Spacecraft flight attitude angle correction method and device and electronic equipment

Abstract

Embodiments of the present application provide a spacecraft flight attitude angle correction method and device and electronic equipment. The method comprises: using a target correction model to process a real-time state space composed of real-time errors of an inertial navigation system and real-time errors of a starlight guidance system to obtain a compensation value of a mathematical platform misalignment angle, correcting the mathematical platform misalignment angle using the compensation value, and correcting the flight attitude of the spacecraft using the corrected mathematical platform misalignment angle. The real-time state space directly uses real error data as input and does not need to rely on a preset fixed model, so when the noise dynamically changes or the model error, the system state can be captured through real-time error feedback, and filter divergence is reduced. In a high dynamic scene, the error caused by linearization can be reduced. Further, the compensation value corresponds to the real-time state space, and the compensation value is dynamically changing, so that the error accumulation chain can be reduced, and the error can always be within a controllable range.

Description

Technical Field

[0001] This application relates to the field of spacecraft technology, and in particular to a method, apparatus and electronic equipment for correcting the flight attitude angle of a spacecraft. Background Technology

[0002] The flight attitude angles of a spacecraft refer to the pitch, roll, and yaw angles of the spacecraft in the inertial navigation system. These angles need to be corrected in real time to ensure guidance accuracy, flight stability, and mission adaptability in complex flight scenarios.

[0003] Currently, starlight guidance is often used, employing attitude correction methods such as Kalman filtering to adjust the spacecraft's flight attitude angles. Starlight guidance uses a star as an inertial reference, measuring the starlight vector through star sensors to correct the spacecraft's flight attitude angles. However, current starlight guidance technology has the following problems:

[0004] First, it relies on guidance system models. Current starlight guidance systems depend on these models, such as the offset formulas of gyroscopes in inertial navigation systems. These models are fixed and cannot be adjusted autonomously, making them prone to filtering divergence when noise changes dynamically or model errors occur. Furthermore, in highly dynamic scenarios, such as rapid and violent spacecraft motion, the spacecraft's state changes exceed the applicability of linear approximations, resulting in significant linearization errors.

[0005] Secondly, the fixed correction strategy and inability to dynamically adjust correction parameters, coupled with the dynamic nature of spacecraft flight scenarios, make the fixed scenario unsuitable for the spacecraft's flight attitude, leading to susceptibility to vibrations and slow convergence. The intermittent nature of starlight guidance corrections results in increasingly larger accumulated errors. For example, if starlight guidance corrects every 0.5 seconds, within that 0.5 seconds, inertial navigation system misalignment causes an attitude angle deviation of 0.3°, and airflow interference causes another 0.2° deviation. This results in an accumulated error of 0.5° between two corrections, significantly increasing the error to be offset in the next correction. Over the long term, this error accumulation will continue to grow.

[0006] Third, there is a contradiction between robustness and real-time performance. Improving robustness requires sacrificing real-time performance, and vice versa. Specifically, improving robustness involves complex robustness algorithms, such as increasing the dimensionality of the error model, but this leads to an exponential increase in computation. Conversely, improving real-time performance simplifies robustness algorithms, such as ignoring some interference factors, which results in decreased robustness.

[0007] In view of this, a new method for correcting the attitude angle of an aircraft is provided to solve one or more of the aforementioned technical problems. Summary of the Invention

[0008] This application provides a method, apparatus, and electronic device for correcting the flight attitude angle of a spacecraft, which can achieve independent guidance system model, non-fixed compensation value, and simultaneously take into account robustness and real-time performance.

[0009] In a first aspect, embodiments of this application provide a method for correcting the flight attitude angle of a spacecraft, characterized in that the method includes:

[0010] The real-time state space is obtained, which indicates the real-time error of the composite guidance system composed of the inertial navigation system and the starlight guidance system.

[0011] The real-time state space is processed using a target correction model to obtain the compensation value of the mathematical platform misalignment angle; the mathematical platform misalignment angle is the misalignment angle obtained through the inertial navigation system.

[0012] The target correction model includes a decision network and a value network. The decision network is used to process the real-time state space to obtain the compensation value. The value network is used to determine the state value of the state space to be trained during the model training phase, and adjust the model parameters of the decision network according to the state value.

[0013] Adjust the misalignment angle of the mathematical platform according to the compensation value;

[0014] The spacecraft's flight attitude angles are corrected based on the adjusted mathematical platform misalignment angles.

[0015] Optionally, the real-time state space includes at least one of the following: real-time attitude angle error, mathematical platform misalignment angle error, velocity error, position error, and star tracking measurement error.

[0016] Optionally, the real-time state space includes real-time attitude angle error, mathematical platform misalignment angle error, velocity error, position error, and star tracking measurement error. Obtaining the real-time state space includes:

[0017] Acquire celestial information and inertial navigation information; the celestial information includes the spacecraft's attitude angle reference value obtained through the celestial guidance system; the inertial navigation information includes the spacecraft's attitude angle measurement value, velocity, position, gyroscope constant drift, and accelerometer constant bias obtained through the inertial navigation system;

[0018] Based on the attitude angle reference value and the attitude angle measurement value, determine the attitude angle error; based on the attitude angle error and the error transformation matrix, determine the mathematical platform misalignment angle;

[0019] Based on the mathematical platform's misalignment angle, velocity, position, gyroscope constant drift, and accelerometer constant bias, the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error are obtained.

[0020] The real-time state space is obtained based on the acceleration error, the velocity error, the position error, the mathematical platform misalignment angle error, and the star tracking measurement error.

