Satellite space-time trajectory data association analysis method based on PNTA

Abstract

The invention discloses a satellite spatio-temporal trajectory data association analysis method based on PNTA, and particularly relates to the technical field of data association analysis. Aggregation and crossing behaviors of trajectories in a sliding window are sensitively captured by introducing trajectory crossing and overlapping risk coefficients; secondly, the modal feature degradation perception coefficient dynamically measures the feature degradation degree caused by shielding, interference or sensor failure of the multi-modal information carried by each track from two dimensions of time sequence stability and spectrum distribution, and the feature degradation degree of the multi-modal information carried by each track is evaluated through joint evaluation. The method can actively trigger a confidence suppression and modal enhancement mechanism when a system enters a high-pseudo-consistency mismatch risk area, avoids the damage of false trajectory combination to subsequent clustering and scheduling path deduction, and can maintain the high efficiency and stability of a matching algorithm in a low-risk state. Therefore, fine control over the misjudgment risk in the trajectory recognition link is achieved, and the accuracy, robustness and system response capability of multi-target trajectory association in a complex environment are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of data correlation analysis technology, and more specifically, to a satellite spatiotemporal trajectory data correlation analysis method based on PNTA. Background Technology

[0002] With the continuous increase in the deployment density of multi-orbit satellite constellations and the widespread collaboration of heterogeneous payload platforms, the fusion and identification of satellite trajectory data are facing unprecedented complex challenges. Especially in low-Earth orbit constellations, a large number of target satellites operate at similar orbital altitudes and velocities, easily leading to spatial intersections and highly overlapping trajectories. Furthermore, under resource constraints or with a single payload mode, trajectory observations of some satellites also suffer from problems such as occlusion interference, data ambiguity, and attitude feature degradation, resulting in decreased accuracy and increased ambiguity for traditional trajectory identification and association methods. Existing trajectory data association methods mostly rely on linear interpolation, fixed sliding windows, or pure spatial distance models, which are difficult to effectively handle nonlinear disturbances between trajectories, asynchronous sampling, and mode degradation caused by occlusion. The Predictive Nonlinear Trajectory Association (PNTA) method, which has emerged in recent years based on predictive modeling and nonlinear matching, while possessing good trajectory evolution trend characterization and confidence association capabilities in most scenarios, still faces significant challenges in complex scenarios where trajectory intersections and overlaps coexist with mode occlusion.

[0003] Specifically, in the intersecting regions of target trajectories, multiple trajectories exhibit highly consistent "apparent features" in terms of spatial position and velocity vectors. Furthermore, modal occlusion further weakens dynamic features used for target identification, such as acceleration and attitude spectra, forcing the PNTA method to rely solely on fuzzy spatial distances and angular relationships for association judgment. This feature dimensionality compression and structural ambiguity easily trigger multiple matching ambiguities and false trajectory merging, ultimately leading to high-confidence but erroneous trajectory matching—the pseudo-trajectory consistency mismatch problem. Once this problem occurs, it severely interferes with subsequent trajectory clustering, track recognition, and scheduling path deduction modules, causing decision-making link interruptions and system performance degradation. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a satellite spatiotemporal trajectory data correlation analysis method based on PNTA to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A method for correlation analysis of satellite spatiotemporal trajectory data based on PNTA includes the following steps: Step S1: Collect multimodal information uploaded from different satellite platforms to construct a trajectory state vector sequence and predict the future state of the trajectory; Step S2: Calculate the initial matching confidence score by performing spatial similarity matching on any two predicted trajectories within the sliding window; Step S3: Obtain trajectory intersection and overlap information between trajectories within the sliding window and calculate the trajectory intersection and overlap risk coefficient; Step S4: Obtain modal feature degradation information between trajectories within the sliding window and calculate the modal feature degradation sensing coefficient; Step S5: Construct a pseudo-trajectory consistency mismatch assessment model based on the trajectory crossover risk coefficient and modal feature degradation perception coefficient, output the pseudo-trajectory consistency mismatch assessment index, and assess the pseudo-trajectory consistency mismatch risk of the current system. Step S6: Dynamically adjust the initial matching confidence based on the pseudo-trajectory consistency mismatch risk of the current system.

