GNSS model self-adaptive optimization method for navigation scene perception enhancement in high-precision positioningTechnical Field
The application relates to the technical field of GNSS, in particular to a GNSS model self-adaptive optimization method for enhancing navigation scene perception in high-precision positioning.
Background
The continuous development of RTD, RTK and other technologies with multi-frequency and multi-mode fusion enables GNSS to provide centimeter-level high-precision dynamic positioning service for users in an open environment. By fusing with carriers such as automobiles, mobile phones and unmanned aerial vehicles, the application environment of the GNSS is gradually changed from static, single, simple to dynamic, various and complex. This puts new demands on the GNSS dynamic positioning performance in complex environments.
The method comprises the steps of obtaining GNSS signals of different scenes, enabling the GNSS observation values of the different scenes to be inconsistent in quality due to the difference loss caused by the GNSS signals, solidifying the existing GNSS threshold model through multiple parameters, wherein the GNSS threshold model is obviously inconsistent with the real quality distribution of the GNSS observation values, reconstructing a random model aiming at a single scene in the aspect of random model optimization for representing the GNSS observation value quality with the difference in the different scenes, and carrying out random model optimization aiming at the single scene in the prior art, wherein the random model solidified through the parameters can not adapt to GNSS positioning requirements in a dynamic complex environment, or adjusting a weighting strategy after the topographic and geomorphic information of the site is detected, and the final effect of the technology depends on the accuracy of the topographic and geomorphic information detection and the granularity of scene classification.
At present, the scene perception based on GNSS signals mostly uses an empirical threshold method or a traditional machine learning algorithm, so that the rough perception of indoor and outdoor open and outdoor shielding scenes is realized, the scene granularity is low, and the complex and changeable scene information requirements in the GNSS dynamic positioning are difficult to meet.
Therefore, it is obviously difficult to adapt to complex and changeable dynamic application scenarios by using a fixed GNSS positioning strategy.
Disclosure of Invention
In order to improve the availability, accuracy and continuity of GNSS positioning services in complex scenes, the GNSS positioning requirements in dynamic complex environments are met. The application provides a GNSS model self-adaptive optimization method for enhancing navigation scene perception in high-precision positioning.
The first object of the present application is achieved by the following technical solutions:
a GNSS model adaptive optimization method for navigation scene perception enhancement in high-precision positioning comprises the following steps:
Acquiring GNSS observation data, and extracting and calculating to obtain GNSS data characteristics, wherein the GNSS data characteristics comprise preset carrier-to-noise ratios, satellite altitude angles, DOP data, satellite numbers and multipaths under different classification scenes;
constructing a time sequence GNSS feature vector of the GNSS data feature through SFFS algorithm;
constructing an LSTM neural network with a time memory function based on the GNSS feature vector;
training the LSTM neural network by combining the GNSS observation data and a scene label preset by a navigation scene to obtain an LSTM model for scene perception;
Obtaining a position true value provided by navigation equipment synchronous with test equipment, substituting a preset inter-satellite-epoch double-difference observation equation of pseudo range and carrier phase, and back calculating a residual error of the pseudo range and the carrier phase;
Fitting a functional relation between the residual error and a preset influence factor, and constructing GNSS random models of different classification scenes, wherein the preset influence factor comprises a carrier-to-noise ratio and a satellite altitude angle;
based on different classification scenes and GNSS observation data, constructing threshold models of the different classification scenes, wherein parameters of the threshold models comprise carrier-to-noise ratio, satellite altitude angle and ratio values;
And after receiving the current environment information in real time and judging the classification scene to which the current environment belongs, the LSTM model for scene perception calls a GNSS random model under the corresponding classification scene to carry out weight determination, and adjusts the carrier-to-noise ratio, satellite altitude angle and ratio value matched with the corresponding classification scene through a threshold model.
Optionally, the acquiring GNSS observation data, extracting and calculating to obtain GNSS data features, where the GNSS data features include a carrier-to-noise ratio, a satellite altitude angle, DOP data, a satellite number, and multipath in a classification scene, and the method includes:
constructing a data acquisition platform through interest equipment and high-precision integrated navigation equipment, and collecting the number of satellites, carrier-to-noise ratio, satellite altitude angle and multipath visible by GNSS of the classified scene;
The multipath obtaining mode includes obtaining the difference between the change amount of the pseudo range and the Doppler observation value to evaluate the influence of NLOS and multipath under different environments, and the specific calculation formula is as follows:
; (1)
In the formula (1), the components are as follows,Is a multipath error; Is a difference operator; Is a pseudo-range observation; is Doppler observed value; Is the speed of light; is the signal frequency; Is a time interval;
The calculation formula of (2) is as follows: Wherein the subscripts、Indicating the time of day.
