A satellite collision avoidance method with full-process closed-loop rapid response

Abstract

The present invention discloses a full-process closed-loop rapid-response satellite collision avoidance method, comprising the following steps: Step 1: Observe and obtain the orbital state of space objects, then fuse multi-source data to construct a space object database; Step 2: Screen out high-risk collision sources that may collide with the satellite; Step 3: Determine the high-risk collision source based on the orbital state history records of the possible high-risk collision sources and the satellite's orbital state; Step 4: Generate an initial avoidance maneuver plan and optimize the initial avoidance maneuver plan using a multi-objective optimization algorithm; Determine the final avoidance maneuver plan for subsequent execution; Step 5: Based on the engine startup and shutdown time and thrust vector included in the final avoidance maneuver plan, control the onboard engine to operate at a specified thrust vector within the given startup and shutdown time to execute the collision avoidance maneuver. This method can achieve a faster response speed than conventional methods.

Description

Technical Field

[0001] The present invention relates to the field of near-Earth space technology, and in particular to a satellite collision avoidance method with a full-process closed-loop rapid response. Background Art

[0002] With the increasing frequency of global space activities, the number of spacecraft has surged, and the amount of space debris generated by launches, failures, and collisions has increased. The number and density of space objects are increasing day by day, leading to an increasing number of space collision incidents. To protect spacecraft safety, my country's space assets, and national security, it is of great significance to study collision avoidance methods for spacecraft such as satellites.

[0003] Satellite collision avoidance generally involves three main components: early warning, decision-making, and execution. Early warning involves monitoring orbiting spacecraft, space debris, and other objects through ground-based tracking and control stations and space-based satellites, forecasting their orbits, and determining whether there is a high risk of collision with a satellite. Decision-making involves formulating a collision avoidance strategy based on a comprehensive consideration of multiple factors, including the satellite's health, maneuverability, mission constraints, and fuel consumption, when a high risk of collision between a satellite and a space object is determined. Execution involves injecting orbit control maneuver commands into the satellite via the tracking and control station according to the formulated collision avoidance strategy, activating the onboard engine, and performing an orbit change maneuver to avoid a possible collision.

[0004] Conventional satellite collision avoidance methods primarily monitor space objects through ground-based tracking and control stations and space-based navigation satellites. This method typically provides long-range advance warning information, requiring collision warning results 48 hours before a collision. The scope of space objects monitored is limited, and measurement accuracy is limited. Conventional methods are ineffective for detecting smaller space debris and unexpected, approaching non-cooperative spacecraft, making it difficult to detect threats in advance and provide sufficient time for decision-making, planning, and execution of evasive maneuvers.

[0005] To provide collision warnings for a spacecraft, its orbit must be compared and calculated with the orbits of all other cataloged space objects. However, the sheer number of space objects necessitates a significant computational effort, necessitating efficient screening algorithms to filter out low-collision-risk objects and calculate only the orbits of high-risk objects. Conventional collision source screening employs a multi-stage process based on different orbital characteristics. This process is complex and requires manual intervention and judgment, resulting in low screening efficiency and impracticality for deployment onboard spacecraft with limited computing resources.

[0006] Developing a collision avoidance maneuver requires factoring in multiple factors, such as the satellite's health, maneuverability, mission constraints, and fuel consumption, as optimization constraints and objectives. This involves performing multi-objective optimization calculations and requiring repeated iterations to arrive at a satisfactory orbital control plan. Conventional iterative optimization methods are computationally intensive, inefficient, difficult to deploy onboard computers, and unable to meet the demands of rapid response.

[0007] Conventional methods often execute collision warnings and collision avoidance decisions on the ground. This requires transmitting instructions for evasive maneuvers from a ground-based tracking and control station to the satellite, which is then executed by the onboard engine. This process of transmitting instructions from the ground station to the satellite causes time delays, making it difficult to achieve instant control. Furthermore, ground-based tracking and control stations struggle to maintain full-time tracking and control of satellites. Instructions must be transmitted only when the satellite passes through areas covered by their own tracking and control stations. If the time between the detection of a high-risk collision source and the expected collision is short, and the satellite is out of range of its own tracking and control station at this time, the satellite will be unable to avoid the collision in a timely manner. Therefore, the conventional ground-based approach of executing warnings, making decisions, and transmitting instructions significantly limits the ability to rapidly respond to collision risks.

