises the following steps: (2) In the formula, Is the current node and is the current node, Is the probability that the current will be, Representing slave Node selection next node Is a function of the probability of (1), () Representing a set of adjacent nodes, d () representing the number of adjacent nodes, generating a set of random walk sequences Wherein ) Represents a random walk algorithm, G represents an undirected graph relationship, Representing a starting node, and T represents the sequence length; Step 4, establishing a cross-modal characterization model, which comprises the following steps: Aiming at the characteristics of the data sets, four sets of undirected graphs are established, random walk sequences are respectively established, and nodes are trained Generating nodes To obtain the probability of actual calculation, defining undirected graph relation Simultaneously training the final loss function F of the four groups of undirected graphs; Step 5, obtaining a prediction strategy, which comprises the steps of carrying out prediction operation after obtaining a residual prediction model, wherein due to the characteristic of pseudo-range residual, nodes which are closer in time are more reliable, the input conditions comprise the elevation angle and azimuth angle of satellite signs and the coordinates of a stand, calculating 10 residual values which are closest to each other in a model mapping space by using the input conditions, and carrying out weighted calculation according to the time of the last occurrence of the history of the 10 residual values to obtain a final prediction result; Step 6, designing a standard single point positioning method, which comprises the following steps of And the difference value between the pseudo range and the geometrical distance of the receiver and the GNSS satellite is expressed, namely the pseudo range residual error.
2. The standard single point positioning method based on big data cross-modal residual model compensation according to claim 1, wherein the step 4 comprises: To be used for Represents undirected graph relationship, wherein X and Y respectively represent a class of nodes, U represents association, Representing the degree of association between nodes, training nodes Generating nodes The probability of (2) is: (3) In the middle of , , Is a node Node And an embedded vector of a node k, wherein k represents all nodes adjacent to the node i, X represents a set of nodes adjacent to the node i, and the actual calculated probability is as follows: (4) In the middle of Is the edge weight of node i and node j, X represents the set of nodes adjacent to i; Definition of the definition The loss function of (2) is: (5) wherein KuL represents KL divergence expressed as Wherein n represents the value of the existence of the random variables p and q, X represents the set of nodes adjacent to i, and Y represents the set of nodes adjacent to j; the final loss function F for the four sets of undirected graphs were trained simultaneously as follows: (6) In the middle of Representing the loss function of the coordinate of the observation station and the residual error, the loss function of the satellite elevation angle and the residual error, the loss function of the satellite azimuth angle and the residual error, and the loss function of the satellite signal and the residual error respectively; in order to mitigate the computation by adopting a negative sampling method, the finally learned loss function S is as follows: (7) Wherein, the () Representing the sigmoid function and M representing the number of negative samples taken, , , Is a node Node And an embedding vector for node k, k representing all nodes adjacent to node i.
3. The standard single point positioning method based on big data cross-modal residual model compensation according to claim 2, wherein in the step 5, the weighting method is as follows: (8) (9) In the middle of As the weight of the material to be weighed, Is the weight of the i-th predictor, The current time, the time of the predicted value in the history model, and the model start time, respectively.
4. A standard single point positioning method based on big data cross-modal residual model compensation according to claim 3, wherein the step 6 includes: (10) (11) (12) (13) in the formula (11) As shown in formula (12), is an intermediate parameter in which Respectively represent Partial derivatives of x, y, z; the pseudorange locations are shown in equation (10) -equation (13), where, 。

Description

Standard single point positioning method based on big data cross-modal residual error model compensation Technical Field The invention belongs to the field of navigation positioning, and particularly relates to a standard single-point positioning method based on big data cross-mode residual error model compensation. Background As one of the original observables of satellite signals, pseudo-range plays a very important role, and although the precision of pseudo-range positioning is not as good as that of the methods such as precise single-point positioning and real-time differential positioning which are continuously developed in recent years, the algorithm is simple, the station setting is convenient, the dependence on external network information is low, and the like, so that the pseudo-range positioning method has an irreplaceable role in a plurality of positioning scenes. However, no detailed rule research on pseudo-range errors exists at present, and residuals of the pseudo-range errors are fully known and can be predicted, so that better understanding of satellite signal systems is facilitated. From standard single point positioning to precise single point positioning to differential positioning. The positioning accuracy of the GNSS is continuously improved, but the improvement of the accuracy brings the disadvantages of lengthening the convergence time, increasing the dependence on external network information, and the like. As the calculation is simplest, the real-time performance is highest, the most widely applied positioning algorithm is adopted, and the standard single-point positioning has unique advantages. The most common standard positioning algorithm is iterated by adopting a least square method or a Kalman filtering method, however, in the iteration, because the pseudo-range residual error is not a known quantity, more is directly ignored or added into the unknown quantity of coordinates and clock errors to solve the unknown quantity, and the method is also an important factor with low standard single-point positioning accuracy. In the prior art, the observed quantity noise is estimated, and the method is wholly divided into three categories, namely, firstly, the observed quantity noise is regarded as a random model based on a height angle, wherein Bernese software is regarded as a cosine function model, GAMIT software is regarded as a sine function model, barnes software is regarded as an exponential function model, secondly, the observed quantity noise is regarded as a random model of a signal-to-noise ratio, and thirdly, the observed quantity residual value is continuously corrected by positioning and least square. None of these are models made purely for pseudorange residuals and are less accurate. Disclosure of Invention In order to grasp the change rule of pseudo-range residual errors more clearly and improve the precision of standard single-point positioning, the invention provides a standard single-point positioning method based on big data cross-mode residual error model compensation. The invention adopts an undirected graph to establish the relation among time and residual error, space and residual error, satellite elevation angle and residual error and satellite azimuth angle and residual error conditions, then utilizes random walk to establish global relation, and finally carries out cross-modal characterization learning on tuples generated by the random walk to establish a history relation model. In the aspect of prediction, a final prediction result is obtained according to the input space condition, the satellite elevation angle condition and the satellite azimuth angle condition and the combination of an output strategy. This residual is eventually compensated into a standard single point positioning algorithm. Compared with the traditional standard single-point positioning method, the method has higher precision, does not need to rely on an external network at any time, and can ensure that the optimal level which can be achieved by the model can be maintained within 50 days for localized calculation by updating the model once in 2 to 3 months. In order to achieve the above purpose, the invention adopts the following technical scheme: A standard single point positioning method based on big data cross-modal residual error model compensation comprises the following steps: step 1, acquiring pseudo-range residual errors; step 2, establishing an undirected graph; Step 3, establishing a random walk tuple; Step 4, establishing a cross-modal characterization model; Step 5, obtaining a prediction strategy; and 6, designing a standard single-point positioning method. Further, the step1 includes: The pseudo-range residual error is calculated by adopting a precise ephemeris and an observation file: In the middle of Is the pseudorange observations of the observation station r and satellite j at frequency i,Its true distance, c is the speed of light, dt r,dt(j) is the receiver clock diffe