[0006] In a preferred embodiment, in step S1, multimodal observation information uploaded from different satellite platforms is collected and fused. For the collected multimodal observation information, a unified time reference is established, and interpolation alignment and sampling normalization are performed. Filters are applied to remove jump points and missing measurement segments, and a trajectory state vector sequence is constructed for each trajectory object. A temporal trajectory representation module with a time attention mechanism is constructed based on the trajectory state vector sequence, and the output trajectory potential state representation sequence is generated. An embedding representation is constructed for the perturbation data. The future time step state is generated by using the trajectory historical state and the perturbation embedding to form a predicted trajectory sequence, thus completing the prediction of the future state of the trajectory.

[0007] In a preferred embodiment, the logic for obtaining the trajectory intersection and overlap risk coefficient is as follows: For any pair of trajectories Construct its temporal compression difference vector ; At the preset distance threshold The following conditions are used to extract low-bifaction segments from continuous temporal sequences: The set of continuous time series low-branch segments that meet the extraction criteria is marked as: ; Calculate the cross-response density coefficient ; Construct a convergent attraction function and output the attraction potential field value between the two trajectories. : ,in The angle between the directions of the two predicted trajectories; The cross-response density coefficient and the attraction potential field value are linked to calculate the trajectory intersection and overlap risk coefficient. : .

[0008] In a preferred embodiment, the logic for obtaining the modal feature degradation perception coefficient is as follows: For each trajectory Within a sliding window, the predicted trajectory state feature data vector time series is extracted, and a manifold learning algorithm is used to map the high-dimensional modal features to a low-dimensional embedding subspace. , ; For each embedded subspace, calculate its state mutation rate sequence along the time dimension: , ,in The mutation rate of the state. As a unit of time, To predict the time step length; extract the extreme value gradient fluctuation ratio. : ,in This is a very small constant to prevent division by zero; For the original trajectory state feature data vector time series Perform wavelet packet transform to obtain its time-spectrum packet distribution in multiple frequency bands: , ,in This is the time-frequency representation of the trajectory state feature data vector. After wavelet packet transform OK, Column time-frequency characteristic matrix, Wavelet packet transform; Calculate the trajectory state characteristic instability distribution value based on the temporal spectrum packet distribution. : , ,in This represents all elements in the h-th column of the time-frequency characteristic matrix. Indicates all rows, This represents the L2 norm distance between vectors; Construct a modal feature degradation mapping function and output the modal feature degradation mapping values. : ; Calculate trajectory pairs Modal feature degradation perception coefficient : ,in For trajectory Modal feature degradation mapping value, For trajectory The modal feature degradation mapping value.

[0009] In a preferred embodiment, a pseudo-trajectory consistency mismatch assessment model is constructed based on the trajectory intersection and overlap risk coefficient and the modal feature degradation perception coefficient, and the pseudo-trajectory consistency mismatch assessment index is output. The formula used in the pseudo-trajectory consistency mismatch assessment model is as follows: In the formula This is a pseudo-trajectory consistency mismatch assessment index. The risk coefficient for trajectory intersection and overlap. The modal feature degradation perception coefficient, These represent the preset proportional coefficients for the trajectory intersection and overlap risk coefficient and the modal feature degradation perception coefficient, respectively. All are greater than 0.

[0010] In a preferred embodiment, the pseudo-trajectory consistency mismatch assessment index is compared with a preset pseudo-trajectory consistency mismatch assessment index threshold to assess the pseudo-trajectory consistency mismatch risk of the current system, as follows: If the pseudo-trajectory consistency mismatch assessment index is greater than the pseudo-trajectory consistency mismatch assessment index threshold, then the pseudo-trajectory consistency mismatch risk of the current system is marked as high pseudo-trajectory consistency mismatch risk. If the pseudo-trajectory consistency mismatch assessment index is less than or equal to the pseudo-trajectory consistency mismatch assessment index threshold, then the pseudo-trajectory consistency mismatch risk of the current system is marked as low pseudo-trajectory consistency mismatch risk.