[0008] In addition, conventional collision avoidance methods usually consist of multiple independent links, and each link relies on manual operation, making it difficult to achieve an automated closed-loop process to avoid multiple potential collision targets, which also limits the ability to respond quickly to collision risks. Summary of the Invention

[0009] The purpose of the present invention is to propose a full-process closed-loop rapid response satellite collision avoidance method to address the shortcomings of the existing technology. It adopts an intelligent algorithm based on neural networks and can be deployed in onboard computers with limited computing resources. Compared with conventional solutions, it has higher computing efficiency and speed. Moreover, since it is deployed directly on the satellite, there is no need to communicate with the ground measurement and control station, and it can achieve a faster response speed than conventional methods.

[0010] In order to solve the above technical problems, the technical solution of the present invention is:

[0011] A satellite collision avoidance method with a full-process closed-loop rapid response includes the following steps:

[0012] Step 1: Observe and obtain the orbital state of space objects, then fuse multi-source data to build a space object database;

[0013] Step 2: Train an intelligent screening neural network model. Use the pre-trained intelligent screening neural network model to screen the orbital states of all space objects and identify high-risk collision sources that may collide with your satellite.

[0014] Step 3: Obtain a confirmed high-risk collision source based on the orbital status history records of the possible high-risk collision source and the own orbital status;

[0015] Step 4: Train an intelligent decision-making neural network model. Use the orbital state of the determined high-risk collision source and the orbital state of the own satellite as input to generate an initial evasive maneuver plan through the pre-trained intelligent decision-making neural network model. Optimize the initial evasive maneuver plan through a multi-objective optimization algorithm to obtain the final evasive maneuver plan for subsequent execution.

[0016] Step 5: Based on the engine on / off time and thrust vector included in the final avoidance maneuver plan, control the onboard engine to operate with the specified thrust vector within the given on / off time to perform the collision avoidance maneuver.

[0017] Step 6: After the maneuver is complete, the satellite status feedback is provided to determine the satellite's orbital state after the maneuver. The avoidance effect is evaluated to determine the collision risk after the avoidance. If the collision risk is still above the threshold, the process returns to Step 4 and repeats. If the collision risk is below the threshold, the process returns to Step 1 and repeats the warning and avoidance of potential high-risk collision sources.

[0018] Preferably, the orbital state refers to any physical quantity that can describe the position and velocity of the spacecraft in space.

[0019] Preferably, in step 1, the method for obtaining the orbital state of the space object includes: ground station observation, GNSS satellite observation, on-board autonomous observation and orbit prediction of known objects.

[0020] Preferably, the training method of the intelligent screening neural network model is:

[0021] First, the space object database is used as input. Geometric feature screening algorithms, temporal feature screening algorithms, and orbital rough prediction screening algorithms are applied to screen space objects, distinguish high-risk and low-risk collision sources, and label them, thereby establishing machine learning samples.

[0022] A basic neural network model is constructed, and then the supervised learning method is used to train the neural network model using the above machine learning sample set to optimize the network parameters. When the optimal network parameters are obtained, the training is completed and a pre-trained intelligent screening neural network model is obtained.

[0023] Preferably, in step 3, the method for obtaining the determined high-risk collision source is:

[0024] Step 3.1, based on the orbital state history of the collision source and the own satellite's orbital state, obtain the orbital state and measurement error covariance matrix at the moment of closest approach between the collision source and the own satellite;

[0025] Step 3.2: Based on the orbital state and measurement error covariance matrix of the collision source at the closest approach, and the orbital state and measurement error covariance matrix of the own satellite at the closest approach, the collision risk is assessed and the high-risk collision source is determined.

[0026] Preferably, in step 3.1, the measurement error covariance matrix at the initial moment is obtained based on the historical record of the orbital state of the collision source; then a high-precision dynamic model is used to make a high-precision prediction of the orbital state of the collision source and the own satellite at the initial moment and the measurement error covariance matrix at the initial moment, so as to obtain the orbital state and measurement error covariance matrix at the moment when the collision source and the own satellite are closest.

[0027] Preferably, in step 3.2, the collision risk assessment method is:

[0028] First, calculate the collision probability, the expression is as follows:

[0029] ;

[0030] in, and are the x and y components of the projection of the joint error covariance onto the meeting plane, Represents the distance between two targets when they are closest. and are the x and y components of the projection of the relative position vector on the encounter plane at the moment of closest approach, represents the natural exponential function;

[0031] Construct a comprehensive collision risk assessment function, the expression is as follows:

[0032] ;

[0033] Among them, a1 to a7 are weight coefficients, R, S, and W are the components of the relative position of the two targets at the time of closest approach in the orbital radial, track direction, and normal direction, and L R 、L S 、L W are the reference approach distances in the orbit radial, track, and normal directions, respectively; T1 and T2 are the time from the current moment to the TCA (i.e., the moment of closest approach) and the time from the most recent orbit measurement to the TCA, respectively; Q is the expert score for orbit determination quality taking into account the observation accuracy and orbit model accuracy; k and K are custom coefficients;

[0034] Calculate the collision risk assessment value of high-risk collision sources. When the collision risk of a space target is higher than the set threshold, it is considered that the satellite needs to avoid the target.

