Precise Orbit Determination Method, Device and Equipment for Low Earth Orbit Satellites Based on Machine Learning

Abstract

The present invention provides a method, device and equipment for precise orbit determination of low-Earth orbit satellites based on machine learning, which relates to the technical field of Earth observation and navigation, and includes: establishing an on-board GNSS heterogeneous multi-mode particle swarm model, and iteratively screening the on-board GNSS observation data to obtain the target on-board GNSS observation data participating in the precise orbit determination solution of the low-Earth orbit satellite for each epoch; complementing the satellite laser ranging data through a deep learning neural network to obtain the corresponding target satellite laser ranging data within the observable arc of the satellite laser ranging ground station for the low-Earth orbit satellite; estimating the pseudo-random pulse parameters through an inverse reinforcement learning model; and performing precise orbit determination according to the target on-board GNSS observation data, the target satellite laser ranging data and the pseudo-random pulse parameters. The present invention effectively solves the problems of on-board GNSS data redundancy and the lack of satellite laser ranging data in rainy and snowy weather, and avoids the uncertainty in the selection of pseudo-random pulse parameters, thereby effectively improving the orbit determination accuracy of low-Earth orbit satellites.

Description

Technical Field

[0001] The present invention relates to the technical field of earth observation and navigation, and in particular to a precise orbit determination method, device and equipment for low-earth orbit satellites based on machine learning. Background Art

[0002] With the continuous development of technologies such as low-earth orbit satellite communication, satellite remote sensing, and low-earth orbit navigation signal enhancement, the role of low-earth orbit satellites in fields such as earth observation, communication, and satellite navigation has become increasingly prominent. High-precision orbit determination is a prerequisite for low-earth orbit satellites to perform various tasks. Therefore, precise orbit determination of low-earth orbit satellites is very important.

[0003] At present, the precise orbit determination technology for low-earth orbit satellites mainly includes on-board GNSS (Global Navigation Satellite System) orbit determination technology and SLR (Satellite Laser Ranging) orbit determination technology. In terms of orbit determination technology: With the continuous enrichment of multi-mode and multi-frequency observation data, the redundancy brought by redundant observations causes uncertainty in orbit determination accuracy and greatly increases the computational amount for on-board GNSS orbit determination technology. Satellite laser ranging has high measurement accuracy, but data loss caused by weather factors such as rain and snow will affect the orbit determination accuracy.

[0004] In addition, the precise orbit determination methods for low-earth orbit satellites mainly include dynamic orbit determination methods, kinematic orbit determination methods, and simplified dynamic orbit determination methods. In terms of orbit determination methods: The dynamic orbit determination method mainly constructs the satellite motion equation according to the mechanical model and continuously corrects the orbit using observation data. This method is greatly affected by the accuracy of the mechanical model. The kinematic orbit determination method completely relies on the geometric information of observation data for orbit determination. The simplified dynamic orbit determination method is between the dynamic method and the kinematic method, and pseudo-random pulse parameters are added during the orbit determination process to balance the dynamic information and the geometric information of observation data. Therefore, it has higher orbit determination accuracy, but the setting of pseudo-random pulse parameters generally uses empirical values and is greatly affected by subjective consciousness.

[0005] In summary, the orbit determination accuracy of existing precise orbit determination technologies / methods for low-earth orbit satellites needs to be improved. Summary of the Invention

[0006] In view of this, the purpose of the present invention is to provide a precise orbit determination method, device and equipment for low-earth orbit satellites based on machine learning, which effectively solves the problems of redundant on-board GNSS data and data loss of satellite laser ranging in rainy and snowy weather, and avoids the uncertainty of the selection of pseudo-random pulse parameters, thereby effectively improving the orbit determination accuracy of low-earth orbit satellites.

[0007] In the first aspect, the present invention provides a precise orbit determination method for low-earth orbit satellites based on machine learning, including:

[0008] Obtain on-board GNSS observation data and satellite laser ranging data observed by low Earth orbit satellites;

[0009] Establish an on-board GNSS heterogeneous multi-mode particle swarm model for the on-board GNSS observation data, and iteratively screen the on-board GNSS observation data through the on-board GNSS heterogeneous multi-mode particle swarm model to obtain the target on-board GNSS observation data participating in the precise orbit determination solution of the low Earth orbit satellite for each epoch;

[0010] In addition, complete the satellite laser ranging data through a pre-trained deep learning neural network to obtain the target satellite laser ranging data corresponding to the observable arc segment of the low Earth orbit satellite by the satellite laser ranging ground station;

[0011] In addition, construct an action set based on the attribute data corresponding to the low Earth orbit satellite, and output the state set corresponding to the action set through a pre-trained inverse reinforcement learning model, where the state set includes the pseudo-random pulse parameters corresponding to the low Earth orbit satellite;

[0012] Determine the precise orbit determination result corresponding to the low Earth orbit satellite according to the target on-board GNSS observation data, the target satellite laser ranging data and the pseudo-random pulse parameters.

[0013] In one implementation manner, establishing an on-board GNSS heterogeneous multi-mode particle swarm model for the on-board GNSS observation data includes:

[0014] Determine the quality index parameters corresponding to the on-board GNSS observation satellite data, where the quality index parameters include multipath index parameters, cycle slip index parameters and signal-to-noise ratio index parameters;

[0015] Establish an on-board GNSS heterogeneous multi-mode particle swarm model based on the multipath index parameters, cycle slip index parameters and signal-to-noise ratio index parameters.

[0016] In one implementation manner, iteratively screening the on-board GNSS observation data through the on-board GNSS heterogeneous multi-mode particle swarm model to obtain the target on-board GNSS observation data participating in the precise orbit determination solution of the low Earth orbit satellite for each epoch includes:

[0017] Perform the following operations on the on-board GNSS observation data observed by the low Earth orbit satellite for multiple GNSS satellites for each epoch:

[0018] Based on the on-board GNSS observation data observed by the low Earth orbit satellite for multiple GNSS satellites in this epoch, generate the initial population corresponding to this epoch. Different particles in the initial population include the on-board GNSS observation data observed by the low Earth orbit satellite for multiple different GNSS satellites and their corresponding quality index parameters;

[0019] Through the spaceborne GNSS heterogeneous multi - mode particle swarm model, the particles included in the initial population corresponding to this epoch are iteratively screened to obtain the target spaceborne GNSS observation data for precise orbit determination of low - earth orbit satellites in this epoch.

