Satellite orbit prediction method and system and electronic equipment
By establishing dynamic heterogeneous graphs and multi-timescale modeling, and combining short-term dynamics and long-term trend models, the satellite orbit prediction results are optimized, solving the problems of multi-source data integration and dynamic environment adaptability, and improving prediction accuracy and adaptability.
Patent Information
- Application Number
- CN202511668163.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-14
- Publication Date
- 2026-03-03
AI Technical Summary
Existing satellite orbit prediction systems cannot simultaneously integrate multi-source data, adapt to dynamic environmental changes, or have adaptive capabilities, resulting in low prediction accuracy.
By acquiring multi-source heterogeneous data, establishing a dynamic heterogeneous graph, constructing a global spatiotemporal feature matrix, using a gating mechanism for multi-timescale modeling, creating short-term dynamic models and long-term trend models, and optimizing prediction results through an error compensation model.
It significantly improves the accuracy and adaptability of orbit prediction, and can adaptively capture short-term disturbances and long-term evolution patterns of the orbital environment, achieving a balance between short-term sensitivity and long-term stability.
Smart Images
Figure CN121598288A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of satellite orbit prediction technology, specifically providing a satellite orbit prediction method, system, and electronic equipment. Background Technology
[0002] With the rapid development of science and technology, satellite orbit prediction technology is playing an increasingly important role in space situational awareness and national security. Traditional orbit prediction methods mainly rely on orbital dynamics models, which model the various perturbations experienced by the satellite during its operation to complete the prediction. However, this method inevitably suffers from low modeling accuracy due to the complexity of the forces acting on the satellite in space and the uncertainty of the satellite target parameters, directly resulting in low accuracy in the target satellite orbit prediction.
[0003] In recent years, the application of deep learning technology has been continuously expanding in various fields, including significant progress in weather forecasting and typhoon path prediction. However, in the field of satellite orbit prediction, existing deep learning methods still have some shortcomings. First, a single data source and model cannot fully capture the multidimensional dynamic characteristics of complex orbital environments, resulting in limited monitoring range and temporal continuity. Second, traditional prediction methods often rely on fixed rules or empirical formulas, making it difficult to process dynamically changing data in real time and respond promptly to complex and ever-changing orbital environments. Finally, existing prediction models lack effective integration and fusion mechanisms when processing multi-source heterogeneous data, making it difficult to fully explore the correlations and synergistic effects between data.
[0004] Accordingly, a new technical solution is needed in this field to solve the above problems. Summary of the Invention
[0005] The present invention aims to solve the above-mentioned technical problems, and to address the issues in the existing technology where satellite orbit prediction systems cannot simultaneously integrate multi-source data, adapt to dynamic environmental changes, and lack adaptive capabilities.
[0006] This invention provides a method for predicting the orbit of a satellite, comprising: acquiring multi-source heterogeneous data, wherein the multi-source heterogeneous data includes IMU data, orbit data, and space environment data; The multi-source heterogeneous data is preprocessed and normalized. Based on the processed multi-source heterogeneous data, a dynamic heterogeneity graph is established. Based on the dynamic heterogeneous graph, a global spatiotemporal feature matrix is constructed; By using a gating mechanism, global spatiotemporal features are modeled at multiple time scales to capture features at different time scales; Based on the characteristics of the different time scales, short-term dynamic models and long-term trend models are created; The short-term dynamic model prediction results and the long-term trend model prediction results are combined to form a fusion result; The orbit prediction result is calculated based on the error between the fusion result and the actual result.