[0035] Preferably, the training method of the intelligent decision-making neural network model is as follows: the orbital state of the space object, the orbital state of the satellite, the constraints, and the collision risk in the historical satellite collision events are used as input features, and the avoidance maneuver strategy is used as the output feature, and a training sample set is constructed using a multi-objective optimization algorithm; a basic neural network model is established, and the neural network model is trained using the above training samples using a supervised learning method to obtain the optimal neural network parameters, and the pre-trained intelligent decision-making neural network model is obtained after the training is completed.

[0036] As a preference, in step 4, the method for optimizing the initial evasive maneuver scheme by a multi-objective optimization algorithm is as follows: the safety factor is the collision risk. F Indicates that the following conditions must be met:

[0037] ;

[0038] in, is the collision risk threshold;

[0039] The time factor refers to the time it takes for the engine to be turned on and off during the avoidance maneuver and the time it takes for the warning to be given. , the time between the two targets when they are closest , the start and end time of the ground-controllable time interval , the maximum start-up time of satellite orbit control engine The satellite orbit control engine power on / off time must meet the following conditions:

[0040] ;

[0041] The fuel factor refers to the fuel consumption of the engine during the evasive maneuver. During optimization, the fuel consumption should be reduced as much as possible. The expression of fuel consumption is as follows:

[0042]

[0043] in represents the satellite mass, Represents the engine thrust;

[0044] Mission factors refer to the satellite's own functions and the work efficiency of the tasks it performs. During optimization, the work efficiency should be improved as much as possible, that is, the impact of avoidance maneuvers on the satellite's own mission should be reduced.

[0045] The present invention has the following characteristics and beneficial effects:

[0046] Neural networks are used to implement intelligent screening algorithms and intelligent decision-making algorithms, replacing or assisting the hierarchical screening and multi-objective optimization algorithms in conventional methods. This can screen risk sources and make avoidance maneuver decisions with higher computing efficiency and faster computing speed, thereby improving the system's autonomous response capabilities.

[0047] After executing a collision avoidance maneuver, the avoidance effect is evaluated in real time and fed back to the decision-making process to determine whether to continue the avoidance maneuver and dynamically adjust the results. After completing the avoidance of the current target, the system returns to the early warning process to continue monitoring and avoiding the next potential collision target, achieving early warning and avoidance of multiple collision targets. Compared to the relatively discrete and independent collision avoidance steps in conventional methods, this solution proposes a full-process closed-loop approach to achieve automated and rapid response to collision risks.

[0048] Each step of the present invention can be fully or partially completed by the satellite's own observation equipment and onboard computer, reducing the communication limitations and time delays associated with data exchange with the ground. Conventional methods typically use ground-based equipment as the primary observation and computing platform, requiring satellite control commands to be injected into the satellite during its visible time interval for satellite maneuvers. This makes it difficult to address emergency collision risks and situations where ground-based measurement and control are unavailable.

[0049] Space objects are observed and monitored using a variety of methods, including onboard autonomous observation equipment, ground-based tracking and control stations, and space-based navigation satellites. Multi-source data fusion methods are used to improve observation accuracy.

[0050] This approach addresses the shortcomings of conventional collision source screening methods, which involve complex, multi-stage screening processes based on different orbital characteristics, require manual operation and judgment, have low screening efficiency, and cannot be deployed on onboard computers with limited computing resources. This problem is addressed by employing an intelligent screening algorithm based on a neural network model, which is trained in advance and then deployed onboard computers to assist in rapid collision source screening.

[0051] This approach addresses the shortcomings of conventional iterative optimization, which suffers from high computational complexity, low efficiency, difficulty deploying on-board computers, and an inability to meet rapid response requirements. This approach addresses these issues by employing an intelligent decision-making algorithm based on a neural network model, trained in advance and then deployed on-board computers. This algorithm facilitates the rapid generation of evasive maneuver solutions, reduces the computational complexity of subsequent optimization and correction algorithms, and thus solves these problems.

[0052] This approach addresses the shortcomings of conventional methods, where collision warning and collision avoidance decisions are often executed on the ground. This requires that instructions for executing evasive maneuvers be transmitted to the satellite via a ground-based tracking and control station, making it incapable of responding to emergencies and lacking support for ground-based tracking and control. This approach addresses the issue by implementing all or part of these steps through onboard equipment, enabling the satellite to autonomously perform the entire collision warning, decision-making, and execution process.