[0020] In one implementation, through the spaceborne GNSS heterogeneous multi - mode particle swarm model, the particles included in the initial population corresponding to this epoch are iteratively screened to obtain the target spaceborne GNSS observation data for precise orbit determination of low - earth orbit satellites in this epoch, including:

[0021] Based on the quality index parameters corresponding to the spaceborne GNSS observation data obtained by the low - earth orbit satellite included in the particle for each GNSS satellite observation, determine the fitness value corresponding to the particle;

[0022] If the fitness value corresponding to the particle is better than the preset individual best value and global best value, then use the fitness value corresponding to the particle to update the individual best value and global best value;

[0023] According to the updated individual best value and global best value, update the particle velocity and particle position corresponding to the particle;

[0024] Determine the next particle according to the updated particle velocity and particle position, and based on the quality index parameters corresponding to the spaceborne GNSS observation data obtained by the low - earth orbit satellite included in the next particle for each GNSS satellite observation, determine the fitness value corresponding to the next particle until the preset iteration stop condition is met, and obtain the target spaceborne GNSS observation data for precise orbit determination of low - earth orbit satellites in this epoch; where the iteration stop condition is that the global best value is less than the preset threshold.

[0025] In one implementation, based on the quality index parameters corresponding to the spaceborne GNSS observation data obtained by each low - earth orbit satellite included in the particle, determine the fitness value corresponding to the particle, including:

[0026] Determine the fitness value corresponding to the particle according to the following formula:

[0027] ;

[0028] Where, is the fitness value corresponding to the th particle , is the weighting factor, is the multipath index parameter corresponding to the spaceborne GNSS observation data obtained by the low - earth orbit satellite in the particle for the th GNSS satellite observation, , is the multipath index parameter corresponding to the spaceborne GNSS observation data obtained by the low - earth orbit satellite in the particle for the th GNSS satellite observation, The cycle slip index parameters corresponding to the on-board GNSS observation data obtained from a GNSS satellite are particles The signal-to-noise ratio index parameters corresponding to the on-board GNSS observation data obtained by the medium and low Earth orbit satellite for the th GNSS satellite

[0029] In one implementation, the laser ranging data is complemented through a pre-trained deep learning neural network to obtain the target laser ranging data corresponding to the observable arc of the low Earth orbit satellite by the laser ranging ground station, including:

[0030] Determine the observable arc of the low Earth orbit satellite by the laser ranging ground station;

[0031] Normalize the laser ranging data to obtain the normalized laser ranging data;

[0032] Through the pre-trained deep learning neural network, based on the normalized laser ranging data, complement the missing laser ranging data within the observable arc to obtain the target laser ranging data corresponding to the observable arc of the low Earth orbit satellite by the laser ranging ground station.

[0033] In one implementation, the action set is constructed based on the low Earth orbit height, low Earth orbit mass, low Earth orbit solar panel area, and orbit resolution time included in the attribute data corresponding to the low Earth orbit satellite; before the state set corresponding to the action set is output through the pre-trained inverse reinforcement learning model, the method further includes:

[0034] Output the historical state set corresponding to the historical action set through the inverse reinforcement learning model;

[0035] Generate an expert policy trajectory based on the historical action set and the historical state set;

[0036] Construct an incentive function according to the expert policy trajectory to train the inverse reinforcement learning model using the incentive function, and the trained inverse reinforcement learning model is used to output the state set corresponding to the action set; where the expression of the incentive function is as follows:

[0037] ;

[0038] where is the incentive function, is the th expert policy trajectory, is the th action in the th expert policy trajectory, , is the The th and th states in the expert policy trajectory, is the weight of the incentive function, is a constant, is the state feature,

[0039] In one implementation, according to the target spaceborne GNSS observation data, target laser satellite ranging data, and pseudo-random pulse parameters, the precise orbit determination result corresponding to the low-earth orbit satellite is determined, including:

[0040] Constructing a spaceborne GNSS carrier phase observation equation corresponding to the low-earth orbit satellite based on the target spaceborne GNSS observation data; and constructing a laser satellite ranging observation equation corresponding to the low-earth orbit satellite based on the target laser satellite ranging data; and constructing a satellite motion equation corresponding to the low-earth orbit satellite based on the pseudo-random pulse parameters;

[0041] Combining the spaceborne GNSS carrier phase observation equation, the laser satellite ranging observation equation, and the satellite motion equation to determine the precise orbit determination result corresponding to the low-earth orbit satellite, and the precise orbit determination result includes the satellite position coordinates and the satellite velocity.

[0042] In a second aspect, the present invention also provides a precise orbit determination device for a low-earth orbit satellite based on machine learning, including:

[0043] A data acquisition module for acquiring spaceborne GNSS observation data and laser satellite ranging data observed by the low-earth orbit satellite;

[0044] A GNSS screening module for establishing a spaceborne GNSS heterogeneous multi-modal particle swarm model for the spaceborne GNSS observation data, and iteratively screening the spaceborne GNSS observation data through the spaceborne GNSS heterogeneous multi-modal particle swarm model to obtain the target spaceborne GNSS observation data participating in the precise orbit determination calculation of the low-earth orbit satellite for each epoch;

[0045] A laser satellite ranging completion module for completing the laser satellite ranging data through a pre-trained deep learning neural network to obtain the target laser satellite ranging data corresponding to the observable arc segment of the laser satellite ranging ground station for the low-earth orbit satellite;

[0046] A pseudo-random pulse generation module for constructing an action set based on the attribute data corresponding to the low-earth orbit satellite, and outputting a state set corresponding to the action set through a pre-trained inverse reinforcement learning model, and the state set includes the pseudo-random pulse parameters corresponding to the low-earth orbit satellite;

[0047] A precise orbit determination module for determining the precise orbit determination result corresponding to the low-earth orbit satellite according to the target spaceborne GNSS observation data, the target laser satellite ranging data, and the pseudo-random pulse parameters.

[0048] In a third aspect, the present invention further provides an electronic device, including a processor and a memory. The memory stores computer-executable instructions that can be executed by the processor, and the processor executes the computer-executable instructions to implement the method according to any one of the first aspect.

[0049] A precise orbit determination method, device and equipment for low-earth orbit satellites based on machine learning provided by the present invention first obtain on-board GNSS observation data and satellite laser ranging data observed by low-earth orbit satellites; then establish an on-board GNSS heterogeneous multi-modal particle swarm model for the on-board GNSS observation data, and perform iterative screening on the on-board GNSS observation data through the on-board GNSS heterogeneous multi-modal particle swarm model to obtain the target on-board GNSS observation data participating in the precise orbit determination solution of the low-earth orbit satellite for each epoch; and, complete the satellite laser ranging data through a pre-trained deep learning neural network to obtain the target satellite laser ranging data corresponding to the observable arc segment of the low-earth orbit satellite by the satellite laser ranging ground station; and, construct an action set based on the attribute data corresponding to the low-earth orbit satellite, and output the state set corresponding to the action set through a pre-trained inverse reinforcement learning model, where the state set includes the pseudo-random pulse parameters corresponding to the low-earth orbit satellite; finally, determine the precise orbit determination result corresponding to the low-earth orbit satellite according to the target on-board GNSS observation data, the target satellite laser ranging data and the pseudo-random pulse parameters. The above method uses the on-board GNSS heterogeneous multi-modal particle swarm model to perform iterative screening on the on-board GNSS observation data, constructs a deep learning neural network model to complete the satellite laser ranging data, constructs an inverse reinforcement learning model to estimate the pseudo-random pulse parameters, and on this basis, fuses the target on-board GNSS observation data and the target satellite laser ranging data, and introduces the pseudo-random pulse parameters into the satellite motion equation for precise orbit determination of the low-earth orbit satellite, effectively solving the problems of redundancy of on-board GNSS data and missing of satellite laser ranging data in rainy and snowy weather, and avoiding the uncertainty of the selection of pseudo-random pulse parameters, thereby effectively improving the orbit determination accuracy of the low-earth orbit satellite.