[0007] By employing the aforementioned technical solution, a dynamic heterogeneous graph was established through the acquisition and processing of multi-source heterogeneous data, effectively integrating data from different sources and types. This overcomes the problem of traditional methods relying on a single data source and struggling to comprehensively capture the multidimensional characteristics of complex orbital environments, providing a richer and more accurate data foundation for subsequent analysis. By constructing a global spatiotemporal feature matrix and utilizing a gating mechanism for multi-timescale modeling, short-term disturbances and long-term evolutionary patterns in the orbital environment can be adaptively captured. This dynamic feature extraction method outperforms traditional methods based on fixed rules or empirical formulas, significantly enhancing the understanding and prediction capabilities of complex and variable orbital environments. By creating and fusing short-term dynamic models and long-term trend models, and optimizing the prediction results based on the error between the fused results and actual results, a balance between short-term sensitivity and long-term stability is achieved. This mechanism continuously optimizes the model, improving the overall accuracy of orbital prediction and the system's adaptability to practical tasks.
[0008] In the specific implementation of the above-mentioned satellite orbit prediction method, the nodes in the dynamic heterogeneous graph represent data sources, and the edges describe the relationships between data sources.
[0009] By employing the aforementioned technical solution, the specific semantics of nodes and edges in dynamic heterogeneous graphs are clearly defined, providing a clear modeling foundation for graph neural networks. This helps to more accurately extract and utilize complex nonlinear relationships between data, improving the accuracy of feature extraction. In the specific implementation of the above-mentioned satellite orbit prediction method, the step of constructing a global spatiotemporal feature matrix based on the dynamic heterogeneous graph specifically includes: A dynamic multimodal graph neural network is used to perform multi-level graph convolution processing on the dynamic heterogeneous graph to extract the nonlinear relationships between multiple data sources; A graph convolutional network is used to iteratively update the features of the nodes to obtain a global spatiotemporal feature matrix.
[0010] By employing the aforementioned technical solution, using a dynamic multimodal graph neural network for multi-layer graph convolution processing and iteratively updating node features, it is possible to more deeply uncover the implicit and complex nonlinear relationships between multi-source data. This helps to construct a more comprehensive and accurate global spatiotemporal feature matrix, thereby enhancing the model's comprehensive perception of orbital dynamics.
[0011] In the specific implementation of the above-mentioned satellite orbit prediction method, the step of modeling global spatiotemporal features at multiple time scales through a gating mechanism to capture features at different time scales further includes: Capture the 30-minute, 2-hour, and 24-hour timescale features in the global spatiotemporal features respectively; Features are extracted through convolution operations to obtain the feature matrix; The feature map is obtained by performing a nonlinear transformation using the Tanh activation function; The feature maps are dynamically fused across multiple time scales using a gating mechanism to obtain features at different times.
[0012] By employing the aforementioned technical solution, and through the capture of features at different time scales and dynamic fusion using a gating mechanism, the model can adaptively focus on evolutionary patterns at different time granularities. This enhances the ability to perceive and represent orbital dynamics at a fine-grained level.
[0013] In a specific implementation of the above-described satellite orbit prediction method, the step of creating a short-term dynamic model and a long-term trend model based on the characteristics of the different time scales further includes: An LSTM network is used to model short-term dynamics and obtain short-term prediction results for 30 minutes and 2 hours. The physical constraint model is used to model the long-term trend and obtain long-term forecast results for 24 hours and 48 hours.
[0014] By employing the aforementioned technical solutions, using LSTM networks for data-driven modeling of short-term dynamics effectively learns complex patterns in historical sequences; simultaneously, utilizing physical constraint models to model long-term trends introduces prior knowledge of orbital dynamics. This combination enhances the model's generalization ability and the physical plausibility of long-term predictions.
[0015] In a specific implementation of the above-mentioned satellite orbit prediction method, the step of fusing the short-term dynamic model prediction results and the long-term trend model prediction results to form a fused result further includes: The prediction results of the short-term dynamic model and the prediction results of the long-term trend model are combined using a weighted method.
[0016] By employing the above technical solution, a weighted fusion approach is used to combine short-term and long-term prediction results, resulting in a simple and flexible mechanism. By adjusting the weighting parameters, the final prediction results can be optimized for different satellite or orbit types, balancing the contributions of short-term accuracy and long-term trends.