[0021] Optionally, the step of obtaining acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error based on the mathematical platform misalignment angle, velocity, position, gyroscope constant drift, and accelerometer constant bias includes:

[0022] The acceleration error is obtained according to formula (1):

[0023] (1);

[0024] in, The inaccuracy angle of the mathematical platform, The acceleration error is... This is the constant bias of the accelerometer; It is Gaussian white noise. This is the transformation matrix from the spacecraft coordinate system to the navigation coordinate system;

[0025] The velocity error and the position error are obtained according to formula (2):

[0026] (2);

[0027] in, Speed ​​error,

[0028] , ;

[0029] ; ... These are the partial derivatives of gravitational acceleration with respect to position coordinates;

[0030] The misalignment angle error of the mathematical platform is obtained according to formula (3):

[0031] (3);

[0032] The misalignment angle error of the mathematical platform is... This represents the constant drift of the gyroscope. Gaussian white noise for the gyroscope model;

[0033] The celestial tracking measurement error is obtained according to formula (4):

[0034] (4);

[0035] in, For star tracking measurement error, The inverse autocorrelation time diagonal matrix of the Markov process in the star tracking measurement error is given. This is the star image drift error angle. .

[0036] Optionally, obtaining the real-time state space based on the acceleration error, the velocity error, the position error, the mathematical platform misalignment angle error, and the star tracking measurement error includes:

[0037] Based on the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and celestial tracking measurement error, combined with the pre-constructed state equation (5), the real-time state space is obtained:

[0038] (5);

[0039] in, The real-time state space is defined by the following components: the mathematical platform misalignment angle error, the acceleration error, the velocity error, the position error, and the star tracking measurement error. For random noise in formulas (1) to (4);

[0040] ;

[0041] .

[0042] Optionally, the target correction model is a trained correction model, and the target correction model uses a reward function as an iteration criterion during training; the reward function includes an error penalty term and an action penalty term;

[0043] The error penalty term is related to the attitude angle error and the system state error, and the system state error includes the velocity error, position error and mathematical platform misalignment angle error;

[0044] The action penalty is related to the compensation value.

[0045] Optionally, the error penalty term is as shown in formula (6):

[0046] (6);

[0047] in, This is the error penalty value. For attitude angle error, System state error For attitude error weights, For system state error weights, It is an L2 norm;

[0048] The action penalty is shown in formula (7):

[0049] (7);

[0050] in, The penalty value for the action. For action penalty weights, This is the compensation value obtained during the training process.

[0051] Optionally, the error in the real-time state space is a standardized error, and the method further includes:

[0052] According to formula (8), the target component in the target error in the real-time state space is standardized to obtain the standardized error. The target error is any error in the real-time state space, and the target component is any component of the target error.

[0053] (8)

[0054] in, For the standardized target components , Let be the theoretical minimum value of the target component s. This represents the theoretical maximum value of the target component s.

[0055] Secondly, embodiments of this application provide a spacecraft flight attitude angle correction device, the device comprising:

[0056] The acquisition unit is used to acquire the real-time state space, which indicates the real-time error of the composite guidance system composed of the inertial navigation system and the starlight guidance system.

[0057] The model processing unit is used to process the real-time state space using the target correction model to obtain the compensation value of the mathematical platform misalignment angle; the mathematical platform misalignment angle is the misalignment angle obtained through the inertial navigation system;

[0058] The target correction model includes a decision network and a value network. The decision network is used to process the real-time state space to obtain the compensation value. The value network is used to determine the state value of the state space to be trained during the model training phase, and adjust the model parameters of the decision network according to the state value.

[0059] An adjustment unit is used to adjust the misalignment angle of the mathematical platform according to the compensation value;

[0060] The correction unit is used to correct the flight attitude angle of the spacecraft based on the adjusted mathematical platform misalignment angle.

[0061] Thirdly, embodiments of this application provide an electronic device, including:

[0062] Memory, used to store computer programs;

[0063] A processor for executing the computer program to implement the method as described in any one of the first aspects.

[0064] Fourthly, embodiments of this application provide a computer program that, when run on a computer, causes the computer to perform the method in any of the possible implementations of any of the above aspects.

[0065] Fifthly, embodiments of this application provide a computer-readable storage medium storing a computer program (also referred to as code or instructions) that, when run on a computer, causes the computer to perform the method in any of the possible implementations of any of the above aspects.

[0066] Sixthly, embodiments of this application provide a chip system including one or more processors for calling and executing instructions stored in memory, causing the methods in any of the above aspects or possible implementations to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0067] This application provides a method, apparatus, and electronic device for correcting the flight attitude angle of a spacecraft. The method includes: acquiring a real-time state space, the real-time state space indicating the real-time error of a composite guidance system, the composite guidance system including an inertial navigation system and a starlight guidance system; processing the real-time state space using a target correction model to obtain a compensation value for the mathematical platform misalignment angle of the inertial navigation system; the target correction model including a decision network and a value network, the decision network being used to process the real-time state space to obtain the compensation value; the value network being used during model training to determine the state value of the state space to be trained, and adjusting the model parameters of the decision network according to the state value; adjusting the real-time mathematical platform misalignment angle according to the compensation value; the real-time mathematical platform misalignment angle being the real-time misalignment angle of the inertial navigation system; and correcting the flight attitude angle of the spacecraft according to the adjusted mathematical platform misalignment angle.

[0068] To determine the compensation value for the misalignment angle of the mathematical platform, this embodiment utilizes a target correction model to process the real-time state space composed of the real-time errors of the inertial navigation system and the starlight guidance system. The real-time state space directly uses real error data as input and does not rely on a preset fixed model. Therefore, when noise dynamically changes or model errors occur, the system state can be captured through real-time error feedback, reducing filter divergence. In highly dynamic scenarios, errors caused by linearization can be reduced. Furthermore, the compensation value corresponds to the real-time state space, and this compensation value is dynamically changing. Therefore, by reducing the error accumulation chain, the error can always remain within a controllable range. Moreover, the target correction model in this embodiment is a dual-network collaborative optimization method, which can ensure correction accuracy without sacrificing real-time performance. Attached Figure Description

[0069] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0070] Figure 1 A flowchart illustrating a method for correcting the flight attitude angle of a spacecraft, provided as an embodiment of this application;

[0071] Figure 2 This application provides a schematic diagram of a spacecraft attitude angle correction device. Detailed Implementation

[0072] To enable those skilled in the art to better understand the present application, the technical solutions in this embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0073] In the implementation of this application, the collection and processing of relevant data should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and carry out subsequent data use and processing within the scope of laws and regulations and the authorization of the personal information subject.