[0050] Other features and advantages of the present invention will be described in the following specification, and, in part, will be obvious from the specification, or will be understood by implementing the present invention. The objectives and other advantages of the present invention are achieved and obtained by the structures specifically pointed out in the specification, claims and drawings.

[0051] To make the above objectives, features and advantages of the present invention more obvious and understandable, the following specific preferred embodiments are given, and in conjunction with the accompanying drawings, the detailed description is as follows. Description of the Drawings

[0052] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0053] Figure 1 Schematic flow chart of a precise orbit determination method for low-earth orbit satellites based on machine learning provided by an embodiment of the present invention;

[0054] Figure 2 Overall schematic flow chart of a precise orbit determination method for low-earth orbit satellites based on machine learning provided by an embodiment of the present invention;

[0055] Figure 3 Schematic structural diagram of a precise orbit determination device for low-earth orbit satellites based on machine learning provided by an embodiment of the present invention;

[0056] Figure 4 Schematic structural diagram of an electronic device provided by an embodiment of the present invention. Specific embodiments

[0057] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions of the present invention in conjunction with the embodiments. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0058] Currently, the orbit determination accuracy of existing precise orbit determination technologies / methods for low-earth orbit satellites needs to be improved. Based on this, the embodiments of the present invention provide a precise orbit determination method, device, and equipment for low-earth orbit satellites based on machine learning, effectively solving the problems of redundancy of on-board GNSS data and missing laser ranging data in rainy and snowy weather, and avoiding the uncertainty of the selection of pseudo-random pulse parameters, thereby effectively improving the orbit determination accuracy of low-earth orbit satellites.

[0059] To facilitate the understanding of this embodiment, first, a precise orbit determination method for low-earth orbit satellites based on machine learning disclosed in the embodiments of the present invention will be introduced in detail. Refer to Figure 1 The schematic flow chart of a precise orbit determination method for low-earth orbit satellites based on machine learning shown, and this method mainly includes the following steps S102 to step S110:

[0060] Step S102, obtain on-board GNSS observation data and laser ranging data observed by the low-earth orbit satellite.

[0061] Step S104: Establish an on-board GNSS heterogeneous multi-modal particle swarm model for on-board GNSS observation data, and perform iterative screening on the on-board GNSS observation data through the on-board GNSS heterogeneous multi-modal particle swarm model to obtain the target on-board GNSS observation data for precise orbit determination of low-Earth orbit satellites at each epoch.

[0062] In one example, corresponding quality index parameters can be determined based on the on-board GNSS observation data, including multipath index parameters, cycle slip index parameters, and signal-to-noise ratio index parameters. On this basis, an on-board GNSS heterogeneous multi-modal particle swarm model is constructed. Through the on-board GNSS heterogeneous multi-modal particle swarm model, based on the multipath index parameters, cycle slip index parameters, and signal-to-noise ratio index parameters, iterative screening is performed on the on-board GNSS observation data obtained by observing multiple GNSS satellites for low-Earth orbit satellites at each epoch to obtain the target on-board GNSS observation data for precise orbit determination of low-Earth orbit satellites at each epoch.

[0063] Step S106: Complement the laser ranging data of satellites through a pre-trained deep learning neural network to obtain the target laser ranging data corresponding to the observable arc segment of the low-Earth orbit satellite by the laser ranging ground station.

[0064] In one example, the observable arc segment of the low-Earth orbit satellite by the laser ranging ground station is determined. After normalizing the laser ranging data, through the pre-trained deep learning neural network, based on the normalized laser ranging data, the missing laser ranging data in the observable arc segment is complemented to obtain the target laser ranging data corresponding to the observable arc segment of the low-Earth orbit satellite by the laser ranging ground station.

[0065] Step S108: Construct an action set based on the attribute data corresponding to the low-Earth orbit satellite, and output the state set corresponding to the action set through a pre-trained inverse reinforcement learning model. The state set includes the pseudo-random pulse parameters corresponding to the low-Earth orbit satellite.

[0066] Among them, the attribute data includes the orbital altitude of the low-Earth orbit satellite, the mass of the low-Earth orbit satellite, the area of the solar panels of the low-Earth orbit satellite, and the orbit determination time. That is, the action set is constructed based on the orbital altitude of the low-Earth orbit satellite, the mass of the low-Earth orbit satellite, the area of the solar panels of the low-Earth orbit satellite, and the orbit determination time included in the attribute data corresponding to the low-Earth orbit satellite. The action set is input into the pre-trained inverse reinforcement learning model so that the inverse reinforcement learning model outputs the state set corresponding to the action set to realize the estimation of the pseudo-random pulse parameters.

[0067] Step S110: Determine the precise orbit determination result corresponding to the low-Earth orbit satellite according to the target on-board GNSS observation data, the target laser ranging data, and the pseudo-random pulse parameters.

[0068] Among them, the precise orbit determination results include the satellite position coordinates and satellite velocity of the low-earth orbit satellite. In one example, according to the target on-board GNSS observation data, target laser ranging data to satellites, and pseudo-random pulse parameters, the on-board GNSS carrier phase observation equation, laser ranging observation equation to satellites, and satellite motion equation are respectively constructed, and the corresponding precise orbit determination results of the low-earth orbit satellite can be obtained by combining the above three equations.

[0069] The method for precise orbit determination of a low-earth orbit satellite based on machine learning provided by the embodiments of the present invention uses an on-board GNSS heterogeneous multi-mode particle swarm model to iteratively screen the on-board GNSS observation data, constructs a deep learning neural network model to complement the laser ranging data to satellites, constructs an inverse reinforcement learning model to estimate the pseudo-random pulse parameters, and on this basis, fuses the target on-board GNSS observation data and the target laser ranging data to satellites, introduces the pseudo-random pulse parameters into the satellite motion equation for precise orbit determination of the low-earth orbit satellite, effectively solves the problems of redundancy of on-board GNSS data and missing of laser ranging data to satellites in rainy and snowy weather, and avoids the uncertainty of the selection of pseudo-random pulse parameters, thereby effectively improving the orbit determination accuracy of the low-earth orbit satellite.