[0053] In conventional methods, the observation accuracy of ground tracking and control stations is limited. This scheme proposes to build a space object database by integrating multi-source data fusion methods of ground-based tracking and control stations, space-based navigation satellites, and autonomous observation equipment on board. By comprehensively considering observation data from multiple aspects, the observation accuracy of space objects is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0055] Figure 1 The present invention is a flowchart of an embodiment of a satellite collision avoidance method with a full-process closed-loop rapid response. DETAILED DESCRIPTION

[0056] It should be noted that, in the absence of conflict, the embodiments of the present invention and the features in the embodiments may be combined with each other.

[0057] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, features defined as "first", "second", etc. may explicitly or implicitly include one or more of the features. In the description of the present invention, unless otherwise specified, "multiple" means two or more.

[0058] In the description of the present invention, it should be noted that, unless otherwise expressly specified or limited, the terms "mounted," "connected," and "connected" should be understood in a broad sense. For example, they may refer to fixed connections, detachable connections, or integral connections; mechanical connections or electrical connections; direct connections or indirect connections through an intermediate medium; and internal communication between two components. Those skilled in the art will understand the specific meanings of the above terms in the present invention based on specific circumstances.

[0059] The present invention provides a full-process closed-loop rapid response satellite collision avoidance method, including a complete system for achieving satellite collision avoidance, wherein the specific technical details are not limited. Figure 1 As shown, it is mainly divided into three components, namely: early warning link, decision-making link and execution link.

[0060] The early warning phase uses ground-based tracking and control stations, space-based navigation satellites, and autonomous onboard observation equipment to monitor and locate space objects. Combined with known space object forecasts, a space object database is established through multi-source data fusion. An intelligent screening neural network model is used to screen high-risk collision sources. Upon detection of a high-risk collision source, this phase issues an alarm, followed by the collision avoidance decision phase. The decision phase collects orbital information, risk assessment data, and information about the collision risk source, along with information about the satellite itself. The intelligent decision neural network model generates an evasive maneuver plan, which is then refined by combining a precise multi-objective optimization algorithm and mission constraints. The execution phase executes the evasive maneuver plan generated by the decision phase through the onboard attitude and orbit control system. The satellite's orbital state and risk assessment information after the maneuver are fed back to the decision phase to evaluate the avoidance effect and determine whether to proceed with the next evasive maneuver, completing the closed-loop approach. Some steps of this approach utilize intelligent neural network-based algorithms, which can be deployed onboard computers with limited computing resources. This approach offers higher computational efficiency and speed than conventional approaches. Furthermore, since it is deployed directly onboard, eliminating the need for communication with ground tracking and control stations, it achieves a faster response than conventional approaches. The specific implementation steps are as follows.

[0061] Early warning link:

[0062] Step 1: Obtain the orbital states of all observable space objects through measurements from ground-based tracking and control stations and space-based navigation satellites. Autonomous onboard observations are performed using ranging and angle-measuring equipment onboard satellites to determine the relative motion of space objects relative to the satellites. Combined with the satellites' own orbital states, the orbital states of space objects are then determined. The orbital states of space objects obtained from all these observation methods are fused together to construct a space object database.

[0063] It should be noted that in this embodiment, the orbital state refers to any physical quantity that can describe the position and velocity of a spacecraft in space, such as Kepler orbital elements (hereinafter referred to as orbital elements), inertial space position and velocity coordinates, etc.

[0064] It should be noted that this embodiment uses a combination of space-based navigation satellites, ground-based measurement and control stations, and autonomous on-board observations to construct a space object database, with each component executed on different device carriers. In the case where the computing power of the on-board computer is limited, all observation information needs to be aggregated to the ground measurement and control station, and the subsequent steps of the early warning phase need to be executed on the ground computer. In the case where the computing power of the on-board computer is sufficient, all observation information can be aggregated to the on-board computer, and the subsequent steps of the early warning phase can be executed. In the implementation of this solution, a processor with sufficient computing power is preferably used as the on-board computer. In addition, the above-mentioned observation methods are all optional. In specific implementation or emergency situations, only some of them can be used for observation.

[0065] Step 2: After obtaining the space object database, use the pre-trained intelligent screening neural network model to screen the orbital states of all space objects in it and screen out high-risk collision sources that may collide with our satellites;

[0066] It should be noted that there are many methods for training this intelligent screening neural network model, which are conventional technical means. There are many types of basic neural network models that can achieve the functions required by this solution, including but not limited to fully connected neural networks, recurrent neural networks, and convolutional neural networks.

[0067] In this embodiment, in one possible implementation, feature screening is performed based on the orbital state of the space object and the orbital state of the own satellite, including geometric feature screening, temporal feature screening, and orbit rough prediction screening. The following briefly describes different screening methods.