[0017] In the specific implementation of the above-mentioned satellite orbit prediction method, the step of calculating the orbit prediction result based on the error between the fusion result and the actual result further includes: Calculate the residual between the fusion result and the actual result; An error compensation model is constructed using the residuals; the model is in the form of linear regression. The parameters of the error compensation model are optimized by using the gradient descent algorithm to minimize the residual between the predicted result and the actual value. The convergence condition is that the number of iterations reaches 1000 or the residual is less than 0.01. The fusion results are substituted into the error compensation model to calculate the trajectory prediction results.
[0018] By employing the above technical solution, the system error is effectively reduced by calculating the residual between the fusion result and the actual result and constructing an error compensation model (such as linear regression) for correction. The gradient descent algorithm is used to optimize the parameters and set explicit convergence conditions, ensuring the effectiveness and efficiency of the compensation model, thereby significantly improving the accuracy of the final trajectory prediction.
[0019] In the specific implementation of the above-mentioned satellite orbit prediction method, the prediction results are prediction results at different time scales.
[0020] By adopting the above technical solution, the prediction results are clearly defined to cover different time scales, enabling the method to simultaneously provide the prediction information required for different application scenarios such as short-term obstacle avoidance, medium-term task planning, and long-term trend analysis, thereby enhancing the practicality and applicability of the method.
[0021] This invention also discloses a satellite orbit prediction system, comprising: The data preprocessing module is used to perform data preprocessing operations on multi-source heterogeneous data to obtain a dynamic heterogeneous graph; The feature extraction module is used to model global spatiotemporal features at multiple time scales through a gating mechanism, capturing features at different time scales. The modeling module is used to create short-term dynamic models and long-term trend models based on the features at the different time scales. The fusion module is used to fuse the short-term dynamic model prediction results and the long-term trend model prediction results to form a fusion result; The error calculation module is used to generate orbit prediction results through the error compensation model and fusion results.
[0022] By adopting the above technical solution, the trajectory prediction system is divided into dedicated modules such as data preprocessing, feature extraction, modeling, fusion, and error calculation, achieving functional decoupling and systematic processing. This architecture not only improves processing efficiency and facilitates maintenance and upgrades, but also allows for flexible configuration and expansion to meet different task requirements.
[0023] The present invention also discloses an electronic device, comprising: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform the satellite orbit prediction method described above.
[0024] When the above technical solution is adopted, the above method can be executed in an electronic device, which provides a hardware carrier for the method. Attached Figure Description
[0025] The preferred embodiments of the present invention are described below with reference to the accompanying drawings, in which: Figure 1 This is a flowchart of the main steps in satellite orbit prediction methods. Detailed Implementation
[0026] Preferred embodiments of this application are described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of this application and are not intended to limit the scope of protection of this application. Those skilled in the art can make adjustments as needed to adapt to specific application scenarios.
[0027] like Figure 1 As shown, to address the problems of existing satellite orbit prediction systems being unable to simultaneously integrate multi-source data, adapt to dynamic environmental changes, and lack adaptive capabilities, this invention provides a satellite orbit prediction method, comprising: S101 acquires multi-source heterogeneous data, including IMU data, orbital data, and space environment data. Specifically, IMU data refers to the raw data collected by the inertial measurement unit that describes the satellite's three-dimensional motion. Orbital data includes the satellite's position vector, velocity vector, orbital elements, and timestamp. Space environment data includes residual atmospheric density at orbital altitude, geomagnetic field, and solar activity parameters. The specific methods for detecting the above data are existing technologies and will not be elaborated here.
[0028] S102 involves preprocessing and normalizing the multi-source heterogeneous data. Specifically, time alignment is performed on the multi-source heterogeneous data, unifying all data to the Coordinated Universal Time (UTC) time base. All data are then normalized, and all normalized features are mapped to a unified interval of [-1, 1] or [0, 1].