Optionally, the acquiring GNSS observation data extracts and calculates GNSS data features, where the GNSS data features include a carrier-to-noise ratio, a satellite altitude angle, DOP data, a satellite number, and multipath in a classification scene, and further includes:
The position accuracy attenuation factor PDOP of the satellite is introduced when the space rectangular coordinates of the receiver are used、、Clock errorWhen the parameters to be estimated are used for positioning calculation, a co-factor matrix is obtained as follows:
; (2)
the satellite's position accuracy attenuation factor PDOP can be expressed as:
。
optionally, the constructing the time sequence GNSS feature vector of the GNSS data feature by using the SFFS algorithm includes:
calculating the carrier-to-noise ratio, satellite altitude angle and GNSS visible satellite number and multipath statistic under different preset satellite azimuth angles under different preset classification scenes through a preset statistical algorithm;
calculating the statistics of the carrier-to-noise ratio under different satellite altitude angles and different satellite azimuth angles;
Calculating satellite altitude statistics under different satellite azimuth angles and a satellite position accuracy attenuation factor PDOP value;
Constructing a feature set for scene perception based on the number of satellites visible by the GNSS, the multipath statistic, the carrier-to-noise ratio statistic, the satellite altitude angle statistic and a position accuracy attenuation factor PDOP value of the satellites;
Initializing the SFFS algorithm by using an empty feature set, and adding the GNSS data features and judging whether the classification performance is improved or not, wherein the number of the GNSS data features added each time comprises two or more;
Outputting a current feature set for navigating the GNSS feature vector of scene perception when the classification performance is no longer improved;
And deleting one GNSS data feature which is not in accordance with the GNSS feature vector condition and has the worst perception result from the GNSS data features when the feature set does not reach the condition for forming the GNSS feature vector after the GNSS data features are added, if the deleted GNSS data feature is not newly added, deleting the one GNSS data feature which is not in accordance with the GNSS feature vector condition and has the worst perception result continuously, otherwise, continuing to select a new GNSS data feature for adding.
Optionally, the LSTM model includes a forgetting gate, an updating gate, and an output gate, where the forgetting gate includes a sigmoid layer, the updating gate and the output gate each include a sigmoid layer and a tanh layer, and the training of the LSTM neural network by combining GNSS observation data and a scene tag preset by a navigation scene to obtain the LSTM model for scene perception includes:
The calculation formula of the sigmoid layer and the tanh layer is as follows:
And;
In the LSTM model, the expressions of the forget gate, the update gate, and the output gate are:
forgetting the door:
;
Update door:
;
;
;
output door:
;
;
The input information of one LSTM unit in the LSTM model comprises the memory cell state of the last epochHidden layer outputFeature information with current epochAfter the input information is obtained, the last epoch hidden layer outputsAnd current epoch characteristic informationFirstly, entering the forgetting gate, determining forgetting of non-important information through a sigmoid layer of the forgetting gate, and outputting the forgetting information as follows;
And (3) withDetermining information to be updated through a sigmoid layer of the update gateCreating new candidate state quantities through the tanh layer,And (3) withMultiplying determines the information that needs to be stored in the current cell state, whileAnd (3) withThe multiplication realizes the information discarding of the last epoch;
And (3) withAdding to obtain the complete cell state of the current epochThe cell state will be output to the LSTM unit of the next epoch, i.e. the update gate constituting the LSTM model;
And (3) withDetermining output information in cell status by sigmoid layer of the output gate,And the tanh layer passing through the output gateThe multiplication results in the final output part, i.e. the output gate constituting the LSTM model.