[0070] For ease of understanding, the embodiments of the present invention provide a specific implementation manner of a method for precise orbit determination of a low-earth orbit satellite based on machine learning. Refer to Figure 2 the overall flow schematic diagram of a method for precise orbit determination of a low-earth orbit satellite based on machine learning shown in the figure, including: (1) obtaining the on-board GNSS observation data and laser ranging data to satellites of the low-earth orbit satellite; (2) screening the multi-mode and multi-frequency on-board GNSS observation data based on the heterogeneous multi-mode particle swarm algorithm to obtain the target on-board GNSS observation data for orbit determination calculation at each epoch; (3) constructing a deep learning neural network to complement the missing laser ranging data to satellites to obtain the target laser ranging data to satellites corresponding to the observable arc segment of the laser ranging ground station to the low-earth orbit satellite; (4) constructing an inverse reinforcement learning model to estimate the pseudo-random pulse parameters; (5) based on the target on-board GNSS observation data, target laser ranging data to satellites, and pseudo-random pulse parameters, using the simplified dynamics method for precise orbit determination of the low-earth orbit satellite.

[0071] (1) Obtain the on-board GNSS observation data and laser ranging data to satellites of the low-earth orbit satellite.

[0072] (2) Screen the multi-mode and multi-frequency on-board GNSS observation data based on the heterogeneous multi-mode particle swarm algorithm to obtain the target on-board GNSS observation data for orbit determination calculation at each epoch.

[0073] (2.1) Determine the quality index parameters corresponding to the on-board GNSS observation satellite data, and the quality index parameters include multi-path index parameters, cycle slip index parameters, and signal-to-noise ratio index parameters.

[0074] In one example, the calculation process of the multi-path index parameters is as follows:

[0075] ;

[0076] Wherein, is the multipath index parameter, is the GNSS satellite PRN number; is the on-board GNSS pseudorange observation; , are respectively and the carrier wavelengths of the frequency points; is the on-board GNSS carrier observation; , , are respectively and the carrier frequencies of the frequency points.

[0077] In one example, the calculation process of the cycle slip index parameter is as follows:

[0078] ;

[0079] ;

[0080] ;

[0081] Wherein, is the cycle slip flag, 1 indicates a cycle slip occurs, 0 indicates no cycle slip occurs, is the GNSS satellite PRN number; is the geometric-free observation; is the MW combination observation; is the on-board GNSS carrier observation; is the on-board GNSS pseudorange observation; , are respectively and the carrier wavelengths of the frequency points; , are respectively and the carrier frequencies of the frequency points; , are two adjacent epoch times.

[0082] In one example, the calculation process of the signal-to-noise ratio index parameter is as follows:

[0083] ;

[0084] Wherein, is the signal-to-noise ratio value; is the carrier power density; is the noise power density.

[0085] (2.2) Based on the multipath index parameter, cycle slip index parameter, and signal-to-noise ratio index parameter, establish a spaceborne GNSS heterogeneous multi-mode particle swarm model. Through the spaceborne GNSS heterogeneous multi-mode particle swarm model, iteratively screen the spaceborne GNSS observation data to obtain the target spaceborne GNSS observation data for precise orbit determination of the low Earth orbit satellite at each epoch.

[0086] For the spaceborne GNSS observation data obtained by the low Earth orbit satellite observing multiple GNSS satellites at each epoch, perform the following steps a to b:

[0087] Step a: Based on the spaceborne GNSS observation data obtained by the low Earth orbit satellite observing multiple GNSS satellites at this epoch, generate the initial population corresponding to this epoch. Different particles in the initial population include the spaceborne GNSS observation data obtained by the low Earth orbit satellite observing multiple different GNSS satellites and their corresponding quality index parameters.

[0088] In one example, at epoch a total of GNSS satellites are observed. Select the observation data of 6 GNSS satellites to generate an initial population with a scale of . The th particle in the initial population is: ;

[0089] ;

[0090] where , , …, represent the data (including spaceborne GNSS observation data and their corresponding quality index parameters) obtained by the low Earth orbit satellite observing the 1st, 2nd, …, 6th GNSS satellites in the th particle, , , …, represent the multipath index parameters corresponding to the spaceborne GNSS observation data obtained by the low Earth orbit satellite observing the 1st, 2nd, …, 6th GNSS satellites in the th particle, , , …, represent the cycle slip index parameters corresponding to the spaceborne GNSS observation data obtained by the low Earth orbit satellite observing the 1st, 2nd, …, 6th GNSS satellites in the th particle, , , …, represent the signal-to-noise ratio index parameters corresponding to the spaceborne GNSS observation data obtained by the low Earth orbit satellite observing the 1st, 2nd, …, 6th GNSS satellites in the The SNR index parameters corresponding to the on-board GNSS observation data obtained by the LEO satellite for the 1st, 2nd, …, 6th GNSS satellites among the particles.

[0091] In addition, initialize the individual optimal value and the global optimal value , and initialize the particle velocity and particle position.

[0092] Step b: Through the on-board GNSS heterogeneous multi-modal particle swarm model, iteratively screen the particles included in the initial population corresponding to this epoch to obtain the target on-board GNSS observation data for the precise orbit determination of the LEO satellite in this epoch. Specifically, it includes the following steps b1 to b4:

[0093] Step b1: Based on the quality index parameters corresponding to the on-board GNSS observation data obtained by the LEO satellite included in the particle for each GNSS satellite, determine the fitness value corresponding to the particle.

[0094] Determine the fitness value corresponding to the particle according to the following formula:

[0095] ;

[0096] where is the fitness value corresponding to the th particle , is the weighting factor, and its value is the PDOP (Position Dilution of Precision) value of the 6 GNSS satellites that make up the particle, is the multipath index parameter corresponding to the on-board GNSS observation data obtained by the LEO satellite in the particle for the th GNSS satellite, is the cycle slip index parameter corresponding to the on-board GNSS observation data obtained by the LEO satellite in the particle for the th GNSS satellite, is the SNR index parameter corresponding to the on-board GNSS observation data obtained by the LEO satellite in the particle for the th GNSS satellite.

[0097] Step b2: If the fitness value corresponding to the particle is better than the preset individual optimal value and global optimal value, then update the individual optimal value and global optimal value using the fitness value corresponding to the particle.

[0098] In one example, the particle fitness value is compared with the individual optimal value. If the particle fitness value is better than the individual optimal value, the particle fitness value is updated to the individual optimal value. The particle fitness value is compared with the global optimal value. If the particle fitness value is better than the global optimal value, the particle fitness value is updated to the global optimal value.

[0099] Step b3: Update the particle velocity and particle position corresponding to the particle according to the updated individual optimal value and global optimal value.

[0100] In one example, the particle velocity and particle position can be updated according to the following function:

[0101] ;

[0102] In the formula, is the number of iterations, represents the th GNSS satellite; is the weight factor; is the particle velocity; is the particle position; and are random factors, and their values are between and.