[0029] S103. Based on the processed multi-source heterogeneous data, a dynamic heterogeneous graph is established. Specifically, nodes in the dynamic heterogeneous graph represent data sources, and edges describe the relationships between data sources. There is an edge between IMU nodes and track nodes, indicating their correlation. There is an edge between environment nodes and track nodes, indicating the influence of the environment on the track. There may also be an edge between environment nodes and IMU nodes, indicating the influence of the environment on IMU measurements. The weights or features of the edges change over time. The features of IMU nodes are the IMU data at time step t; the features of track nodes are the track data at time step t; and the features of environment nodes are the environment data at time step t. An attention mechanism is used to calculate the weight of each edge at the current time step, which considers the features of the two connected nodes at the current time step. For each time step t, a graph G(t) = (V, E(t)) is generated, where V is the set of nodes, E(t) is the set of edges at time t, and each edge has a weight α. The dynamic heterogeneous graph {G(t1), G(t2), ..., G(tn)} can be formed using the above data. It should be noted that the above is only a simple description of dynamic heterogeneous graphs. The specific steps for forming dynamic heterogeneous graphs are existing technologies and will not be elaborated here.
[0030] S104, based on dynamic heterogeneous graphs, constructs a global spatiotemporal feature matrix. Specifically, a dynamic multimodal graph neural network is used to perform multi-level graph convolution processing on the dynamic heterogeneous graph to extract nonlinear relationships between multiple data sources. A graph convolutional network is then used to iteratively update the node features to obtain the global spatiotemporal feature matrix. Thus, by using a dynamic multimodal graph neural network for multi-level graph convolution processing and iteratively updating node features, it is possible to more deeply uncover the implicit and complex nonlinear relationships between multi-source data. This helps to construct a more comprehensive and accurate global spatiotemporal feature matrix, thereby improving the model's comprehensive perception capability of orbital dynamics.
[0031] S105 employs a gating mechanism to model global spatiotemporal features across multiple time scales, capturing features at different time scales. Specifically, it captures 30-minute, 2-hour, and 24-hour time scale features from the global spatiotemporal features. Features are extracted through convolution operations to obtain a feature matrix. A nonlinear transformation using the Tanh activation function yields feature maps. These feature maps are then dynamically fused across multiple time scales using a gating mechanism to obtain features at different times. Thus, by capturing features at different time scales (e.g., 30 minutes, 2 hours, 24 hours) from the global spatiotemporal features and dynamically fusing them using a gating mechanism, the model can adaptively focus on evolutionary patterns at different time granularities. This enhances the fine-grained perception and representation of dynamic changes in the orbit.
[0032] S106, based on characteristics at different time scales, creates short-term dynamic models and long-term trend models. Specifically, an LSTM network is used to model short-term dynamics, obtaining short-term predictions for 30 minutes and 2 hours. A physical constraint model is used to model long-term trends, obtaining long-term predictions for 24 hours and 48 hours. Thus, using an LSTM network for data-driven modeling of short-term dynamics effectively learns complex patterns in historical sequences; simultaneously, using a physical constraint model to model long-term trends introduces prior knowledge of orbital dynamics. This combination enhances the model's generalization ability and the physical plausibility of long-term predictions.
[0033] S107, the fusion result is formed by combining the predictions of the short-term dynamic model and the long-term trend model. Specifically, a weighted method is used to fuse the predictions of the short-term dynamic model and the long-term trend model. This weighted fusion method combines short-term and long-term predictions, offering a simple and flexible mechanism. By adjusting the weight parameters, the final prediction result can be optimized for different satellite or orbit types, balancing the contributions of short-term accuracy and long-term trends.