[0074] First, the technical terms involved in the embodiments of this application will be explained.

[0075] (1) The flight attitude angle of the spacecraft.

[0076] In the embodiments of this application, the flight attitude angles of a spacecraft refer to the pitch angle, roll angle, and heading angle of the spacecraft in the inertial navigation system.

[0077] In star-guided navigation, parallel light from a star is focused and imaged by a charge-coupled device (CCD) array after passing through an optical system. The center of the star image is determined using the energy center method. Based on the focusing geometry, the direction of the starlight vector in the star sensor coordinate system is further calculated. Then, through a transformation using the star sensor mounting matrix, the observed vector of the starlight vector in the spacecraft coordinate system is obtained, thus acquiring the spacecraft's three-axis attitude information: pitch angle θ, yaw angle ψ, and roll angle γ. This attitude information is also called the attitude angle reference value. The three-axis attitude information of the spacecraft determined by the inertial navigation system, also known as the attitude angle measurements, are pitch angle θ0, yaw angle ψ0, and roll angle γ0. The spacecraft coordinate system is a right-handed orthogonal coordinate system with the spacecraft's center of mass as the origin, the x-axis pointing forward along the spacecraft's longitudinal axis, the y-axis pointing horizontally to the right, and the z-axis pointing vertically upward.

[0078] In this embodiment, the attitude angle error is the difference between the attitude angle reference value and the attitude angle measurement value. That is, the attitude angle error of the spacecraft in the three axes of the missile coordinate system is as shown in formula (1):

[0079] (1)

[0080] This refers to the attitude angle error. If the attitude angle error is determined, the aircraft's attitude angle under the inertial navigation system can be determined based on the attitude angle reference value determined by the star sensor, as shown below: , , .in, The pitch angle is measured using an inertial navigation system. For the heading angle measured under the inertial navigation system, This is the roll angle measured using an inertial navigation system.

[0081] Since the attitude angle error is error information under the starlight guidance system, in this embodiment, the attitude angle error needs to be converted into the mathematical platform misalignment angle under the inertial navigation system before calibration can be performed. The specific conversion is as follows:

[0082] ;

[0083] In the formula, This is the error angle transformation matrix. For the inaccurate angle of the mathematical platform, This refers to the attitude angle error. Specifically, the mathematical platform misalignment angle refers to the angular deviation between the inertial navigation system and the actual navigation coordinate system.

[0084] (2) Starlight guidance

[0085] Starlight guidance is a technology that uses stars as inertial references and measures the starlight vectors of stars through star sensors to correct the flight attitude and position deviations of spacecraft.

[0086] Currently, starlight guidance employs attitude correction algorithms such as Kalman filtering or PID control to correct carrier attitude and position deviations. However, current technology has the following problems:

[0087] First, it relies on guidance system models. Current starlight guidance systems depend on these models, such as the offset formulas of gyroscopes in inertial navigation systems. These models are fixed and cannot be adjusted autonomously, making them prone to filtering divergence when noise changes dynamically or model errors occur. Furthermore, in highly dynamic scenarios, such as rapid and violent spacecraft motion, the spacecraft's state changes exceed the applicability of linear approximations, resulting in significant linearization errors.

[0088] Secondly, the fixed correction strategy and inability to dynamically adjust correction parameters, coupled with the dynamic nature of spacecraft flight scenarios, make the fixed scenario unsuitable for the spacecraft's flight attitude, leading to susceptibility to vibrations and slow convergence. The intermittent nature of starlight guidance corrections results in increasingly larger accumulated errors. For example, if starlight guidance corrects every 0.5 seconds, within that 0.5 seconds, inertial navigation system misalignment causes an attitude angle deviation of 0.3°, and airflow interference causes another 0.2° deviation. This results in an accumulated error of 0.5° between two corrections, significantly increasing the error to be offset in the next correction. Over the long term, this error accumulation will continue to grow.

[0089] Third, there is a contradiction between robustness and real-time performance. Improving robustness requires sacrificing real-time performance, and vice versa. Specifically, improving robustness involves complex robustness algorithms, such as increasing the dimensionality of the error model, but this leads to an exponential increase in computation. Conversely, improving real-time performance simplifies robustness algorithms, such as ignoring some interference factors, which results in decreased robustness.

[0090] To address one or more of the aforementioned technical problems, embodiments of this application provide a method for correcting the flight attitude angle of a spacecraft, which can be applied to a correction system. The correction system performs the following operations: acquiring a real-time state space, the real-time state space indicating the real-time error of a composite guidance system, the composite guidance system including an inertial navigation system and a starlight guidance system; processing the real-time state space using a target correction model to obtain a compensation value for the mathematical platform misalignment angle of the inertial navigation system; the target correction model includes a decision network and a value network, the decision network being used to process the real-time state space to obtain the compensation value; the value network being used during the model training phase to determine the state value of the state space to be trained, and adjusting the model parameters of the decision network based on the state value; adjusting the real-time mathematical platform misalignment angle based on the compensation value; the real-time mathematical platform misalignment angle being the real-time misalignment angle of the inertial navigation system; and correcting the flight attitude angle of the spacecraft based on the adjusted mathematical platform misalignment angle.

[0091] Compared to attitude correction algorithms such as Kalman filtering or PID control, this method determines the compensation value for the mathematical platform's misalignment angle. It then uses a target correction model to process the real-time state space, composed of the real-time errors of the inertial navigation system and the starlight guidance system. The real-time state space directly uses real error data as input, without relying on a pre-set fixed model. Therefore, when noise dynamically changes or model errors occur, the system state can be captured through real-time error feedback, reducing filter divergence. In highly dynamic scenarios, it can reduce errors caused by linearization. Furthermore, the compensation value corresponds to the real-time state space, and this compensation value is dynamically changing. Therefore, by reducing the error accumulation chain, the error can always be kept within a controllable range. Moreover, the target correction model in this embodiment is a dual-network collaborative optimization method, which can ensure correction accuracy without sacrificing real-time performance.