Optionally, the obtaining a position true value provided by a navigation device synchronized with the test device substitutes a preset inter-satellite-inter-epoch double difference observation equation of a pseudo range and a carrier phase, and inversely calculates a residual error of the pseudo range and the carrier phase, including:
The GNSS pseudo-range observation equation and the carrier phase observation equation are as follows:
; (3)
; (4)
Wherein the subscriptIndicating the receiver, superscriptRepresenting a satellite;、 pseudo-range observations and carrier phase observations, respectively; For the geometrical distance between the receiver and the satellite, the calculation formula is as follows,、The positions of the receiver and the satellite under the ECEF frame;
in order to achieve the light velocity, the light beam is,、Receiver clock difference and satellite clock difference,Is ionospheric delay; is a tropospheric delay; is a multipath error;、、、 the method comprises the steps of respectively carrying out hardware delay on a receiver pseudo-range, a satellite pseudo-range, a receiver phase and a satellite phase of the receiver; Is a carrier wavelength; Is a non-poor integer ambiguity;、 pseudo-range observation noise and carrier phase observation noise are respectively;
Acquiring GNSS observation data of the formulas (3) and (4), performing inter-planet and inter-epoch double difference processing, and constructing a double difference observation equation;
First utilizing reference satelliteWith the target satelliteThe inter-satellite single difference observation equation is constructed by the inter-satellite observation quantity:; (5); (6)
In the formula,Is a single difference operator between the stars,、The difference between the pseudo-range and carrier phase observations on duty for the week, respectively; for receiver and reference satelliteTarget satelliteThe difference in the geometric distance between them,As reference satelliteClock error, target satelliteThe difference in the clock difference between the two,As reference satelliteWith the target satelliteIs used to determine the difference in ionospheric delay,As reference satelliteWith the target satelliteIs used to determine the difference in tropospheric delay,As reference satelliteWith the target satelliteMultipath error differences of (a);、 Respectively, reference satellitesPseudo-range, target satellitePhase hardware delays the inter-star difference; As reference satelliteWith the target satelliteSingle difference integer ambiguity between the satellites;
And carrying out inter-epoch difference on the basis of the inter-epoch single-difference observation equation to form an inter-epoch-inter-epoch double-difference observation equation as follows:
; (7)
;(8)
In the formula,Is an inter-satellite-epoch double difference operator, and subscript、Is a time mark; AndInter-epoch double difference observations of pseudorange and carrier phase, respectively; For the reference satelliteTarget satelliteAt the moment of timeTime of dayInter-epoch double difference of geometric distance between the two satellites; inter-epoch double difference being ionosphere delay; Inter-satellite-epoch double difference being troposphere delay; inter-epoch double difference, which is a multipath error; is inter-epoch double difference integer ambiguity;
Obtaining broadcast ephemeris, UNB3m model and Klobuchar model to calculate satellite clock biasAcquiring tropospheric delayAnd ionospheric delayThe correction precision of the Klobuchar model is 2-8 TECU, the satellite clock precision is 2.5ns, and the forecast precision of the UNB3m model in the zenith direction is 5.2cm;
when no cycle slip occurs between epochs, after correction of the broadcast ephemeris, the UNB3m model and the Klobuchar model, the formulas (7) and (8) can be simplified as:
; (9)
; (10)
Substituting the true values of the receiver coordinates into the formulas (9) and (10), and back-calculating the pre-test residual terms comprising the observed value noise and the multipath errors reflecting the environmental influence, wherein the pre-test residual terms are as follows:
; (11)
; (12)
when the single frequency data is single frequency data, the single frequency data comprises three systems of GPS, BDS and GALILEO, and the single frequency pseudo-range and carrier phase double-difference residual error can be expressed as:
; (13)
;(14)
in the formulas (13) and (14),、AndAre known, and the multipath errors of the three systems GPS, BDS and GALILEO can be calculated by the formulas (13) and (14)Pseudo range observation noiseIs a comprehensive error term of (2)And multipath errorNoise observed with carrier phaseIs a comprehensive error term of (2)According to the error propagation law, the relationship between the middle error of the combined residual error and the middle error of the carrier phase non-difference residual error is as follows:
; (15)
; (16)
Wherein,And (3) with、And (3) with、And (3) with、And (3) withFor the integrated error term of adjacent epochs of the same satellite, the equations (15) and (16) can be expressed as:
; (17)
。 (18)。
Optionally, the fitting the functional relation between the residual error and a preset influence factor, and constructing GNSS random models of different classification scenes, where the preset influence factor includes the carrier-to-noise ratio and the satellite altitude angle, includes:
Selecting a satellite with the targetSatellites with similar influence factors are the reference satellitesFor reference, formulae (17) and (18) at this time can be expressed as:
; (19)
; (20)
acquiring the GNSS observations of different classification scenarios for computationReconstructing a classified scene pseudo-range random model under different scenes by carrier-to-noise ratio and satellite altitude angle: Wherein CNR is carrier-to-noise ratio, ELE is satellite altitude angle, and positioning is performed by usingAnd calculating the carrier phase observed quantity weight.