[0034] S108 calculates the trajectory prediction result based on the error between the fused result and the actual result. Specifically, it calculates the residual between the fused result and the actual result. An error compensation model is constructed using the residual, with the model taking the form of linear regression. The parameters of the error compensation model are optimized using the gradient descent algorithm to minimize the residual between the predicted result and the actual value. The convergence condition is that the number of iterations reaches 1000 or the residual is less than 0.01. The fused result is then substituted into the error compensation model to calculate the trajectory prediction result. Thus, by calculating the residual between the fused result and the actual result and constructing an error compensation model (such as linear regression) for correction, the system error is effectively reduced. The use of the gradient descent algorithm to optimize the parameters and setting explicit convergence conditions ensures the effectiveness and efficiency of the compensation model, thereby significantly improving the accuracy of the final trajectory prediction. The prediction results are presented at different time scales. This explicit coverage of different time scales allows the method to simultaneously provide the prediction information required for different application scenarios such as short-term obstacle avoidance, medium-term task planning, and long-term trend analysis, enhancing the method's practicality and applicability. By acquiring and processing multi-source heterogeneous data, a dynamic heterogeneous graph was established, effectively integrating data from different sources and types. This overcomes the problem of traditional methods relying on a single data source and struggling to comprehensively capture the multidimensional characteristics of complex orbital environments, providing a richer and more accurate data foundation for subsequent analysis. By constructing a global spatiotemporal feature matrix and utilizing a gating mechanism for multi-timescale modeling, short-term disturbances and long-term evolution patterns in the orbital environment can be adaptively captured. This dynamic feature extraction method outperforms traditional methods based on fixed rules or empirical formulas, significantly enhancing the understanding and prediction capabilities of complex and variable orbital environments. By creating and fusing short-term dynamic models and long-term trend models, and optimizing the prediction results based on the error between the fused results and the actual results, a balance between short-term sensitivity and long-term stability is achieved. This mechanism continuously optimizes the model, improving the overall accuracy of orbit prediction and the system's adaptability to practical tasks.
[0035] This invention also discloses a satellite orbit prediction system, comprising: a data preprocessing module for preprocessing multi-source heterogeneous data to obtain a dynamic heterogeneous map; a feature extraction module for modeling global spatiotemporal features at multiple time scales using a gating mechanism to capture features at different time scales; a modeling module for creating short-term dynamic models and long-term trend models using features at different time scales; a fusion module for fusing the prediction results of the short-term dynamic model and the prediction results of the long-term trend model to form a fused result; and an error calculation module for generating orbit prediction results using an error compensation model and the fused result. Thus, by dividing the orbit prediction system into dedicated modules such as data preprocessing, feature extraction, modeling, fusion, and error calculation, functional decoupling and systematic processing are achieved. This architecture not only improves processing efficiency and facilitates maintenance and upgrades but also allows for flexible configuration and expansion to meet different mission requirements.
[0036] The present invention also discloses an electronic device, comprising: at least one processor; a memory communicatively connected to the at least one processor; the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the aforementioned satellite orbit prediction method. Thus, the aforementioned method can be executed in the electronic device, which provides the hardware carrier for the method.
[0037] Those skilled in the art will understand that all or part of the processes in the method of the above embodiment of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable storage medium can include any entity or device capable of carrying computer program code, media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable storage medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.
[0038] The various component embodiments of the present invention can be implemented in hardware, or as software modules running on one or more processors, or a combination thereof. Those skilled in the art will understand that microprocessors or digital signal processors (DSPs) can be used in practice to implement some or all of the functions of some or all of the components in a satellite platform or ground platform according to embodiments of the present invention. The present invention can also be implemented as a device or apparatus program (e.g., a PC program and PC program products) for performing some or all of the methods described herein. Such programs implementing the present invention can be stored on a PC-readable medium, or can be in the form of one or more signals. Such signals can be downloaded from an Internet website, provided on a carrier signal, or provided in any other form.
[0039] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after such changes or substitutions will all fall within the scope of protection of the present invention.