[0092] In practical applications, the correction system may include a software system, which can be provided to the user as a software package for deployment, such as on a local physical server or in a private cloud. In some possible implementations, the correction system may also be deployed in a public cloud and provided to the user as a cloud service. For example, a cloud service provider could offer a one-stop system service integrating the functions of the aforementioned core components.

[0093] The following description, in conjunction with the accompanying drawings, illustrates a method for correcting the flight attitude angle of a spacecraft according to an embodiment of this application. It should be noted that the method is illustrated using a correction system as an example.

[0094] Appendix Figure 1 A flowchart of a method for correcting the flight attitude angle of a spacecraft is provided for an embodiment of this application. The method includes steps S10 to S40.

[0095] S10. Obtain the real-time state space.

[0096] The state space is used to reflect the errors of the composite guidance system, which includes an inertial navigation system and a stellar guidance system. The state space is a state vector, including, but not limited to, object tracking measurement errors, velocity errors, position errors, mathematical platform misalignment angle errors, and attitude angle errors. In this embodiment, the mathematical platform misalignment angle error is determined by random noise.

[0097] In this embodiment, the state space also satisfies real-time performance and dimensional controllability to avoid redundant information causing slow convergence of the target correction model. Specifically, in this embodiment, the state space is a real-time state space used to indicate the real-time error of the composite guidance system. It should be noted that the acquisition cycle of the real-time state space is synchronized with the data update cycles of the inertial navigation system and the starlight guidance system.

[0098] Furthermore, due to the different dimensions of each error component, numerical differences exist during training, leading to an imbalance in network weights. In this embodiment, the real-time state space is a standardized state vector. For example, for any error component in the real-time state space, such as the target component of the target error... The target error is calculated using the following formula. Standardization process:

[0099] ;

[0100] in, For the standardized target components, This represents the theoretical minimum value of the target component. This represents the theoretical maximum value of the target component. For example, if the target error is a velocity error, and the target component is the velocity error in the x-axis direction, s = 10 m / s, the theoretical minimum value of this velocity error is... -20 m / s, theoretical maximum If the speed is 20 m / s, then the components included in the state space are... =0.75.

[0101] In this embodiment, a real-time state space can be obtained by unifying acceleration error, velocity error, position error, mathematical platform misalignment angle error, attitude angle error, and / or celestial tracking measurement errors through state equations. In this embodiment, the state equations include, but are not limited to, one or more of the following: acceleration error equation, velocity-position error equation, position error equation, mathematical platform misalignment angle error equation, and celestial tracking measurement error equation.

[0102] S20. The real-time state space is processed using the target correction model to obtain the compensation value of the mathematical platform misalignment angle of the inertial navigation system.

[0103] The target correction model includes a decision network and a value network. The decision network is used to process the real-time state space to obtain compensation values. The value network is used to determine the state value of the state space to be trained during the model training phase, and adjust the model parameters of the decision network according to the state value.

[0104] The decision network comprises an input layer, hidden layers, and an output layer. The input layer acquires the real-time state space, transforms it into a form recognizable by the neural network, and passes it to the hidden layers. The hidden layers extract high-dimensional features from the input real-time state space to capture the coupling relationships between various errors, such as the correlation between the mathematical platform misalignment angle and attitude angle errors, and the propagation relationship between velocity and position errors. The hidden layers send the extracted high-dimensional features to the output layer, which outputs the compensation value (also known as the action) for the mathematical platform misalignment angle.

[0105] The compensation value of the mathematical platform misalignment angle output by the decision network :

[0106] ;

[0107] in, For compensation value The component along the x-axis in the spacecraft coordinate system, For compensation value The component along the x-axis in the spacecraft coordinate system, For compensation value The component in the z-axis direction of the spacecraft coordinate system.

[0108] It should be noted that in this embodiment, the motion space is directly matched with the attitude update mechanism of the inertial navigation system. The output of this motion space can directly affect the correction amount of the inertial navigation system, and the motion amplitude needs to be limited to avoid vibration of the inertial navigation system.

[0109] In one example, the compensation values ​​output by this embodiment are continuous values, which constitute the action space. The output layer of the decision network includes a variance sub-output layer and a mean sub-output layer, the variance sub-output layer being used to output the action mean. The mean sub-output layer is used to output the action variance. Then through Sampling yields compensation value Where I is the identity matrix. This method allows for precise fine-tuning and avoids abrupt changes in amplitude.

[0110] It should be noted that starlight guidance systems are subject to random interference such as star tracking measurement noise and gyroscope Gaussian white noise. Traditional deterministic correction methods are easily affected by this interference, leading to correction oscillations. By employing Gaussian distribution sampling, even if the input state fluctuates slightly due to noise, the sampling action will not exhibit extreme abrupt changes, thus reducing correction oscillations. Action variance It can dynamically respond to interference intensity, further enhancing robustness against random interference and improving anti-interference capabilities.

[0111] For example, Table 1 shows the architecture of a decision network provided in an embodiment of this application. In this architecture, the decision network includes an input layer, two hidden layers (hidden layer 1 and hidden layer 2), and two output layers (mean output layer and variance output layer).

[0112] Table 1

[0113]

[0114] Furthermore, in this embodiment of the application, the decision network can employ the following decision loss function during training:

[0115] ;

[0116] in, This represents the probability ratio between the old and new strategies; The dominant function reflects the action. The advantage of relative average; Here, W is the clipping parameter; W is the weight matrix; and b is the bias parameter. For the decision loss value, These are the model parameters for the decision network. ) is the clipping function.