Optionally, the constructing a threshold model of different classification scenes based on different preset classification scenes and GNSS observation data, where the threshold model includes a carrier-to-noise ratio, a satellite altitude angle, and a ratio value includes:
Based on the GNSS observation data, extracting the distribution conditions of the download noise ratio and the satellite altitude angle of different classification scenes, and selecting the download noise ratio and the altitude angle threshold value of different classification scenes according to a triple sigma principle;
calculating the inter-station difference on the basis of the formula (6), and forming a carrier phase station star double-difference observed quantity:
(21)
in formula (21), subscriptIs a mobile station,Is a reference station; The carrier phase station star double difference observation value; Station star double difference for ionosphere delay; Standing star double difference for troposphere delay; Station star double difference with multipath error; The true station star double-difference integer ambiguity is calculated back as a reference value after substituting the true value coordinates of the receiver for the station star double-difference integer ambiguity;
Performing GNSS RTK positioning, comparing and positioning to solve the station star double-difference integer ambiguity and the real station star double-difference ambiguity, judging whether to realize true fixation, acquiring the distribution condition of the ratio values under different classification scenes, and outputting a fixed rate and a correct fixed rate, wherein the correct fixed rate is the true fixed rate, the fixed rate is the successfully fixed rate, and a ratio threshold is set through the fixed rate and the correct fixed rate.
The second object of the present application is achieved by the following technical solutions:
a computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, the processor implementing the steps of a navigation scene perception enhanced GNSS model adaptive optimization method in high precision positioning as described above when the computer program is executed.
The third object of the present application is achieved by the following technical solutions:
a computer readable storage medium storing a computer program which when executed by a processor implements the steps of a navigation scene perception enhanced GNSS model adaptive optimization method in high precision positioning described above.
In summary, the present application includes at least one of the following beneficial technical effects:
1. The invention provides a navigation scene perception enhanced self-adaptive GNSS positioning model, which is used for constructing a multi-dimensional GNSS feature vector by considering the time correlation characteristic of GNSS feature information under the condition of using only a GNSS single signal source, designing and training a long-term memory LSTM neural network based on a large number of actual GNSS observation data of a complex scene, and realizing outdoor complex navigation scene perception; based on the set classification scene and a large amount of GNSS observation data, constructing a carrier-to-noise ratio, altitude angle and ratio threshold model of the classification scene, and reversely calculating GNSS observation value residual errors by using a coordinate truth value brought into a pseudo range/carrier phase double-difference observation equation, and constructing a classification scene GNSS random model by using the same;
2. considering that there is a certain connectivity between the navigation scene epochs, that is, the current epoch scene is more likely to be consistent with the last epoch scene, the cyclic neural network (RNN) with the current feature and the previous feature as common input is more suitable for navigation scene sensing, because the RNN introduces the hidden layer concept, the neurons with self-feedback capability contained therein can be used for processing time sequence signals with any length. As a variant of RNN, LSTM is further added into a gating structure, so that the problem that gradient possibly appears in long-time sensing disappears is better solved than RNN, and therefore LSTM is selected for navigation scene sensing;
3. In complex environments, multipath is serious, and the pseudo-range measurement error caused by the multipath may reach several meters or even more than ten meters, and the influence of the multipath error on the carrier phase is at most only in the centimeter level. As a comparison, the main influencing term in the formula (9) isTherefore, the pseudo-range observation residual error can be calculated by the formula (9) to construct a pseudo-range classification scene random model, and the formula (10) possibly also comprises the influence of other error items, but because the carrier and the pseudo-range observation environments are the same, after the pseudo-range classification scene random model is reconstructed and the influence of the environments on the quality of the GNSS observation value is reflected, the residual error proportion of the pseudo-range and the carrier phase observation value is calculated, and the specific weight of the carrier phase observation value in the corresponding scene and the GNSS positioning is obtained by using the corresponding proportion.
4. The carrier-to-noise ratio or satellite altitude is usually chosen as an influencing factor when constructing a stochastic model. The method comprises the steps of selecting two kinds of influence factors to jointly construct a random model, and after the influence factors are selected, considering the precision of a GNSS observation value to obey certain function distribution of the influence factors, namely the precision of the observation value of the same influence factor is the same. Thus, select and target starSatellites with influence factors as close as possible are used as reference satellites。
Drawings
FIG. 1 is a flowchart of a SFFS algorithm in an embodiment of a GNSS model adaptive optimization method for enhancing navigation scene perception in high-precision positioning according to the present application;
FIG. 2 is a flowchart illustrating an implementation of an embodiment of a GNSS model adaptive optimization method for enhanced navigation scene perception in high-precision positioning according to the present application;
fig. 3 is a schematic structural diagram of a computer device according to the present application.
Detailed Description
The present application will be described in further detail with reference to the accompanying drawings.