[0103] Step b4: Determine the next particle according to the updated particle velocity and particle position, and repeat steps b1 to b3 until the preset iteration stop condition is satisfied, so as to obtain the target on-board GNSS observation data for precise orbit determination of the low-earth orbit satellite in this epoch.

[0104] Among them, the iteration stop condition is that the global optimal value is less than the preset threshold. For example, set the iteration condition to .

[0105] (3) Construct a deep learning neural network to complete the missing laser satellite ranging data, and obtain the target laser satellite ranging data corresponding to the observable arc segment of the low-earth orbit satellite by the laser satellite ranging ground station.

[0106] (3.1) Determine the observable arc segment of the low-earth orbit satellite by the laser satellite ranging ground station.

[0107] In one example, the orbit of the low-earth orbit satellite is determined by solving the laser satellite ranging data to obtain the position coordinates of the low-earth orbit satellite. Based on the coordinates of the laser satellite ranging ground station and the position coordinates of the low-earth orbit satellite, the altitude angle of the low-earth orbit satellite relative to the laser satellite ranging ground station is calculated, and the cut-off altitude angle is set to 10 degrees, so as to obtain the observable arc of the low-earth orbit satellite by the laser satellite ranging ground station .

[0108] (3.2) Normalize the laser satellite ranging data to obtain the normalized laser satellite ranging data.

[0109] In one case, the laser ranging data to satellites is normalized according to the following formula:

[0110] ;

[0111] where is the time series of the normalized laser ranging data to satellites, ; is the original sequence, ; denotes taking the expectation.

[0112] (3.3) Based on the normalized laser ranging data to satellites, the missing laser ranging data within the observable arc segment is completed through a pre-trained deep learning neural network, and the corresponding target laser ranging data within the observable arc segment of the ground station for low-earth orbit satellites is obtained.

[0113] In one example, a single-layer deep learning neural network can be constructed to complete the laser ranging data:

[0114] ;

[0115] where is the predicted value; , , are all weight matrices; is the memory variable; is the bias value; is the neuron activation function, and the Relu function is adopted.

[0116] Furthermore, pre-training is required before using the above single-layer deep learning neural network to complete the laser ranging data. In the above neuron training, there is a bias value, which is represented by the mean square error:

[0117] ;

[0118] where is the number of training samples; is the true value of the laser ranging; is the predicted value of the laser ranging.

[0119] The bias value is used as the training end constraint condition, and its threshold is set to 0.999. If the bias value is less than or equal to the threshold, it is considered convergent and the training ends.

[0120] (IV) Construct an inverse reinforcement learning model to estimate the pseudo-random pulse parameters.

[0121] First, the state set and action set are explained:

[0122] Set of states : , that is, the set of states consists of pseudo-random pulse parameters ;

[0123] Set of actions : . The elements of the set of actions are determined by the low-earth orbit satellite altitude, satellite mass, satellite solar panel area, and orbit resolution time. The calculation formula is:

[0124] ;

[0125] ;

[0126] In the formula, is the low-earth orbit satellite mass; is the low-earth orbit satellite altitude; is the low-earth orbit satellite solar panel area; is a constant coefficient determined by the orbit resolution time; is the orbit resolution time, expressed in days of the year.

[0127] Based on the above set of states and set of actions , an embodiment of the present invention provides a specific process for constructing an inverse reinforcement learning model, including:

[0128] (4.1) Output the historical state set corresponding to the historical action set through the inverse reinforcement learning model.

[0129] (4.2) Generate an expert policy trajectory based on the historical action set and the historical state set. The expression of the expert policy trajectory is as follows:

[0130] ; In the formula, is the expert policy trajectory.

[0131] (4.3) Construct an incentive function according to the expert policy trajectory to use the incentive function to train the inverse reinforcement learning model. The trained inverse reinforcement learning model is used to output the state set corresponding to the action set.

[0132] Among them, the expression of the incentive function is as follows:

[0133] ;

[0134] Among them, is the incentive function, is the th expert policy trajectory, is the The th action in and are the th and th states in the th expert policy trajectory. is the weight of the activation function, is a constant, is the state feature, is the state transition function.

[0135] By iterating over all expert trajectories, the final activation function is obtained. Based on the activation function, a set of states composed of pseudo-random pulse parameters can be output for the input action set, thereby estimating the pseudo-random pulse parameters required in orbit determination. .

[0136] (5) Based on the target spaceborne GNSS observation data, target laser satellite ranging data, and pseudo-random pulse parameters, the simplified dynamics method is used for precise orbit determination of low Earth orbit satellites.

[0137] (5.1) Construct the spaceborne GNSS carrier phase observation equation corresponding to the low Earth orbit satellite according to the target spaceborne GNSS observation data; and construct the laser satellite ranging observation equation corresponding to the low Earth orbit satellite according to the target laser satellite ranging data; and construct the satellite motion equation corresponding to the low Earth orbit satellite according to the pseudo-random pulse parameters.

[0138] Construct the spaceborne GNSS carrier phase observation equation:

[0139] ;

[0140] In the formula, is the spaceborne GNSS carrier phase observation value of the low Earth orbit satellite; is the geometric distance from the low Earth orbit satellite to the GNSS satellite; is the tropospheric mapping function; is the tropospheric zenith wet delay; is the ionospheric delay; is the carrier wavelength; is the spaceborne GNSS phase ambiguity of the low Earth orbit satellite; is the spaceborne GNSS carrier phase noise; is the speed of light in vacuum; is the GNSS satellite clock error; is the spaceborne GNSS receiver clock error of the low Earth orbit satellite.

[0141] Construct the laser satellite ranging observation equation:

[0142] ;

[0143] wherein, is the observation value of satellite laser ranging for low Earth orbit satellites; is the geometric distance between the laser retroreflector of the low Earth orbit satellite and the ground station for satellite laser ranging; tropospheric delay correction; is the correction value between the centroid of the low Earth orbit satellite and the geometric center of the laser retroreflector; is the eccentricity correction of the ground station for satellite laser ranging; is the measurement noise.

[0144] Construct the satellite motion equation:

[0145] ;

[0146] wherein, is the acceleration of the low Earth orbit satellite in the inertial system; is the position of the low Earth orbit satellite in the inertial system; is the universal gravitational constant of the Earth; is the perturbed acceleration by perturbations.

[0147] Furthermore, add the pseudo-random pulse parameter to the above satellite motion equation to absorb the perturbed acceleration generated by non-conservative perturbing forces.

[0148] (5.2) Combine the on-board GNSS carrier phase observation equation, the satellite laser ranging observation equation and the satellite motion equation to determine the precise orbit determination result corresponding to the low Earth orbit satellite. The precise orbit determination result includes the satellite position coordinates and the satellite velocity.