[0117] In this embodiment, the value network includes an input layer, a hidden layer, and an output layer. During training, the input layer acquires the sample state space, converts it into a form recognizable by the neural network, and passes it to the hidden layer. The hidden layer extracts high-dimensional features from the input sample state space to capture the coupling relationships between various errors, such as the correlation between mathematical platform misalignment angle and attitude angle errors, and the propagation relationship between velocity and position errors. The hidden layer sends the extracted high-dimensional features to the output layer, which outputs the state value corresponding to the state space.

[0118] For example, Table 2 is an architecture table of a value network provided in an embodiment of this application. As shown in Table 2, the hidden layers of this value network include hidden layer 1 and hidden layer 2.

[0119] Table 2

[0120]

[0121] In this embodiment, the loss function used by the value network during training is:

[0122] ;

[0123] in, For target value, Value loss These are the model parameters for the value network. Value of the current state. This represents the mathematical expectation under strategy π.

[0124] Furthermore, embodiments of this application also provide a method for training a target correction model, the training method comprising the following steps:

[0125] A1. Initialization configuration.

[0126] In this embodiment of the application, the network parameters of the decision network are initialized. Network parameters of value network The sampling period T is set. The sampling period T is the same as the data update period of the starlight guidance system.

[0127] A2. Training samples are collected based on the sampling period T.

[0128] The training samples include historical system states, which include, but are not limited to, information such as historical velocity errors, historical position errors, historical mathematical platform misalignment angles, and historical attitude angle errors.

[0129] Specifically, read the current system state. ,Will After standardization, the inputs are fed into the decision network and the value network. The decision network outputs the average action value. and action variance Sampling yields correction actions :

[0130]

[0131] Will Inputting data into the composite guidance system updates the spacecraft's flight attitude parameters, and the system transitions to the new state. And calculate the instant reward based on the preset reward function. Storage Experience Group The sampling continues until the number of sampling steps reaches a preset length N, where N is a positive integer, for example, N is 1000.

[0132] Furthermore, generalized advantage estimation (GAE-) can be used. Calculate the advantage function .

[0133] It should be noted that, to improve training performance, the training samples also include random noise to enhance the algorithm's robustness to noise and airflow interference from the star tracker measurements. Robustness is also added to the reward function to avoid drastic fluctuations.

[0134] In this embodiment, the reward function includes an error penalty term and an action penalty term. The error penalty term is related to the calibrated attitude angle error and system state error; the action penalty term is related to the corrected action.

[0135] For example, the error penalty term is as follows:

[0136] ;

[0137] in, This is the error penalty value. The attitude angle error is... The system state error includes velocity error. Position error The comprehensive error vector of the mathematical platform's misalignment angle error For attitude error weights, For system state error weights, It is an L2 norm.

[0138] The penalty for the action is shown in the following formula:

[0139] ;

[0140] in, The penalty value for the action. For action penalty weights, The reward function both penalizes pose errors and corrects excessive actions, and provides positive incentives for high-precision corrections, guiding the algorithm towards directional optimization.

[0141] A3. Use training samples to train the target and correct the model.

[0142] Based on N training samples from the experience pool and the calculated advantage function Update the network parameters of the decision network and the value network.

[0143] S30. Adjust the mathematical platform misalignment angle according to the compensation value.

[0144] That is, the target correction model can be achieved through output Real-time compensation is performed on the mathematical platform misalignment angle of the inertial navigation system, and the corrected platform misalignment angle is obtained. for:

[0145]

[0146] S40. Correct the spacecraft's flight attitude angle based on the adjusted mathematical platform misalignment angle.

[0147] To determine the compensation value for the misalignment angle of the mathematical platform, this embodiment utilizes a target correction model to process the real-time state space composed of the real-time errors of the inertial navigation system and the starlight guidance system. The real-time state space directly uses real error data as input and does not rely on a preset fixed model. Therefore, when noise dynamically changes or model errors occur, the system state can be captured through real-time error feedback, reducing filter divergence. In highly dynamic scenarios, errors caused by linearization can be reduced. Furthermore, the compensation value corresponds to the real-time state space, and this compensation value is dynamically changing. Therefore, by reducing the error accumulation chain, the error can always remain within a controllable range. Moreover, the target correction model in this embodiment is a dual-network collaborative optimization method, which can ensure correction accuracy without sacrificing real-time performance.

[0148] Furthermore, in relation to S10, this application embodiment also provides a state equation, which is as follows:

[0149] ;

[0150] in, The components are the mathematical platform misalignment angle, the velocity error, the position error, and the star tracking measurement error, respectively. For each component, there is random noise;

[0151] ;

[0152]

[0153] in, , ... These are the partial derivatives of gravitational acceleration with respect to position coordinates, respectively. It is Gaussian white noise, including components along the x-axis in the space coordinate system. Components in the y-axis direction and the component in the z-axis direction , This is the transformation matrix from the spacecraft coordinate system to the navigation coordinate system. The constant drift of the gyroscope includes the component in the x-axis direction of the space coordinate system. Components in the y-axis direction and the component in the z-axis direction. This is the time diagonal matrix of the Markov process inverse autocorrelation in the star tracking measurement error.

[0154] The embodiments of this application can integrate the acceleration error equation, the velocity position error equation, the mathematical platform misalignment angle error equation, and the star tracking measurement error equation to obtain the state equation.

[0155] The acceleration error equation is as follows:

[0156] ;

[0157] In the formula, For accelerometer measurement error; W is the antisymmetric matrix composed of the specific forces sensitive to the accelerometer. The inaccuracy angle of the mathematical platform; This is the constant bias of the accelerometer; It is Gaussian white noise; This is the transformation matrix from the spacecraft coordinate system to the navigation coordinate system.

[0158] The acceleration error equation is composed of accelerometer error and mathematical platform offset error, which form the accelerometer measurement error. This error ignores the first and second order error terms of the accelerometer and installation error, so as to reduce the amount of calculation while ensuring the accuracy of the error.