[0011] In a preferred embodiment, if the pseudo-trajectory consistency mismatch risk of the system is marked as high pseudo-trajectory consistency mismatch risk, a confidence risk feedback adjustment function is constructed based on the pseudo-trajectory consistency mismatch assessment index output by the pseudo-trajectory consistency mismatch assessment model: ,in This is the corrected match confidence score. The initial matching confidence level is dynamically adjusted.

[0012] The technical effects and advantages of this invention are as follows: 1. This invention introduces a trajectory overlap risk coefficient to keenly capture the aggregation and intersection behavior of trajectories within a sliding window. Its value directly reflects the degree of structural confusion caused by the "apparent consistency" between trajectories. Secondly, the modal feature degradation perception coefficient dynamically measures the degree of feature degradation caused by occlusion, interference, or sensor failure of the multimodal information (such as attitude spectrum, infrared texture, radar echo, etc.) carried by each trajectory from two dimensions: temporal stability and spectral distribution. Through the joint evaluation of these two risk factors, confidence suppression and modal enhancement mechanisms can be proactively triggered when the system enters a high pseudo-consistency mismatch risk zone, avoiding the disruption of subsequent clustering and scheduling path deduction caused by false trajectory merging. It can also maintain the high efficiency and stability of the matching algorithm under low-risk conditions, thereby achieving refined control over the risk of misjudgment in the trajectory recognition link and significantly improving the accuracy, robustness, and system response capability of multi-target trajectory association in complex environments. Attached Figure Description

[0013] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings; Figure 1 This is a flowchart of a method according to an embodiment of the present invention. Detailed Implementation