[0068] Specifically, the geometric feature screening method is: let P o and A o are the perigee and apogee of a space target, respectively, P s and A s are the perigee and apogee of our satellite respectively. When the altitude difference between the orbit of the satellite and that of the satellite is large, the collision is considered to be unlikely.

[0069] Furthermore, based on the orbital elements of the space object and the satellite, the Kepler orbital equation is used to calculate the minimum distance between the two orbits. If the distance is greater than the threshold, a collision is considered unlikely. The threshold distance is set based on the specific application scenario.

[0070] The time feature screening method is as follows: according to the orbital elements of the space object and its own satellite, the Kepler orbital equations are used based on the true anomaly to obtain the time of approach (TCA) of the two orbits, and then the collision risk time domain is obtained.

[0071] The coarse orbit prediction screening method uses the SGP4 / SDP4 prediction model to perform coarse orbit prediction within the collision risk time domain determined by the temporal feature screening method. The prediction step size and threshold distance are set according to specific application requirements. If the distance between a space object and a satellite falls below the threshold within a certain step size, the object is considered to have a high collision risk and requires subsequent high-precision screening.

[0072] It is understandable that if the conditions are met for ground calculation and sending instructions to satellites, the above-mentioned feature screening method can be directly used, or combined with intelligent screening, to screen high-risk collision sources.

[0073] Using this method, as much historical data on space objects as possible is screened to distinguish high-risk and low-risk collision sources. This historical data is used as training samples, with the object's orbital state as the input feature and whether it is a high-risk or low-risk collision source as the label. A basic neural network model is established and trained using supervised learning methods using the training samples to obtain optimal neural network parameters, resulting in a pre-trained intelligent screening neural network model. This trained intelligent screening neural network model, when fed with the space object's state, directly outputs whether it is a high-risk collision source. This intelligent screening neural network model is deployed in the onboard computer to enable rapid intelligent screening and identify potential high-risk collision sources.

[0074] Step 3: According to the orbital status history records of the possible high-risk collision sources and the own orbital status, the determined high-risk collision source is obtained.

[0075] Specifically, the initial measurement error covariance matrix is ​​derived from the orbital state history of the potential high-risk collision source. A high-precision dynamics model is used to accurately predict the initial orbital state and the initial measurement error covariance matrix, resulting in the orbital state and measurement error covariance matrix at the moment of closest approach between the collision source and the satellite.

[0076] It should be noted that the high-precision dynamics model is a conventional model. The orbital state and measurement error covariance matrix at the initial time are obtained from the space object database and used as input for the high-precision forecast. The orbital state and measurement error covariance matrix at the time of closest approach are also obtained through high-precision forecasting.

[0077] The collision risk is assessed based on the orbital state and measurement error covariance matrix of the high-risk collision source at the time of closest approach, as well as the orbital state and measurement error covariance matrix of the friendly satellite at the time of closest approach. For ease of understanding, the following briefly describes a quantitative assessment method for collision risk.

[0078] First, calculate the collision probability. The relative position vector and relative velocity vector of the space target at the moment of closest approach to the satellite are perpendicular. The two targets are located on a plane perpendicular to the relative velocity, which is defined as the encounter plane. Based on the orbital states of the two targets, the relative position vector, relative velocity vector, and joint error covariance of the space target relative to the satellite are calculated and projected onto the encounter plane. The collision probability is calculated as follows:

[0079]

[0080] in, and are the x and y components of the projection of the joint error covariance onto the meeting plane, Represents the distance between two targets when they are closest. and are the x and y components of the projection of the relative position vector on the encounter plane at the moment of closest approach, Represents the natural exponential function.

[0081] Construct a comprehensive collision risk assessment function, the expression is as follows:

[0082]

[0083] Among them, a1 to a7 are weight coefficients, R, S, and W are the components of the relative position of the two targets at the time of closest approach in the orbital radial, track direction, and normal direction, and L R 、L S 、L W are the reference approach distances in the radial, track, and normal directions, respectively; T1 and T2 are the current time from the TCA and the time from the most recent orbit measurement from the TCA, respectively; Q is the expert score for orbit determination quality, which takes into account observation accuracy and orbit model accuracy; and k and K are custom coefficients. Specifically, the weight coefficients, custom coefficients, and reference approach distances should be set according to actual application requirements.

[0084] The collision risk assessment function calculates the collision risk assessment value of high-risk collision sources. When the collision risk of a space target exceeds the set threshold, it is considered that the satellite needs to avoid the target and enters the collision avoidance decision-making phase.