Claims
1. A method for predicting the orbit of a satellite, characterized in that, include: Acquire multi-source heterogeneous data, including IMU data, orbital data, and space environment data; The multi-source heterogeneous data is preprocessed and normalized. Based on the processed multi-source heterogeneous data, a dynamic heterogeneity graph is established. Based on the dynamic heterogeneous graph, a global spatiotemporal feature matrix is constructed; By using a gating mechanism, global spatiotemporal features are modeled at multiple time scales to capture features at different time scales; Based on the characteristics of the different time scales, short-term dynamic models and long-term trend models are created; The short-term dynamic model prediction results and the long-term trend model prediction results are combined to form a fusion result; The orbit prediction result is calculated based on the error between the fusion result and the actual result.
2. The satellite orbit prediction method according to claim 1, characterized in that, In the dynamic heterogeneous graph, nodes represent data sources, and edges describe the relationships between data sources.
3. The satellite orbit prediction method according to claim 2, characterized in that, The step of constructing a global spatiotemporal feature matrix based on the dynamic heterogeneous graph specifically includes: A dynamic multimodal graph neural network is used to perform multi-level graph convolution processing on the dynamic heterogeneous graph to extract the nonlinear relationships between multiple data sources; A graph convolutional network is used to iteratively update the features of the nodes to obtain a global spatiotemporal feature matrix.
4. The satellite orbit prediction method according to claim 2, characterized in that, The step described above uses a gating mechanism to model global spatiotemporal features across multiple time scales, and capturing features at different time scales further includes: Capture the 30-minute, 2-hour, and 24-hour timescale features in the global spatiotemporal features respectively; Features are extracted through convolution operations to obtain the feature matrix; The feature map is obtained by performing a nonlinear transformation using the Tanh activation function; The feature maps are dynamically fused across multiple time scales using a gating mechanism to obtain features at different times.
5. The satellite orbit prediction method according to claim 4, characterized in that, The step of creating short-term dynamic models and long-term trend models based on the characteristics of the different time scales further includes: An LSTM network is used to model short-term dynamics and obtain short-term prediction results for 30 minutes and 2 hours. The physical constraint model is used to model the long-term trend and obtain long-term forecast results for 24 hours and 48 hours.
6. The satellite orbit prediction method according to claim 5, characterized in that, The step of fusing the short-term dynamic model prediction results and the long-term trend model prediction results to form a fused result further includes: The prediction results of the short-term dynamic model and the prediction results of the long-term trend model are combined using a weighted method.
7. The satellite orbit prediction method according to claim 6, characterized in that, The step of calculating the orbit prediction result based on the error between the fusion result and the actual result also includes: Calculate the residual between the fusion result and the actual result; An error compensation model is constructed using the residuals; the model is in the form of linear regression. The parameters of the error compensation model are optimized by using the gradient descent algorithm to minimize the residual between the predicted result and the actual value. The convergence condition is that the number of iterations reaches 1000 or the residual is less than 0.
01. The fusion results are substituted into the error compensation model to calculate the trajectory prediction results.
8. The satellite orbit prediction method according to claim 7, characterized in that, The prediction results are predictions at different time scales.
9. A satellite orbit prediction system, characterized in that, include: The data preprocessing module is used to perform data preprocessing operations on multi-source heterogeneous data to obtain a dynamic heterogeneous graph; The feature extraction module is used to model global spatiotemporal features at multiple time scales through a gating mechanism, capturing features at different time scales. The modeling module is used to create short-term dynamic models and long-term trend models based on the features at the different time scales. The fusion module is used to fuse the short-term dynamic model prediction results and the long-term trend model prediction results to form a fusion result; The error calculation module is used to generate orbit prediction results through the error compensation model and fusion results.
10. An electronic device, characterized in that, include: At least one processor; A memory that is communicatively connected to the at least one processor; The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the orbit prediction method for the satellite according to any one of claims 1 to 8.