[0014] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0015] Example: Figure 1 This invention presents a satellite spatiotemporal trajectory data correlation analysis method based on PNTA, comprising the following steps: Step S1: Collect multimodal information uploaded from different satellite platforms to construct a trajectory state vector sequence and predict the future state of the trajectory; Step S2: Calculate the initial matching confidence score by performing spatial similarity matching on any two predicted trajectories within the sliding window; Step S3: Obtain trajectory intersection and overlap information between trajectories within the sliding window and calculate the trajectory intersection and overlap risk coefficient; Step S4: Obtain modal feature degradation information between trajectories within the sliding window and calculate the modal feature degradation sensing coefficient; Step S5: Construct a pseudo-trajectory consistency mismatch assessment model based on the trajectory crossover risk coefficient and modal feature degradation perception coefficient, output the pseudo-trajectory consistency mismatch assessment index, and assess the pseudo-trajectory consistency mismatch risk of the current system. Step S6: Dynamically adjust the initial matching confidence based on the current system's pseudo-trajectory consistency mismatch risk; In step S1, multimodal observation information from different satellite platforms is collected and fused, including trajectory state feature data such as spatial position, velocity vector, attitude spectrum, acceleration spectrum, and inter-satellite link status. For the collected multimodal observation information, a unified time reference (e.g., GPS time → UTC) is established, and interpolation alignment and sampling normalization are performed. Filters (e.g., Kalman or wavelet filters) are applied to remove jump points and missing segments, constructing a trajectory state vector sequence. ,in For each trajectory object, the trajectory state feature data vector is... A temporal trajectory representation module with a time attention mechanism is constructed based on the trajectory state vector sequence: ,in This is the hidden state vector of the trajectory, used as a starting point to continue deducing the future trend of the trajectory in the subsequent decoding process. A recurrent neural network (RNN) structure for processing time series data. For trajectory encoders, the output sequence of potential trajectory states is: ; For perturbation data Constructing embedded representations: ,in For the perturbation embedding vector, For perturbation embedding networks, future time-step states are generated using historical trajectory states and perturbation embeddings to form predicted trajectory sequences: ,in The decoded hidden state vector at the prediction time. For the historical context of the trajectory, For trajectory decoder, This is the trajectory state feature data vector at the predicted time. For trajectory state output network, To predict the time step length; ultimately, each trajectory object Future predicted trajectory Represented as: This allows for the prediction of the future state of the trajectory. It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. In step S2, within the sliding window, for any two predicted trajectories... Perform spatial similarity matching: ,in To predict trajectory pairs Spatial similarity score, They represent the predicted trajectories respectively. and Trajectory state feature data vector, The Euclidean distance is the distance between the trajectory state feature data vectors. The angle between the directions of the two predicted trajectories. , They represent the predicted trajectories respectively. and The trajectory direction vector, , These represent the preset scaling factors for Euclidean distance and the included angle, respectively. , All are greater than 0; It should be noted that, , The settings should be tailored to the specific circumstances. For example, an expert-empowered approach could be adopted, where experts in relevant fields are invited to determine the pre-defined proportions for each indicator through professional opinion surveys and comprehensive evaluations. , It can be 0.5 or 0.5; Calculate the initial matching confidence. : ; It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. Step S3: Obtain trajectory intersection and overlap information between trajectories within the sliding window and calculate the trajectory intersection and overlap risk coefficient; In this invention, the trajectory overlap risk coefficient is used to quantify the degree to which different target satellite trajectories highly overlap, exhibit similar paths, or have consistent motion directions in the spatiotemporal dimension. Its purpose is to identify potential trajectory matching confusion risks. A high trajectory overlap risk coefficient indicates that multiple trajectories within the region have experienced significant spatial overlap or convergence of motion characteristics within a short period. This may manifest as "apparent consistency" in the predicted trajectories of different satellites across multiple time steps, significantly increasing the probability of incorrect matching in trajectory association judgment. Especially when the target exhibits modal feature degradation or lacks attitude data, the impact of overlap risk is further amplified, causing the system to mistakenly treat multiple different targets as the same object for matching and clustering. Conversely, a low trajectory overlap risk coefficient indicates that the trajectories exhibit good distinguishability in terms of spatial location, motion direction, and temporal trends. Their predicted paths maintain stable distance and angular separation, and their disturbance responses and attitude change patterns are relatively independent. This allows the system to make accurate trajectory attribution judgments based on nonlinear state evolution trends, greatly reducing the risk of incorrect matching and improving the system's trajectory recognition accuracy and the stability of the downstream decision chain. The beneficial effects of this mechanism are reflected in several aspects: First, it can effectively alleviate matching ambiguities caused by similar orbital altitudes or dense deployment of multiple targets in complex constellation operating environments; second, by linking evaluation with the modal feature degradation perception coefficient, it can further identify the risk of "pseudo-trajectory consistency mismatch" and provide early warning of potential target misjudgment or trajectory fusion errors in the system. In summary, this invention, by introducing a trajectory intersection and overlap risk coefficient, effectively compensates for the problems of ambiguous judgment criteria and insufficient handling of dynamic feature degradation in traditional trajectory association methods in high-density trajectory scenarios.