[0149] In one example, combine the satellite motion equation, the on-board GNSS carrier phase observation equation, and the satellite laser ranging observation equation, and use the least squares method to estimate the position coordinates and velocity of the low Earth orbit satellite.

[0150] In summary, the on-board GNSS observation data is developing towards the multi-mode and multi-frequency direction. How to reduce the computational amount while ensuring the accuracy is a problem to be considered; the satellite laser ranging data has high-precision measurement accuracy, but data loss is prone to occur due to weather influence; the determination of the pseudo-random pulse parameter in the simplified dynamics method often relies on empirical values and cannot accurately adapt to each satellite. In view of the above problems, the embodiment of the present invention proposes a method for precise orbit determination of low Earth orbit satellites based on machine learning. First, screen the on-board GNSS observation data based on the heterogeneous multi-mode particle swarm algorithm to obtain the on-board GNSS observation data participating in the calculation; then construct a deep learning neural network to complete the satellite laser ranging data missing due to weather reasons; then construct an inverse reinforcement learning model to estimate the pseudo-random pulse parameter; finally, realize the precise orbit determination of the low Earth orbit satellite based on the simplified dynamics method, which can effectively solve the above problems.

[0151] Based on the foregoing embodiments, an embodiment of the present invention provides a precise orbit determination device for low-Earth orbit satellites based on machine learning. Refer to Figure 3 the structural schematic diagram of a precise orbit determination device for low-Earth orbit satellites based on machine learning shown in

[0152] A data acquisition module 302, configured to acquire on-board GNSS observation data and laser ranging data observed by the low-Earth orbit satellite;

[0153] A GNSS screening module 304, configured to establish an on-board GNSS heterogeneous multi-mode particle swarm model for the on-board GNSS observation data, and iteratively screen the on-board GNSS observation data through the on-board GNSS heterogeneous multi-mode particle swarm model to obtain the target on-board GNSS observation data participating in the precise orbit determination solution of the low-Earth orbit satellite for each epoch;

[0154] A laser ranging data completion module 306, configured to complete the laser ranging data through a pre-trained deep learning neural network to obtain the target laser ranging data corresponding to the observable arc segment of the laser ranging ground station for the low-Earth orbit satellite;

[0155] A pseudo-random pulse generation module 308, configured to construct an action set based on the attribute data corresponding to the low-Earth orbit satellite, and output a state set corresponding to the action set through a pre-trained inverse reinforcement learning model, where the state set includes pseudo-random pulse parameters corresponding to the low-Earth orbit satellite;

[0156] A precise orbit determination module 310, configured to determine the precise orbit determination result corresponding to the low-Earth orbit satellite according to the target on-board GNSS observation data, the target laser ranging data, and the pseudo-random pulse parameters.

[0157] The precise orbit determination device for low-Earth orbit satellites based on machine learning provided by the embodiment of the present invention uses an on-board GNSS heterogeneous multi-mode particle swarm model to iteratively screen the on-board GNSS observation data, constructs a deep learning neural network model to complete the laser ranging data, constructs an inverse reinforcement learning model to estimate the pseudo-random pulse parameters, and on this basis, fuses the target on-board GNSS observation data and the target laser ranging data, and introduces the pseudo-random pulse parameters into the satellite motion equation for precise orbit determination of the low-Earth orbit satellite, effectively solving the problems of redundancy of on-board GNSS data and missing of laser ranging data in rainy and snowy weather, and avoiding the uncertainty of the selection of pseudo-random pulse parameters, thereby effectively improving the precision of low-Earth orbit satellite orbit determination.

[0158] In an implementation manner, the GNSS screening module 304 includes a heterogeneous multi-mode particle swarm model unit, configured to:

[0159] Determine the quality index parameters corresponding to the on-board GNSS observation satellite data, where the quality index parameters include multipath index parameters, cycle slip index parameters, and signal-to-noise ratio index parameters;

[0160] Based on the multipath index parameters, cycle slip index parameters, and signal-to-noise ratio index parameters, establish an on-board GNSS heterogeneous multi-mode particle swarm model.

[0161] In one implementation, the GNSS screening module 304 includes a data iterative screening unit for:

[0162] Perform the following operations on the on-board GNSS observation data obtained by the low-earth orbit satellite for multiple GNSS satellites at each epoch:

[0163] Based on the on-board GNSS observation data obtained by the low-earth orbit satellite for multiple GNSS satellites at this epoch, generate the initial population corresponding to this epoch. Different particles in the initial population include the on-board GNSS observation data obtained by the low-earth orbit satellite for multiple different GNSS satellites and their corresponding quality index parameters;

[0164] Through the on-board GNSS heterogeneous multi-mode particle swarm model, iteratively screen the particles included in the initial population corresponding to this epoch to obtain the target on-board GNSS observation data for the precise orbit determination of the low-earth orbit satellite at this epoch.

[0165] In one implementation, the data iterative screening unit includes:

[0166] An adaptation value determination subunit for: determining the adaptation value corresponding to a particle based on the quality index parameters corresponding to the on-board GNSS observation data obtained by the low-earth orbit satellite included in the particle for each GNSS satellite;

[0167] An optimal value update subunit for: if the adaptation value corresponding to a particle is better than the preset individual optimal value and global optimal value, then update the individual optimal value and global optimal value using the adaptation value corresponding to the particle;

[0168] A velocity and position update subunit for: updating the particle velocity and particle position corresponding to the particle according to the updated individual optimal value and global optimal value;

[0169] An iteration subunit for: determining the next particle according to the updated particle velocity and particle position, and determining the adaptation value corresponding to the next particle based on the quality index parameters corresponding to the on-board GNSS observation data obtained by the low-earth orbit satellite included in the next particle, until the preset iteration stop condition is met, to obtain the target on-board GNSS observation data for the precise orbit determination of the low-earth orbit satellite at this epoch; where the iteration stop condition is that the global optimal value is less than the preset threshold.

[0170] In one embodiment, the fitness value determination subunit is specifically configured to:

[0171] Determine the fitness value corresponding to the particle according to the following formula:

[0172] ;

[0173] where is the fitness value corresponding to the th particle , is the weighting factor, is the multi-path index parameter corresponding to the on-board GNSS observation data obtained by the low-earth orbit satellite in the particle for the th GNSS satellite, is the cycle slip index parameter corresponding to the on-board GNSS observation data obtained by the low-earth orbit satellite in the particle for the th GNSS satellite, is the signal-to-noise ratio index parameter corresponding to the on-board GNSS observation data obtained by the low-earth orbit satellite in the particle for the th GNSS satellite.

[0174] In one embodiment, the laser satellite ranging data completion module 306 is specifically configured to:

[0175] Determine the observable arc of the laser satellite ranging ground station for the low-earth orbit satellite;

[0176] Normalize the laser satellite ranging data to obtain the normalized laser satellite ranging data;

[0177] Complete the missing laser satellite ranging data within the observable arc based on the normalized laser satellite ranging data through a pre-trained deep learning neural network to obtain the target laser satellite ranging data corresponding to the observable arc of the laser satellite ranging ground station for the low-earth orbit satellite.