[0085] Decision-making stage:

[0086] Step 4: The evasive maneuver plan is determined by generating initial values ​​using an intelligent decision-making neural network model. The orbital state of the collision target and the orbital state of the own satellite are input into the intelligent decision-making neural network model to obtain the initial values ​​for the evasive maneuver plan. Based on the orbital states, constraints, and collision risk of the collision target and the own satellite, a multi-objective optimization algorithm is used to optimize the initial values ​​of the evasive maneuver plan. The plan is constrained based on the target value range, and the optimized evasive maneuver plan is modified to obtain the final evasive maneuver plan for subsequent execution. The intelligent decision-making algorithm's purpose is to quickly generate an appropriate evasive maneuver plan, reducing the computational complexity of the subsequent multi-objective optimization algorithm and improving rapid response capabilities.

[0087] The basic idea of ​​a satellite performing a collision avoidance maneuver is to generate thrust through the onboard engine, change the satellite's orbit, and separate it from the orbit of the collision target to avoid a possible collision. There are two main operating modes of satellite engines: pulse maneuvers and continuous low-thrust maneuvers. In the former, the satellite has sufficient thrust to significantly change the satellite's orbital speed instantly, and the engine's work can be regarded as a pulse. In the latter, the satellite has a smaller thrust and needs to be continuously pushed for a period of time to achieve effective avoidance. This scheme takes the continuous low-thrust collision avoidance method as an example. The avoidance maneuver scheme is expressed as ,in The engine start and shut down time, It is the components of the thrust vector in the radial, track and normal directions of the orbit when the engine is working.

[0088] Specifically, the training method of the intelligent decision-making neural network model is as follows:

[0089] The first step is to create a machine learning sample set using a multi-objective algorithm. Based on historical satellite collision events, including the orbital state of the space object, the orbital state of the satellite, constraints, and collision risk, a multi-objective optimization algorithm is used to determine the optimal avoidance maneuver strategy. The space object's orbital state, satellite's orbital state, constraints, and collision risk are used as input features, and the avoidance maneuver strategy is used as the output feature to construct a training sample set.

[0090] Next, the neural network model is trained using the machine learning sample set. Various basic neural network models can achieve the functions required by this solution, including but not limited to fully connected neural networks, recurrent neural networks, and convolutional neural networks. Using open-source neural network toolkits such as Pytorch, the neural network model, including trainable network parameters, is initialized. The neural network model is then trained using the machine learning sample set using supervised learning methods to optimize the network parameters. Training is complete when the optimal network parameters are achieved.

[0091] The trained neural network model can mimic the aforementioned multi-objective optimization algorithm, taking as input the orbital state of a space object, the orbital state of a satellite, constraints, and collision risk, and directly outputs an evasive maneuver plan that meets the requirements. This neural network model is deployed in an onboard computer to enable rapid, intelligent decision-making.

[0092] In this embodiment, after receiving a collision warning and determining the orbital states, constraints, and collision risk of the target and the own satellite, the intelligent decision-making algorithm onboard the satellite's onboard computer generates initial values ​​for an evasive maneuver plan, performs a high-precision orbital prediction of the satellite's orbital state under the current plan, and calculates constraints and optimization objectives. If the optimization objectives and constraints fail to meet the avoidance requirements, a multi-objective optimization algorithm is used to optimize the evasive maneuver plan.

[0093] Specifically, the multi-objective optimization algorithm is as follows:

[0094] The safety factor is represented by the collision risk F, which needs to meet the following conditions:

[0095]

[0096] in The collision risk threshold value needs to be set according to the specific application scenario. The time factor refers to the time it takes for the engine to start and stop during the avoidance maneuver and the time it takes for the warning to be received. , the time between the two targets when they are closest , the start and end time of the ground-controllable time interval , the maximum start-up time of satellite orbit control engine The power on / off time must meet the following conditions:

[0097]

[0098] The above optimization objectives include fuel factors and mission factors. The fuel factor refers to the fuel consumption of the engine during the evasive maneuver. During the optimization, the fuel consumption should be reduced as much as possible. The expression of fuel consumption is as follows:

[0099]

[0100] in represents the satellite mass, Represents the engine thrust. In the case of this scheme, due to the use of continuous small thrust mode, the thrust size is expressed as Mission factors refer to the performance of the satellite's own functions and the mission it performs. During optimization, this performance should be maximized, meaning the impact of evasive maneuvers on the satellite's mission should be minimized. Mission factors are determined based on the specific satellite type and mission scenario being deployed. For example, for Earth observation satellites, mission factors can be represented by three metrics: temporal resolution of coverage, total coverage duration, and mission target point coverage.

[0101] The multi-objective optimization algorithm described above adjusts the evasive maneuver plan X through the optimization algorithm, ensuring that the safety and time constraints within the constraints meet the aforementioned conditions, and minimizing the fuel factor and maximizing the mission factor within the optimization objective. There are many specific implementation methods for the optimization algorithm, which are not limited in this proposal.