In an embodiment, as shown in fig. 1-3, the application discloses a navigation scene perception enhanced GNSS model adaptive optimization method in high-precision positioning, which specifically comprises the following steps:
s1, acquiring GNSS observation data, and extracting and calculating to obtain GNSS data characteristics, wherein the GNSS data characteristics comprise preset carrier-to-noise ratios, satellite altitude angles, DOP data, satellite numbers and multipaths under different classification scenes;
In this embodiment, the integrated navigation device and the interest device constitute a platform for data acquisition of GNSS observation data, and the interest device includes a golden dimension DM711 device, which provides a data base for construction of a scene perception model and a GNSS positioning model.
GNSS signal characteristics can be divided into four general categories, satellite visibility, satellite signal strength, multipath, satellite geometry. From these four classes, GNSS feature vectors of inter-scene variability are constructed. Specifically, the number of GNSS visible satellites, the signal carrier-to-noise ratio and the satellite altitude angle under different classification scenes are collected. In addition, considering that urban canyon scene signal shielding mainly comes from two sides of a road, under-overhead scene signal shielding in different classification scenes mainly comes from zenith directions and the like, namely, scene information is also contained in satellite orientations. Thus, the satellite azimuth is calculated in combination with the carrier attitude information provided by the integrated navigation device. In addition, the influence of NLOS and multipath under different environments is estimated by using the difference between the pseudo-range variation and the Doppler observation value, and the specific calculation formula of the multipath acquisition mode is as follows:
; (1)
In the formula (1), the components are as follows,Is a multipath error; Is a difference operator; Is a pseudo-range observation; is Doppler observed value; Is the speed of light; is the signal frequency; Is a time interval.The calculation formula of (2) is as follows: Wherein the subscripts、Indicating the time of day.
In order to evaluate the geometric distribution of satellites, the geometric intensity of the satellites is reflected by introducing a position accuracy attenuation factor PDOP of the satellites while statistics of various satellite altitude angle statistics are carried out. When the receiver is in a space rectangular coordinate system、、Clock errorWhen the parameters to be estimated are used for positioning calculation, a co-factor matrix is obtained as follows:
; (2)
the satellite position accuracy attenuation factor PDOP can be expressed as:。
s2, constructing a time sequence GNSS feature vector of the GNSS data feature through SFFS algorithm;
in this embodiment, the carrier-to-noise ratio, satellite altitude, and GNSS-visible satellite number and multipath statistics under different preset classification scenes are calculated by combining statistical methods such as polar error, quartile range, mean, standard deviation, ratio (current value/scene maximum value), etc.;
calculating the statistics of the carrier-to-noise ratio under different satellite altitude angles and different satellite azimuth angles;
Calculating satellite altitude statistics under different satellite azimuth angles and a satellite position accuracy attenuation factor PDOP value;
And constructing a feature set for scene perception based on the number of satellites visible by the GNSS, the multipath statistic, the carrier-to-noise ratio statistic, the satellite altitude angle statistic and the position accuracy attenuation factor PDOP value of the satellites.
And constructing a feature quantity set for scene perception by using the calculated values, and constructing a final feature vector by selecting the feature quantity by using SFFS (Sequential Forward Floating Selection) algorithm, wherein the flow of SFFS algorithm is shown in fig. 2.
The SFFS algorithm is initialized by an empty feature set, and by adding GNSS data features and judging whether classification performance is improved, the number of the GNSS data features added each time comprises two or more;
Outputting a current feature set for navigating the GNSS feature vector of scene perception when the classification performance is no longer improved;
And deleting one GNSS data feature with the worst perception result in the added GNSS data features when the feature set does not reach the condition for forming the GNSS feature vector after the GNSS data features are added, continuously deleting the one GNSS data feature which does not meet the condition for forming the GNSS feature vector and has the worst perception result if the deleted GNSS data feature is not newly added, otherwise, continuously selecting new GNSS data features to be added.