[0159] The velocity-position error equation is as follows:

[0160] ;

[0161] In the formula, This refers to the velocity-position error;

[0162] , ;

[0163]

[0164] These represent the velocity and position errors along different axes in the space coordinate system; coefficients ... Let be the partial derivative of gravitational acceleration with respect to position coordinates.

[0165] In the inertial coordinate system at the launch point, this error equation simplifies the gravitational field model into a position error equation for the ballistic velocity constructed under a central force field at the center of the sphere.

[0166] The mathematical platform's misalignment angle error equation is as follows:

[0167] ;

[0168] In the formula, It is a diagonal matrix composed of constant drift values ​​of the gyroscopes; It is Gaussian white noise in the gyroscope model. 'This refers to the inaccuracy angle error of the mathematical platform.'

[0169] This equation is obtained by neglecting the errors of the first and second terms of the gyroscope.

[0170] The equation for the celestial body tracking measurement error is as follows:

[0171] ;

[0172] In the formula, The inverse autocorrelation time diagonal matrix of the Markov process in the star tracking measurement error is given. This is the star image drift error angle. , This refers to the measurement error in celestial tracking.

[0173] In this error equation, the star tracking measurement error consists of a first-order Markov process and white noise. The error equation is obtained by describing the slowly changing star tracking through the Markov process.

[0174] In one specific implementation, the correction system can obtain the state space through the following steps:

[0175] B1. Acquire celestial information and inertial navigation information; celestial information includes the spacecraft's attitude angle reference value; inertial navigation information includes the spacecraft's attitude angle measurement value, velocity, position, gyroscope constant drift, and accelerometer constant bias obtained through the inertial navigation system.

[0176] B2. Determine the attitude angle error based on the attitude angle reference value and the attitude angle measurement value; determine the mathematical platform misalignment angle based on the attitude angle error and the error transformation matrix;

[0177] B3. Based on the mathematical platform misalignment angle, velocity, position, gyroscope constant drift, and accelerometer constant bias, and combined with the above dispersed error equations, the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error are obtained respectively.

[0178] Among them, the dispersion error equations are the acceleration error equation, velocity position error equation, mathematical platform misalignment angle error equation, and star tracking measurement error equation mentioned above.

[0179] B4. Based on the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error, and combined with the state equation, obtain the real-time state space.

[0180] This real-time state space comprehensively reflects the dynamic changes of various error factors during spacecraft flight, laying a solid data foundation for subsequent precise correction of spacecraft attitude angles. Through in-depth analysis and processing of the real-time state space, the correction system can more accurately capture key error information affecting attitude angles, thereby achieving dynamic, real-time, and high-precision correction of spacecraft attitude angles.

[0181] This application provides a method for correcting the flight attitude angle of a spacecraft. To determine the compensation value for the misalignment angle of a mathematical platform, this application utilizes a target correction model to process the real-time state space composed of the real-time errors of the inertial navigation system and the starlight guidance system. The real-time state space directly uses real error data as input and does not rely on a preset fixed model. Therefore, when noise dynamically changes or model errors occur, the system state can be captured through real-time error feedback, reducing filter divergence. In highly dynamic scenarios, it can reduce errors caused by linearization. Furthermore, the compensation value corresponds to the real-time state space, and this compensation value is dynamically changing. Therefore, by reducing the error accumulation chain, the error can always be kept within a controllable range. Moreover, the target correction model in this application is a dual-network collaborative optimization method, which can ensure correction accuracy without sacrificing real-time performance.

[0182] against Figure 1 The method for correcting the flight attitude angle of a spacecraft shown in the present application also provides a device for correcting the flight attitude angle of a spacecraft.

[0183] Appendix Figure 2 This application provides a schematic diagram of a spacecraft flight attitude angle correction device, the device 200 including:

[0184] Acquisition unit 201 is used to acquire the real-time state space, wherein the real-time state space indicates the real-time error of the composite guidance system composed of the inertial navigation system and the starlight guidance system;

[0185] The model processing unit 202 is used to process the real-time state space using the target correction model to obtain the compensation value of the mathematical platform misalignment angle; the mathematical platform misalignment angle is the misalignment angle obtained through the inertial navigation system;

[0186] The target correction model includes a decision network and a value network. The decision network is used to process the real-time state space to obtain the compensation value. The value network is used to determine the state value of the state space to be trained during the model training phase, and adjust the model parameters of the decision network according to the state value.

[0187] Adjustment unit 203 is used to adjust the mathematical platform misalignment angle according to the compensation value;

[0188] The correction unit 204 is used to correct the flight attitude angle of the spacecraft based on the adjusted mathematical platform misalignment angle.

[0189] Optionally, the state space includes at least one of attitude angle error, mathematical platform misalignment angle error, velocity error, position error, and star tracking measurement error.

[0190] Optionally, the state space includes attitude angle error, mathematical platform misalignment angle error, velocity error, position error, and star tracking measurement error, and the acquisition of the state space includes:

[0191] Acquire celestial information and inertial navigation information; the celestial information includes the spacecraft's attitude angle reference value; the inertial navigation information includes the spacecraft's attitude angle measurement value, velocity, position, gyroscope constant drift, and accelerometer constant bias obtained through the inertial navigation system;

[0192] Based on the attitude angle reference value and the attitude angle measurement value, the attitude angle error is determined; based on the attitude angle error and the error transformation matrix, the mathematical platform misalignment angle is determined.

[0193] Based on the mathematical platform misalignment angle, the velocity, the position, the gyroscope constant drift, and the accelerometer constant bias, the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error are obtained.

[0194] The state space is obtained based on the acceleration error, the velocity error, the position error, the mathematical platform misalignment angle error, and the star tracking measurement error.

[0195] Optionally, the step of obtaining acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error based on the mathematical platform misalignment angle, velocity, position, gyroscope constant drift, and accelerometer constant bias includes:

[0196] The acceleration error is obtained according to the following formula:

[0197] ;

[0198] in, The inaccuracy angle of the mathematical platform, The acceleration error is... This is the constant bias of the accelerometer; It is Gaussian white noise. is the transformation matrix from the spacecraft coordinate system to the navigation coordinate system; W is the antisymmetric matrix composed of the specific forces sensitive to the accelerometer.