[0102] Execution stage:

[0103] Step 5: After obtaining an evasive maneuver strategy that satisfies all constraints and achieves the optimal optimization objective, the execution phase begins. The following steps are divided into two execution paths, depending on whether the previous steps are executed by the onboard computer.

[0104] Due to limited computing power onboard, early warning and decision-making processes are primarily performed by ground-based computers. Therefore, evasive maneuvers must be uploaded to the satellite via ground tracking and control stations before execution. First, a high-precision orbit prediction model is used to predict the satellite's orbital state from the current moment to the time of closest approach (TCA) to the collision target. This state is then compared with the coverage of all available ground tracking and control stations to determine if it falls within the ground tracking and control range. This results in a visible time interval and the corresponding visible tracking and control stations. During the closest visible time interval, the evasive maneuver is uploaded to the satellite from the corresponding visible tracking and control station.

[0105] If the computing power of the onboard computer is sufficient, the early warning and decision-making stages are executed on the onboard computer. The final evasive maneuver plan has been stored in the onboard computer and does not need to be recorded on the ground. It can directly enter the next step of the execution stage.

[0106] Given the optimized evasive maneuver plan in the decision link, the actuator response step is performed, that is, according to the evasive maneuver plan The engine on / off time and thrust vector contained in it are used to control the onboard engine to operate with the specified thrust vector within the given on / off time and perform collision avoidance maneuvers.

[0107] Step 6. After the actuator responds, satellite status feedback and avoidance effect evaluation are performed. Satellite status feedback includes determining the current orbital state of the satellite through ground-based tracking and control station observations, space-based navigation satellite positioning, and on-board autonomous observations. Avoidance effect evaluation includes determining the current orbital state of the space object using the same method as satellite status feedback, and calculating the collision risk F based on the orbital state and measurement error covariance matrix of the high-risk collision source at the current moment, and the orbital state and measurement error covariance matrix of the own satellite at the current moment, using the aforementioned method, to determine whether it is higher than the collision risk threshold. If F is higher than , then return to the decision-making phase, repeat the above decision-making and execution phase steps, and perform evasive maneuvers again. , the current high-risk collision source is considered to have been successfully avoided, the collision avoidance process ends, and the system returns to the early warning phase to continue monitoring for new collision risks. By feeding the results of the completed evasive maneuver back to the previous phase, the avoidance effect is dynamically corrected, achieving a closed-loop process.

[0108] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the described embodiments. It will be apparent to those skilled in the art that various changes, modifications, substitutions, and variations of these embodiments, including components, without departing from the principles and spirit of the present invention are still within the scope of protection of the present invention.

Claims

1. A full-process closed-loop rapid response satellite collision avoidance method, characterized in that: The steps include: Step 1: Observe and obtain the orbital state of space objects, then fuse multi-source data to build a space object database; Step 2: Train an intelligent screening neural network model. Use the pre-trained intelligent screening neural network model to screen the orbital states of all space objects and identify high-risk collision sources that may collide with your satellite. Step 3: Obtain a confirmed high-risk collision source based on the orbital status history records of the possible high-risk collision source and the own orbital status; Step 4: Train an intelligent decision-making neural network model. Use the orbital state of the determined high-risk collision source and the orbital state of the own satellite as input to generate an initial evasive maneuver plan through the pre-trained intelligent decision-making neural network model. Optimize the initial evasive maneuver plan through a multi-objective optimization algorithm to obtain the final evasive maneuver plan for subsequent execution. Step 5: Based on the engine startup and shutdown times and thrust vectors included in the final avoidance maneuver plan, control the onboard engines to operate at the specified thrust vector within the given startup and shutdown times to perform the collision avoidance maneuver; Step 6: After the maneuver is completed, the satellite status feedback is performed to obtain the satellite orbit status after the maneuver, and the avoidance effect is evaluated to obtain the collision risk after the avoidance. If the collision risk is still higher than the threshold value, return to step 4 and repeat the above process. If the collision risk is lower than the threshold, return to step 1 and repeat the warning and avoidance of potential high-risk collision sources.

2. The satellite collision avoidance method with a full-process closed-loop rapid response according to claim 1, characterized in that: The orbital state refers to any physical quantity that can describe the position and velocity of a spacecraft in space.

3. The satellite collision avoidance method with a full-process closed-loop rapid response according to claim 1, characterized in that: In step 1, the method for obtaining the orbital state of the space object includes: ground station observation, GNSS satellite observation, on-board autonomous observation and orbit prediction of known objects.