S3, constructing an LSTM neural network with a time memory function based on the GNSS feature vector;
S4, training an LSTM neural network by combining GNSS observation data and a scene label preset by a navigation scene to obtain an LSTM model for scene perception;
in this embodiment, the LSTM unit is shown in the following table 1, where the LSTM model includes a forgetting gate, an updating gate, and an output gate, the forgetting gate includes a sigmoid layer, the updating gate and the output gate both include a sigmoid layer and a tanh layer, and in the following table 1, the calculation formulas of the sigmoid layer and the tanh layer are:、。
TABLE 1 LSTM cell
Specifically, the input of one LSTM cell in the LSTM model includes the memory cell state of the last epochHidden layer outputFeature information with current epoch. After the input is obtained, the last epoch hidden layer outputAnd current epoch characteristic informationFirstly, entering a forgetting gate, determining forgetting of non-important information through a sigmoid layer of the forgetting gate, and outputting the forgetting information as follows;
Then, the process is carried out,And (3) withDetermining information to be updated through a sigmoid layer of an update gateCreating new candidate state quantities through the tanh layer,And (3) withMultiplying determines the information that needs to be stored in the current cell state, whileAnd (3) withMultiplication achieves the discarding of information of the last epoch,
And (3) withAdding to obtain the complete cell state of the current epochThe cell state will be output to the LSTM cell of the next epoch, which constitutes the update gate of LSTM;
And (3) withDetermining output information in cell status through sigmoid layer of output gate,And the tanh layer passing through the output gateThe multiplication results in the final output part, which constitutes the output gate of the LSTM. LSTM solves the problem of gradient disappearance in the long-time memory process of the neural network through the gating structure.
To obtain accurate and reliable navigation scene sensing results, LSTM super parameters suitable for GNSS feature vectors are set. The LSTM super parameters to be set are mainly three types, namely the number of hidden layer layers, the number of hidden layer neurons and a time window. In general, the more the number of hidden layers, the more the number of hidden layer neurons, the longer the time window, the more expressive the model, and at the same time, the longer the model training time and the greater the risk of model overfitting. Therefore, on the premise of considering training accuracy/loss, testing accuracy/loss and training time, the super parameters are set to be 2 layers of hidden layers, 26 numbers of hidden layer neurons and 7s of time window.
S5, obtaining a position true value provided by navigation equipment synchronous with the test equipment, substituting a preset inter-satellite-epoch double-difference observation equation of the pseudo range and the carrier phase, and inversely calculating a residual error of the pseudo range and the carrier phase;
The GNSS pseudo-range observation equation and the carrier phase observation equation are as follows:
; (3)
; (4)
Wherein the subscriptRepresenting the receiver [ ]Can be a mobile stationOr reference station) Superscript (I)Representing a satellite;、 pseudo-range observations and carrier phase observations, respectively; for the geometrical distance between the receiver and the satellite, the calculation formula is as follows,、The positions of the receiver and the satellite under the ECEF frame; Is the speed of light;、 Receiver clock error and satellite clock error respectively; is ionospheric delay; is a tropospheric delay; is a multipath error;、、、 the method comprises the steps of respectively carrying out hardware delay on a receiver pseudo-range, a satellite pseudo-range, a receiver phase and a satellite phase of a receiver; Is a carrier wavelength; Is a non-poor integer ambiguity;、 pseudo-range observation noise and carrier phase observation noise, respectively.
In order to obtain purer observables which only contain reaction environment information (multipath error) and observation noise, the method is used for constructing a classification scene GNSS random model, and a double-difference observation equation is constructed by utilizing the original observation values of the formulas (3) and (4) and the GNSS observation data of the formulas (3) and (4) to carry out double-difference processing between planets and epochs.
Using reference satellitesWith the target satelliteThe inter-satellite single difference observation equation is constructed by the inter-satellite observation quantity:
; (5)
; (6)
In the formula,Is a single difference operator between the stars,、The difference between the pseudo-range and carrier phase observations on duty for the week, respectively; for receiver and reference satelliteTarget satelliteThe difference in the geometric distance between them,As reference satelliteClock error, target satelliteThe difference in the clock difference between the two,As reference satelliteWith the target satelliteIs used to determine the difference in ionospheric delay,As reference satelliteWith the target satelliteIs used to determine the difference in tropospheric delay,As reference satelliteWith the target satelliteMultipath error differences of (a);、 Respectively, reference satellitesPseudo-range, target satellitePhase hardware delays the inter-star difference; As reference satelliteWith the target satelliteSingle difference integer ambiguity between the stars.
In order to further weaken space correlation errors such as satellite end errors, atmospheric errors (ionospheric delay, tropospheric delay), orbit errors and the like, inter-epoch difference is carried out on the basis of inter-epoch difference to form an inter-epoch-inter double-difference observation equation as follows:
; (7)
;(8)
In the formula,Is an inter-satellite-epoch double difference operator, and subscript、Is a time mark; AndInter-epoch double difference observations of pseudorange and carrier phase, respectively; For the reference satelliteTarget satelliteAt the moment of timeTime of dayInter-epoch double difference of geometric distance between the two satellites; inter-epoch double difference being ionosphere delay; Inter-satellite-epoch double difference being troposphere delay; inter-epoch double difference, which is a multipath error; is inter-epoch double difference integer ambiguity.