[0199] The velocity error and the position error are obtained according to the following formulas:

[0200] ;

[0201] in, .

[0202] These are the velocity error components for the corresponding axial directions. These are the position error components for the corresponding axial directions;

[0203] , ;

[0204] ; ... These are the partial derivatives of gravitational acceleration with respect to position coordinates;

[0205] The misalignment angle error of the mathematical platform is obtained according to the following formula:

[0206]

[0207] The misalignment angle error of the mathematical platform is... The drift of the gyroscope constant in the x-axis, y-axis, and z-axis directions of the space coordinate system is given by the following values. Gaussian white noise for the gyroscope model;

[0208] The celestial tracking measurement error is obtained using the following formula:

[0209]

[0210] in, For star tracking measurement error, The inverse autocorrelation time diagonal matrix of the Markov process in the star tracking measurement error is given. This is the star image drift error angle. .

[0211] Optionally, obtaining the state space based on the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error includes:

[0212] Based on the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error, combined with the pre-constructed state equations (as follows), the state space is obtained:

[0213] d ;

[0214] in, The components are the mathematical platform misalignment angle error, acceleration error, velocity error, position error, and star tracking measurement error, respectively. This refers to the random noise in the above error equation;

[0215] ;

[0216] .

[0217] Optionally, the target correction model is a trained correction model, and the target correction model uses a reward function as an iteration criterion during training; the reward function includes an error penalty term and an action penalty term;

[0218] The error penalty term is related to the attitude angle error and the system state error, and the system state error includes the velocity error, position error and mathematical platform misalignment angle error;

[0219] The action penalty is related to the compensation value.

[0220] Optionally, the error penalty term is as shown in the formula:

[0221] ;

[0222] in, This is the error penalty value. The attitude angle error is... The system state error For attitude error weights, For system state error weights, It is an L2 norm;

[0223] The penalty for the action is shown in the formula:

[0224] ;

[0225] in, The penalty value for the action. For action penalty weights, This is the compensation value obtained during the training process.

[0226] Optionally, the error in the state space is a standardized error, which can be standardized in the following way:

[0227] The target component in the target error in the state space is standardized according to the following formula to obtain the standardized error, where the target error is any error in the state space, and the target component is any component of the target error:

[0228] ;

[0229] in, For the standardized target components, For the target component, This represents the theoretical minimum value of the target component. This is the theoretical maximum value of the target component.

[0230] This application provides a spacecraft attitude angle correction device. The device obtains the real-time state space by processing the real-time errors of the inertial navigation system and the starlight guidance system, based on a target correction model, after determining the compensation value for the misalignment angle of a mathematical platform. The real-time state space directly uses real error data as input and does not rely on a preset fixed model. Therefore, when noise changes dynamically or model errors occur, the system state can be captured through real-time error feedback, reducing filter divergence. In highly dynamic scenarios, it can reduce errors caused by linearization. Furthermore, the compensation value corresponds to the real-time state space and is dynamically changing. Therefore, by reducing the error accumulation chain, the error can always be kept within a controllable range. Moreover, the target correction model in this application is a dual-network collaborative optimization method, which can ensure correction accuracy without sacrificing real-time performance.

[0231] According to the method provided in the embodiments of this application, this application also provides a chip system, which includes one or more processors for calling and executing instructions stored in memory, thereby causing the method described in the embodiments of this application to be executed. The chip system may be composed of chips or may include chips and other discrete devices.

[0232] The chip system may include input circuits or interfaces for transmitting information or data, and output circuits or interfaces for receiving information or data.

[0233] According to the method provided in the embodiments of this application, this application also provides a computer program product, which includes: computer program code, which, when run on a computer, causes the computer to execute the various steps or processes executed by the network device or terminal device in any of the foregoing method embodiments.

[0234] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium storing program code, which, when run on a computer, causes the computer to execute the various steps or processes executed by the network device or terminal device in any of the foregoing method embodiments.

[0235] The computer-readable storage medium may be the aforementioned volatile memory or non-volatile memory, or it may include both volatile memory and non-volatile memory.

[0236] In the embodiments of this application, the terms and English abbreviations are exemplary examples given for ease of description and should not be construed as limiting the application in any way. This application does not preclude the possibility of defining other terms that can achieve the same or similar functions in existing or future agreements.

[0237] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When these computer instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated.