4. The satellite collision avoidance method with a full-process closed-loop rapid response according to claim 1, characterized in that: The training method of the intelligent screening neural network model is: First, the space object database is used as input. Geometric feature screening algorithms, temporal feature screening algorithms, and orbital rough prediction screening algorithms are applied to screen space objects, distinguish high-risk and low-risk collision sources, and label them, thereby establishing machine learning samples. A basic neural network model is constructed, and then the supervised learning method is used to train the neural network model using the above machine learning sample set to optimize the network parameters. When the optimal network parameters are obtained, the training is completed and a pre-trained intelligent screening neural network model is obtained.

5. The satellite collision avoidance method with a full-process closed-loop rapid response according to claim 1, characterized in that: In step 3, the method for determining the high-risk collision source is: Step 3.1, based on the orbital state history of the collision source and the own satellite's orbital state, obtain the orbital state and measurement error covariance matrix at the moment of closest approach between the collision source and the own satellite; Step 3.2: Based on the orbital state and measurement error covariance matrix of the collision source at the closest approach, and the orbital state and measurement error covariance matrix of the own satellite at the closest approach, the collision risk is assessed and the high-risk collision source is determined.

6. The satellite collision avoidance method with a full-process closed-loop rapid response according to claim 5, characterized in that: In step 3.1, the measurement error covariance matrix at the initial moment is obtained based on the historical record of the orbital state of the collision source; then a high-precision dynamic model is used to perform high-precision prediction of the orbital state of the collision source and the own satellite at the initial moment and the measurement error covariance matrix at the initial moment, and the orbital state and measurement error covariance matrix at the moment when the collision source and the own satellite are closest are obtained.

7. The satellite collision avoidance method with a full-process closed-loop rapid response according to claim 5, characterized in that: In step 3.2, the collision risk assessment method is: First, calculate the collision probability, the expression is as follows: ； in, and are the x and y components of the projection of the joint error covariance onto the meeting plane, Represents the distance between two targets when they are closest. and are the x and y components of the projection of the relative position vector on the encounter plane at the moment of closest approach, represents the natural exponential function; Construct a comprehensive collision risk assessment function, the expression is as follows: ； Among them, a1 to a7 are weight coefficients, R, S, and W are the components of the relative position of the two targets at the time of closest approach in the orbital radial, track direction, and normal direction, and L R 、L S 、L W are the reference approach distances in the orbit radial, track, and normal directions, respectively; T1 and T2 are the time from the current moment to the TCA (i.e., the moment of closest approach) and the time from the most recent orbit measurement to the TCA, respectively; Q is the expert score for orbit determination quality taking into account the observation accuracy and orbit model accuracy; k and K are custom coefficients; Calculate the collision risk assessment value of high-risk collision sources. When the collision risk of a space target is higher than the set threshold, it is considered that the satellite needs to avoid the target.

8. The satellite collision avoidance method with a full-process closed-loop rapid response according to claim 1, characterized in that: The training method of the intelligent decision-making neural network model is as follows: the orbital state of the space object, the orbital state of the satellite, the constraints, and the collision risk in the historical satellite collision events are used as input features, and the avoidance maneuver strategy is used as the output feature, and a multi-objective optimization algorithm is used to construct a training sample set; Then, a basic neural network model is established, and the supervised learning method is used to train the neural network model using the above training samples to obtain the optimal neural network parameters. After the training is completed, a pre-trained intelligent decision-making neural network model is obtained.

9. The satellite collision avoidance method with a full-process closed-loop rapid response according to claim 1, characterized in that: In step 4, the method for optimizing the initial evasive maneuver plan using a multi-objective optimization algorithm is as follows: The safety factor is represented by the collision risk F, which needs to meet the following conditions: ； in, is the collision risk threshold; The time factor refers to the time it takes for the engine to be turned on and off during the avoidance maneuver and the time it takes for the warning to be given. , the time between the two targets when they are closest , the start and end time of the ground-controllable time interval , the maximum start-up time of satellite orbit control engine The satellite orbit control engine power on / off time must meet the following conditions: ； The fuel factor refers to the fuel consumption of the engine during the evasive maneuver. During optimization, the fuel consumption should be reduced as much as possible. The expression of fuel consumption is as follows: ； in represents the satellite mass, Represents the engine thrust; Mission factors refer to the satellite's own functions and the work efficiency of the tasks it performs. During optimization, the work efficiency should be improved as much as possible, that is, the impact of avoidance maneuvers on the satellite's own mission should be reduced.

10. A full-process closed-loop rapid response satellite collision avoidance method according to any one of claims 1 to 9, characterized in that: The basic network used by the intelligent screening neural network model and the intelligent decision-making neural network model is any one of a fully connected neural network, a recursive neural network, and a convolutional neural network.

Images