[0178] In one embodiment, the action set is constructed based on the low-earth orbit satellite altitude, low-earth orbit satellite mass, low-earth orbit satellite solar panel area, and orbit determination time included in the attribute data corresponding to the low-earth orbit satellite; it further includes an inverse reinforcement learning model training module, which is used to:

[0179] Output the historical state set corresponding to the historical action set through the inverse reinforcement learning model;

[0180] Generate an expert policy trajectory based on the historical action set and the historical state set;

[0181] Construct an incentive function according to the expert policy trajectory to train the inverse reinforcement learning model using the incentive function. The trained inverse reinforcement learning model is used to output the state set corresponding to the action set. The expression of the incentive function is as follows:

[0182] ;

[0183] where, is the incentive function, is the th expert policy trajectory, is the th action in the th expert policy trajectory, , are the th th and th states in the th expert policy trajectory, is the weight of the incentive function, is a constant, is the state feature,

[0184] In one implementation, the precise orbit determination module 310 is specifically configured to:

[0185] Construct the on-board GNSS carrier phase observation equation corresponding to the low-orbit satellite according to the target on-board GNSS observation data; and construct the laser ranging observation equation corresponding to the low-orbit satellite according to the target laser ranging data; and construct the satellite motion equation corresponding to the low-orbit satellite according to the pseudo-random pulse parameters;

[0186] Combine the on-board GNSS carrier phase observation equation, the laser ranging observation equation and the satellite motion equation to determine the precise orbit determination result corresponding to the low-orbit satellite, and the precise orbit determination result includes the satellite position coordinates and the satellite velocity.

[0187] The device provided by the embodiments of the present invention has the same implementation principle and the same technical effects as those of the foregoing method embodiments. For the sake of brief description, for the parts not mentioned in the device embodiments, reference may be made to the corresponding contents in the foregoing method embodiments.

[0188] The embodiments of the present invention provide an electronic device. Specifically, the electronic device includes a processor and a storage device; a computer program is stored on the storage device, and the computer program executes the method according to any one of the foregoing implementation manners when being run by the processor.

[0189] Figure 4A schematic structural diagram of an electronic device provided by an embodiment of the present invention. The electronic device 100 includes: a processor 40, a memory 41, a bus 42, and a communication interface 43. The processor 40, the communication interface 43, and the memory 41 are connected through the bus 42. The processor 40 is configured to execute an executable module stored in the memory 41, such as a computer program.

[0190] Among them, the memory 41 may include a high-speed random access memory (RAM), and may also include a non-volatile memory, such as at least one disk memory. Through at least one communication interface 43 (which can be wired or wireless), a communication connection is established between this system network element and at least one other network element, and the Internet, wide area network, local area network, metropolitan area network, etc. can be used.

[0191] The bus 42 can be an ISA bus, a PCI bus, an EISA bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 4 only a bidirectional arrow is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0192] Among them, the memory 41 is used to store a program. After receiving an execution instruction, the processor 40 executes the program. The method executed by the device defined by the flow process disclosed in any one of the foregoing embodiments of the present invention can be applied to the processor 40 or implemented by the processor 40.

[0193] The processor 40 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in the processor 40 or the instructions in the form of software. The above-mentioned processor 40 may be a general-purpose processor, including a central processing unit (CPU for short), a network processor (NP for short), etc.; it may also be a digital signal processor (DSP for short), an application specific integrated circuit (ASIC for short), a field-programmable gate array (FPGA for short), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components. It can implement or execute various methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc. The steps of the method disclosed in combination with the embodiments of the present invention can be directly embodied as being executed and completed by the hardware decoding processor, or executed and completed by the combination of the hardware and software modules in the decoding processor. The software module may be located in a mature storage medium in the art such as a random access memory, a flash memory, a read-only memory, a programmable read-only memory, or an electrically erasable programmable memory, a register, etc. This storage medium is located in the memory 41, and the processor 40 reads the information in the memory 41 and combines its hardware to complete the steps of the above method.

[0194] The computer program product of the readable storage medium provided by the embodiments of the present invention includes a computer-readable storage medium storing program code, and the instructions included in the program code can be used to execute the methods described in the foregoing method embodiments. For the specific implementation, reference can be made to the foregoing method embodiments and will not be elaborated here.

[0195] When the above-mentioned functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0196] Finally, it should be noted that the above-mentioned embodiments are only specific implementation manners of the present invention, used to illustrate the technical solutions of the present invention, rather than limiting it. The protection scope of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: any person skilled in the art within the technical scope disclosed by the present invention can still modify the technical solutions recorded in the foregoing embodiments or easily conceive of changes, or perform equivalent replacements on some of the technical features; and these modifications, changes, or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered by the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. A method for precise orbit determination of low-orbit satellites based on machine learning, characterized in that: include: Obtain onboard GNSS observation data and laser satellite measurement data of low-orbit satellite observations; A satellite-borne GNSS heterogeneous multi-mode particle swarm model is established for the satellite-borne GNSS observation data, and the satellite-borne GNSS observation data is iteratively screened by the satellite-borne GNSS heterogeneous multi-mode particle swarm model to obtain target satellite-borne GNSS observation data participating in the precise orbit determination solution of the low-orbit satellite at each epoch; And, the laser satellite measurement data is supplemented by a pre-trained deep learning neural network to obtain the target laser satellite measurement data corresponding to the observable arc segment of the laser satellite measurement ground station for the low-orbit satellite; And, constructing an action set based on the attribute data corresponding to the low-orbit satellite, and outputting a state set corresponding to the action set through a pre-trained inverse reinforcement learning model, wherein the state set includes pseudo-random pulse parameters corresponding to the low-orbit satellite; The precise orbit determination result corresponding to the low-orbit satellite is determined according to the target satellite-borne GNSS observation data, the target laser satellite measurement data and the pseudo-random pulse parameters.

2. The method for precise orbit determination of low-orbit satellites based on machine learning according to claim 1, characterized in that: A spaceborne GNSS heterogeneous multi-mode particle swarm model is established for the spaceborne GNSS observation data, including: Determine quality index parameters corresponding to the onboard GNSS observation data, wherein the quality index parameters include a multipath index parameter, a cycle slip index parameter, and a signal-to-noise ratio index parameter; Based on the multipath index parameter, the cycle slip index parameter and the signal-to-noise ratio index parameter, a space-borne GNSS heterogeneous multi-mode particle swarm model is established.