[0238] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

Claims

1. A method for correcting the flight attitude angle of a spacecraft, characterized in that, The method includes: The real-time state space is obtained, which indicates the real-time error of the composite guidance system composed of the inertial navigation system and the starlight guidance system. The real-time state space is processed using a target correction model to obtain the compensation value of the mathematical platform misalignment angle; the mathematical platform misalignment angle is the misalignment angle obtained through the inertial navigation system. The target correction model includes a decision network and a value network. The decision network is used to process the real-time state space to obtain the compensation value. The value network is used to determine the state value of the state space to be trained during the model training phase, and adjust the model parameters of the decision network according to the state value. Adjust the misalignment angle of the mathematical platform according to the compensation value; Based on the adjusted mathematical platform misalignment angle, correct the spacecraft's flight attitude angle; The real-time state space includes real-time attitude angle error, mathematical platform misalignment angle error, velocity error, position error, and star tracking measurement error. Acquiring the real-time state space includes: Acquire celestial information and inertial navigation information; the celestial information includes the spacecraft's attitude angle reference value obtained through the celestial guidance system; the inertial navigation information includes the spacecraft's attitude angle measurement value, velocity, position, gyroscope constant drift, and accelerometer constant bias obtained through the inertial navigation system; Based on the attitude angle reference value and the attitude angle measurement value, determine the attitude angle error; based on the attitude angle error and the error transformation matrix, determine the mathematical platform misalignment angle; Based on the mathematical platform's misalignment angle, velocity, position, gyroscope constant drift, and accelerometer constant bias, the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error are obtained. The real-time state space is obtained based on the acceleration error, the velocity error, the position error, the mathematical platform misalignment angle error, and the star tracking measurement error. The acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error are obtained based on the mathematical platform's misalignment angle, velocity, position, gyroscope constant drift, and accelerometer constant bias, including: The acceleration error is obtained according to formula (1): （1）； in, The inaccuracy angle of the mathematical platform, The acceleration error is... This is the constant bias of the accelerometer; It is Gaussian white noise. is the transformation matrix from the spacecraft coordinate system to the navigation coordinate system; W is the antisymmetric matrix composed of the specific forces sensitive to the accelerometer. The velocity error and the position error are obtained according to formula (2): （2）； in, Speed ​​error, ， ； ; ... These are the partial derivatives of gravitational acceleration with respect to position coordinates; The misalignment angle error of the mathematical platform is obtained according to formula (3): （3）； The misalignment angle error of the mathematical platform is... This represents the constant drift of the gyroscope. Gaussian white noise for the gyroscope model; The celestial tracking measurement error is obtained according to formula (4): （4）； in, For star tracking measurement error, The inverse autocorrelation time diagonal matrix of the Markov process in the star tracking measurement error is given. This is the star image drift error angle. .

2. The method according to claim 1, characterized in that, The step of obtaining the real-time state space based on the acceleration error, the velocity error, the position error, the mathematical platform misalignment angle error, and the star tracking measurement error includes: Based on the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and celestial tracking measurement error, combined with the pre-constructed state equation (5), the real-time state space is obtained: （5）； in, The real-time state space is defined by the following components: the mathematical platform misalignment angle error, the acceleration error, the velocity error, the position error, and the star tracking measurement error. For random noise in formulas (1) to (4); ； 。 3. The method according to claim 1, characterized in that, The target correction model is a trained correction model, and the target correction model uses a reward function as an iteration criterion during training; the reward function includes an error penalty term and an action penalty term; The error penalty term is related to the attitude angle error and the system state error, and the system state error includes the velocity error, position error and mathematical platform misalignment angle error; The action penalty is related to the compensation value.

4. The method according to claim 3, characterized in that, The error penalty term is shown in formula (6): （6）； in, This is the error penalty value. For attitude angle error, System state error For attitude error weights, For system state error weights, It is an L2 norm; The action penalty is shown in formula (7): （7）； in, The penalty value for the action. For action penalty weights, This is the compensation value obtained during the training process.

5. The method according to claim 1, characterized in that, The error in the real-time state space is a standardized error, and the method further includes: According to formula (8), the target component in the target error in the real-time state space is standardized to obtain the standardized error. The target error is any error in the real-time state space, and the target component is any component of the target error. （8） in, For the standardized target components , Let be the theoretical minimum value of the target component s. This represents the theoretical maximum value of the target component s.

6. A spacecraft flight attitude angle correction device, characterized in that, The device includes: An acquisition unit is used to acquire the real-time state space, which indicates the real-time error of the composite guidance system composed of the inertial navigation system and the starlight guidance system. The model processing unit is used to process the real-time state space using the target correction model to obtain the compensation value of the mathematical platform misalignment angle; the mathematical platform misalignment angle is the misalignment angle obtained through the inertial navigation system; The target correction model includes a decision network and a value network. The decision network is used to process the real-time state space to obtain the compensation value. The value network is used to determine the state value of the state space to be trained during the model training phase, and adjust the model parameters of the decision network according to the state value. An adjustment unit is used to adjust the misalignment angle of the mathematical platform according to the compensation value; The correction unit is used to correct the flight attitude angle of the spacecraft based on the adjusted mathematical platform misalignment angle; The real-time state space includes real-time attitude angle error, mathematical platform misalignment angle error, velocity error, position error, and star tracking measurement error. Acquiring the real-time state space includes: Acquire celestial information and inertial navigation information; the celestial information includes the spacecraft's attitude angle reference value obtained through the celestial guidance system; the inertial navigation information includes the spacecraft's attitude angle measurement value, velocity, position, gyroscope constant drift, and accelerometer constant bias obtained through the inertial navigation system; Based on the attitude angle reference value and the attitude angle measurement value, determine the attitude angle error; based on the attitude angle error and the error transformation matrix, determine the mathematical platform misalignment angle; Based on the mathematical platform's misalignment angle, velocity, position, gyroscope constant drift, and accelerometer constant bias, the acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error are obtained. The real-time state space is obtained based on the acceleration error, the velocity error, the position error, the mathematical platform misalignment angle error, and the star tracking measurement error. The acceleration error, velocity error, position error, mathematical platform misalignment angle error, and star tracking measurement error are obtained based on the mathematical platform's misalignment angle, velocity, position, gyroscope constant drift, and accelerometer constant bias, including: The acceleration error is obtained according to formula (1): （1）； in, The inaccuracy angle of the mathematical platform, The acceleration error is... This is the constant bias of the accelerometer; It is Gaussian white noise. is the transformation matrix from the spacecraft coordinate system to the navigation coordinate system; W is the antisymmetric matrix composed of the specific forces sensitive to the accelerometer. The velocity error and the position error are obtained according to formula (2): （2）； in, Speed ​​error, ， ； ; ... These are the partial derivatives of gravitational acceleration with respect to position coordinates; The misalignment angle error of the mathematical platform is obtained according to formula (3): （3）； The misalignment angle error of the mathematical platform is... This represents the constant drift of the gyroscope. Gaussian white noise for the gyroscope model; The celestial tracking measurement error is obtained according to formula (4): （4）； in, For star tracking measurement error, The inverse autocorrelation time diagonal matrix of the Markov process in the star tracking measurement error is given. This is the star image drift error angle. .

7. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the spacecraft flight attitude angle correction method according to any one of claims 1-5.

Images