Calculating satellite clock error by using broadcast ephemeris, UNB3m model based on weather forecast and Klobuchar modelTropospheric delayDelay from ionosphere. According to the product information issued by the IGS, the correction precision of the Klobuchar model is 2-8 TECU, the satellite clock precision is 2.5ns, and the forecast precision of the UNB3m model in the zenith direction is 5.2cm. Due to satellite clock errorsIonospheric delayTropospheric delaySatellite hardware delayAre slow variables, and after no cycle slip occurs between epochs and the broadcast ephemeris, UNB3m model and Klobuchar model are corrected, the formulas (7) and (8) can be simplified as follows:
; (9)
; (10)
It is noted that under complex environments, multipath is serious, and the pseudo-range measurement error caused by the multipath may reach several meters or even more than ten meters, and the influence of the multipath error on the carrier phase is only at the level of centimeters at maximum. As a comparison, the main influencing term in the formula (9) isTherefore, the pseudo-range observation residual error can be calculated by the formula (9) to construct a pseudo-range classification scene random model, and the formula (10) possibly also comprises the influence of other error items, but because the carrier and the pseudo-range observation environments are the same, after the pseudo-range classification scene random model is reconstructed and the influence of the environments on the quality of the GNSS observation value is reflected, the residual error proportion of the pseudo-range and the carrier phase observation value is calculated, and the specific weight of the carrier phase observation value in the corresponding scene and the GNSS positioning is obtained by using the corresponding proportion.
The true value of the receiver coordinate is brought into the formulas (9) and (10), and the pre-test residual error term mainly comprising the multipath errors of observed value noise and reaction environment influence is calculated as follows:
(11)
(12)
in this embodiment, taking GPS, BDS, and GALILEO single frequency data as examples, the three-system single frequency pseudo-range and carrier phase double difference residual error can be expressed as:
; (13)
;(14)
In the formulas (13) and (14),、、Are all known terms, and can be calculated by the formulas (13) and (14) to obtain three-system multipath errorsPseudo range observation noiseIs a comprehensive error term of (2)Multipath errorNoise observed with carrier phaseIs a comprehensive error term of (2). According to the error propagation law, the relationship between the middle error of the combined residual error and the error in the pseudo-range and carrier phase non-difference residual error is as follows:
; (15)
; (16)
Wherein,And (3) with、And (3) with、And (3) with、And (3) withAs for the integrated error term of the adjacent epoch of the same satellite, since the time interval of the adjacent epoch of the GNSS is generally less than or equal to 1s, the integrated error variation of the same satellite within 1s is considered to be small, so the expression (15) and the expression (16) can be expressed as:
; (17)
。 (18)
s6, fitting a functional relation between the residual error and a preset influence factor, and constructing GNSS random models of different classification scenes, wherein the preset influence factor comprises a carrier-to-noise ratio and a satellite altitude angle;
In this embodiment, the carrier-to-noise ratio or satellite altitude is generally selected as the influencing factor when constructing the stochastic model. The height of the satellite altitude angle mainly reflects the intensity of atmospheric errors such as troposphere delay and the like, the magnitude of the carrier-to-noise ratio mainly reflects diffraction errors and the like caused by an observation environment in the satellite signal transmission process, and the two are common random model influence factors. The invention fully utilizes the comprehensive advantages of the two factors, and selects the two influencing factors to jointly construct a random model. After the influence factors are selected, the precision of the GNSS observations is considered to be subject to a certain function distribution of the influence factors, namely, the precision of the observations of the same influence factors is the same. Thus, the satellite is selected and targetedSatellites with influence factors as close as possible are used as reference satellites, i.e. as reference satellitesIn this case, the formulas (17) and (18) can be expressed as:
; (19)
; (20)
GNSS observation data calculation using a large number of different classification scenariosCarrier to noise ratioSatellite altitude angleReconstructing a classified scene pseudo-range random model under different scenes: Wherein CNR is carrier-to-noise ratio, ELE is satellite altitude angle, and positioning is performed by usingCalculating the observed weight of carrier phase
S7, constructing threshold models of different classification scenes based on the different classification scenes and GNSS observation data, wherein parameters of the threshold models comprise carrier-to-noise ratio, satellite altitude angle and ratio value;
in this embodiment, based on a large amount of GNSS observation data, the download noise ratio CNR and the satellite altitude angle ELE distribution conditions of different classification scenes are extracted, and the download noise ratio CNR and the altitude angle ELE threshold under different classification scenes are selected according to the triple sigma principle.