3. The method for precise orbit determination of low-orbit satellites based on machine learning according to claim 2, characterized in that: The onboard GNSS observation data are iteratively screened by the onboard GNSS heterogeneous multi-mode particle swarm model to obtain target onboard GNSS observation data participating in the low-orbit satellite precise orbit determination solution at each epoch, including: The following operations are performed for the onboard GNSS observation data obtained by observing multiple GNSS satellites by a low-orbit satellite in each epoch: Based on the onboard GNSS observation data obtained by the low-orbit satellite observing multiple GNSS satellites at the epoch, generating an initial population corresponding to the epoch, wherein different particles in the initial population include the onboard GNSS observation data obtained by the low-orbit satellite observing multiple different GNSS satellites and the corresponding quality indicator parameters; The particles contained in the initial population corresponding to the epoch are iteratively screened through the onboard GNSS heterogeneous multi-mode particle swarm model to obtain the target onboard GNSS observation data participating in the low-orbit satellite precise orbit determination solution at the epoch.

4. The method for precise orbit determination of low-orbit satellites based on machine learning according to claim 3, characterized in that: The particles contained in the initial population corresponding to the epoch are iteratively screened by the onboard GNSS heterogeneous multi-mode particle swarm model to obtain the target onboard GNSS observation data participating in the low-orbit satellite precise orbit determination solution for the epoch, including: Determining an adaptation value corresponding to the particle based on the quality indicator parameter corresponding to the onboard GNSS observation data obtained by the low-orbit satellite included in the particle for each of the GNSS satellites; If the fitness value corresponding to the particle is better than the preset individual optimal value and the global optimal value, the individual optimal value and the global optimal value are updated using the fitness value corresponding to the particle; According to the updated individual optimal value and the global optimal value, updating the particle speed and particle position corresponding to the particle; The next particle is determined according to the updated particle velocity and the particle position, and the fitness value corresponding to the next particle is determined based on the quality indicator parameter corresponding to the onboard GNSS observation data obtained by the low-orbit satellite included in the next particle for each of the GNSS satellite observations, until a preset iteration stop condition is met, and the target onboard GNSS observation data participating in the low-orbit satellite precise orbit determination solution at this epoch is obtained; wherein the iteration stop condition is that the global optimal value is less than a preset threshold.

5. The method for precise orbit determination of low-orbit satellites based on machine learning according to claim 4, characterized in that: Determining the fitness value corresponding to the particle based on the quality indicator parameter corresponding to the onboard GNSS observation data obtained by observing each of the low-orbit satellites included in the particle, comprises: The fitness value corresponding to the particle is determined according to the following formula: ； in, For the Particles The corresponding fitness value is is the weighting factor, For particles The low-orbit satellite mentioned in the A multipath indicator parameter corresponding to the onboard GNSS observation data obtained by observing the GNSS satellites, For particles The low-orbit satellite mentioned in the A cycle slip index parameter corresponding to the onboard GNSS observation data obtained by observing the GNSS satellites, For particles The low-orbit satellite mentioned in the The signal-to-noise ratio index parameter corresponding to the onboard GNSS observation data obtained by observing the GNSS satellites.

6. The method for precise orbit determination of low-orbit satellites based on machine learning according to claim 1, characterized in that: The laser satellite measurement data is supplemented by a pre-trained deep learning neural network to obtain target laser satellite measurement data corresponding to the observable arc of the laser satellite measurement ground station for the low-orbit satellite, including: Determine the observable arc of the laser satellite ground station to the low-orbit satellite; Normalizing the laser satellite measurement data to obtain normalized laser satellite measurement data; Through the deep learning neural network obtained by pre-training, based on the normalized laser satellite measurement data, the missing laser satellite measurement data in the observable arc segment is supplemented to obtain the target laser satellite measurement data corresponding to the observable arc segment of the low-orbit satellite by the laser satellite measurement ground station.

7. The method for precise orbit determination of low-orbit satellites based on machine learning according to claim 1, characterized in that: The action set is constructed based on the low-orbit satellite orbit height, low-orbit satellite mass, low-orbit satellite solar sail panel area and orbit solution time contained in the attribute data corresponding to the low-orbit satellite; Before outputting the state set corresponding to the action set through the pre-trained inverse reinforcement learning model, the method further includes: Output the historical state set corresponding to the historical action set through the inverse reinforcement learning model; Generate an expert strategy trajectory based on the historical action set and the historical state set; An activation function is constructed according to the expert strategy trajectory to train the inverse reinforcement learning model using the activation function. The trained inverse reinforcement learning model is used to output a state set corresponding to the action set. The expression of the activation function is as follows: ； in, is the activation function, For the Expert strategy trajectories, For the The expert strategy trajectory Actions, , For the The expert strategy trajectory , status, is the activation function weight, is a constant, is the state feature, is the state transfer function.

8. The method for precise orbit determination of low-orbit satellites based on machine learning according to claim 1, characterized in that: Determining a precise orbit determination result corresponding to the low-orbit satellite according to the target onboard GNSS observation data, the target laser satellite measurement data and the pseudo-random pulse parameters, including: Constructing a satellite-borne GNSS carrier phase observation equation corresponding to the low-orbit satellite according to the target satellite-borne GNSS observation data; and constructing a laser satellite measurement observation equation corresponding to the low-orbit satellite according to the target laser satellite measurement data; and constructing a satellite motion equation corresponding to the low-orbit satellite according to the pseudo-random pulse parameters; The onboard GNSS carrier phase observation equation, the laser satellite measurement observation equation and the satellite motion equation are combined to determine the precise orbit determination result corresponding to the low-orbit satellite, wherein the precise orbit determination result includes the satellite position coordinates and the satellite velocity.

9. A low-orbit satellite precise orbit determination device based on machine learning, characterized in that: include: The data acquisition module is used to obtain the onboard GNSS observation data and laser satellite measurement data of low-orbit satellite observations; A GNSS screening module is used to establish a satellite-borne GNSS heterogeneous multi-mode particle swarm model for the satellite-borne GNSS observation data, and iteratively screen the satellite-borne GNSS observation data through the satellite-borne GNSS heterogeneous multi-mode particle swarm model to obtain target satellite-borne GNSS observation data participating in the precise orbit determination solution of low-orbit satellites at each epoch; A laser satellite measurement and completion module is used to complete the laser satellite measurement and completion data through a pre-trained deep learning neural network to obtain the target laser satellite measurement and completion data corresponding to the observable arc segment of the laser satellite measurement and completion ground station for the low-orbit satellite; A pseudo-random pulse generation module, used to construct an action set based on the attribute data corresponding to the low-orbit satellite, and output a state set corresponding to the action set through a pre-trained inverse reinforcement learning model, wherein the state set includes pseudo-random pulse parameters corresponding to the low-orbit satellite; The precise orbit determination module is used to determine the precise orbit determination result corresponding to the low-orbit satellite according to the target satellite-borne GNSS observation data, the target laser satellite measurement data and the pseudo-random pulse parameters.

10. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores computer executable instructions that can be executed by the processor, and the processor executes the computer executable instructions to implement the method according to any one of claims 1 to 8.

Images