In order to obtain proper ratio thresholds under different classification scenes, inter-station difference calculation is carried out on the basis of a formula (6), so as to form carrier phase station star double-difference observed quantity:
(21)
satellite clock error of satellite end correlation error through inter-station errorDelay with satellite hardwareIs completely eliminated. In formula (21), subscriptIs a mobile station,Is a reference station; The carrier phase station star double difference observation value; Station star double difference for ionosphere delay; Standing star double difference for troposphere delay; Station star double difference with multipath error; The true station star double-difference integer ambiguity is calculated back as a reference value after substituting the true value coordinates of the receiver for the station star double-difference integer ambiguity;
performing GNSS RTK positioning, comparing and positioning to solve the station star double-difference integer ambiguity with the real station star double-difference ambiguity, judging whether true fixation is realized, acquiring the distribution condition of the ratio values under different classification scenes, outputting a fixed rate and a correct fixed rate, wherein the correct fixed rate is the true fixed rate, the fixed rate is the successfully fixed rate, and setting a ratio threshold value through the fixed rate and the correct fixed rate.
S8, after receiving the current environment information in real time and judging the classification scene to which the current environment belongs, the LSTM model for scene perception calls a GNSS random model under the corresponding classification scene to carry out weight determination, and adjusts the carrier-to-noise ratio, satellite altitude angle and ratio value matched with the corresponding classification scene through a threshold model.
In an embodiment, referring to fig. 2, a plurality of GNSS observation data are extracted, and GNSS data features such as a carrier-to-noise ratio, a satellite altitude angle, a DOP, a satellite number and the like in a scene of interest are calculated, according to the feature of connectivity between navigation scenes, a time sequence GNSS feature vector is constructed through a SFFS algorithm, and according to the feature, an LSTM model super-parameter with time memory is designed, and a scene perception model is obtained by training LSTM by combining a plurality of actually measured GNSS data with a set scene tag. The method comprises the steps of obtaining a position true value provided by high-precision integrated navigation equipment synchronously mounted with test equipment, carrying out inverse calculation on pseudo range and carrier phase residual error by using an inter-satellite-epoch double-difference observation equation of pseudo range and carrier phase, comprehensively considering two influence factors of carrier-to-noise ratio and satellite altitude, fitting a function relationship between the residual error and the influence factors, obtaining GNSS random models of classified scenes, setting CNR and ELE thresholds in different scenes according to a triple sigma principle, carrying out inverse calculation on true double-difference ambiguity by using the position true value and the inter-satellite double-difference observation equation of the pseudo range and the carrier phase, and comprehensively considering a fixed rate and a correct fixed rate to set different scene ratio thresholds. After the LSTM model and the GNSS positioning model are obtained, environment information is obtained by utilizing the LSTM model in the positioning process, and the random model and the threshold model are adjusted accordingly, so that the self-adaptive optimization of the GNSS positioning model with enhanced navigation scene perception is realized.
It should be understood that the sequence number of each step in the foregoing embodiment does not mean that the execution sequence of each process should be determined by the function and the internal logic, and should not limit the implementation process of the embodiment of the present application.
In one embodiment, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 3. The computer device includes a processor, a memory, a network interface, and a database connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program, when executed by a processor, implements a GNSS model adaptive optimization method for navigation scene perception enhancement in high-precision positioning.
In one embodiment, a computer device is provided that includes a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing a GNSS model adaptive optimization method for navigation scene perception enhancement in high-precision positioning when executing the computer program.
In one embodiment, a computer readable storage medium is provided having a computer program stored thereon, which when executed by a processor, implements a GNSS model adaptive optimization method for navigation scene perception enhancement in high accuracy positioning.
Those skilled in the art will appreciate that implementing all or part of the above described methods may be accomplished by way of a computer program stored on a non-transitory computer readable storage medium, which when executed, may comprise the steps of the embodiments of the methods described above. Any reference to memory, storage, database, or other medium used in embodiments provided herein may include non-volatile and/or volatile memory. The nonvolatile memory can include Read Only Memory (ROM), programmable ROM (PROM), electrically Programmable ROM (EPROM), electrically Erasable Programmable ROM (EEPROM), or flash memory. Volatile memory can include Random Access Memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms such as Static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double Data Rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (SYNCHLINK) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), among others.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-described division of the functional units and modules is illustrated, and in practical application, the above-described functional distribution may be performed by different functional units and modules according to needs, i.e. the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-described functions.
The foregoing embodiments are merely illustrative of the technical solutions of the present application, and not restrictive, and although the present application has been described in detail with reference to the foregoing embodiments, it should be understood by those skilled in the art that modifications may still be made to the technical solutions described in the foregoing embodiments or equivalent substitutions of some technical features thereof, and that such modifications or substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present application.