[0016] The logic for obtaining the trajectory intersection and overlap risk coefficient is as follows: For any pair of trajectories Construct its temporal compression difference vector : ,in As a unit of time, For predicting trajectories At the predicted time Trajectory state feature data vector, For predicting trajectories At the predicted time Trajectory state feature data vector, To predict the time step length, It refers to the L2 norm distance between trajectory state feature data vectors. for The square of; At the preset distance threshold The following conditions are used to extract low-bifaction segments from continuous temporal sequences: The set of continuous time series low-branch segments that meet the extraction criteria is marked as: ; Calculate the cross-response density coefficient This is used to measure the "density" of two trajectories simultaneously maintaining high similarity / intersection within the perturbation state space over a future period of time. Its calculation is expressed as follows: ,in This represents the number of time steps in the set of low-bifaction segments of a continuous time series that satisfy the extraction criteria. The temporal resolution for trajectory prediction, i.e., the time resolution between two adjacent points. The actual time interval; Construct a convergent attraction function and output the attraction potential field value between the two trajectories. : ,in The angle between the directions of the two predicted trajectories; The cross-response density coefficient and the attraction potential field value are linked to calculate the trajectory intersection and overlap risk coefficient. : ; It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. Step S4: Obtain modal feature degradation information between trajectories within the sliding window and calculate the modal feature degradation sensing coefficient; In this invention, the modal feature degradation perception coefficient is used to measure the comprehensive degree of feature information degradation or occlusion anomalies of the target satellite in multimodal observation data. It reflects the impact of the incompleteness of the observation dimensions and the decline in the reliability of dynamic features on the matching accuracy of the trajectory recognition system. When the coefficient value is large, it indicates that the target has significant problems such as missing modal information, sensor limitations, feature value degradation, or interference occlusion in the observation window, such as blurred visible light trajectory, radar breakpoints, and decreased signal-to-noise ratio of attitude spectrum. This makes it impossible for the system to extract discriminative trajectory features from multiple modal dimensions, resulting in a weakened ability to distinguish the target's identity and thus exacerbating the risk of trajectory confusion and false matching. When the modal feature degradation perception coefficient value is small, it indicates that the current trajectory has stable, clear, and complete observation features in multimodal data. The signal-to-noise ratio of each dimension feature (such as acceleration spectrum, attitude spectrum, infrared profile, micro-dynamic displacement, etc.) is good and the time sequence is continuous. The system can effectively establish a high-dimensional discriminative representation from it, which helps to accurately distinguish similar trajectories and improve matching robustness. By introducing a modal feature degradation perception coefficient, the system can dynamically assess the reliability of data in trajectory association, directly mapping changes in observation quality to a pre-factor of potential recognition errors. Compared to traditional matching mechanisms that rely solely on spatial location or orientation, this coefficient introduces perceptual reliability into the decision factor, making trajectory recognition more closely reflect the observation complexity under actual operational conditions. In conjunction with the trajectory overlap risk coefficient, the modal feature degradation perception coefficient can further improve the accuracy of assessing the risk of "pseudo-trajectory consistency mismatch." In trajectory pairs converging in spatial location or orientation, if the modal features of a target degrade due to occlusion, interference, or loss, the system is highly susceptible to incorrect pairing based on the remaining fuzzy features. Therefore, the modal feature degradation perception coefficient, as a quantification of the degree of feature dimension compression, provides the system with an early warning capability for the coupling of "data degradation + converging paths," effectively assisting the PNTA association algorithm in avoiding multiple matching ambiguities and false trajectory fusion problems caused by decreased perception capabilities in complex trajectory recognition scenarios, thereby improving the overall trajectory discrimination accuracy and the robustness of spatiotemporal consistency matching.

[0017] The logic for obtaining the modal feature degradation perception coefficient is as follows: For each trajectory Extract the predicted trajectory state feature data vector time series within the sliding window: Manifold learning algorithms (such as T-SNE, Isomap, UMAP) are used to map high-dimensional modal features to low-dimensional embedding subspaces. , ,in This is the low-dimensional embedded trajectory state feature data vector representation form. For embedding functions (such as T-SNE, Isomap, UMAP), Indicates the Kth row after embedding, The column feature matrix, where each row is a time-step embedding vector, has a dimension of . ; For each embedded subspace, calculate its state mutation rate sequence along the time dimension: , ,in The mutation rate of the state. As a unit of time, To predict the time step length; extract the extreme value gradient fluctuation ratio. : ,in This is to prevent division by zero by a very small constant (generally taken as...). ); For the original trajectory state feature data vector time series Perform wavelet packet transform (WPT) to obtain its time-spectrum packet distribution in multiple frequency bands: , ,in This is the time-frequency representation of the trajectory state feature data vector. After wavelet packet transform OK, Column time-frequency characteristic matrix, Wavelet packet transform; Calculate the trajectory state characteristic instability distribution value based on the temporal spectrum packet distribution. : , ,in This represents all elements in the h-th column of the time-frequency characteristic matrix, i.e., a column vector of length K. Indicates all rows, This refers to the index number of a column in the time-frequency feature matrix. This represents the L2 norm distance between vectors; Construct a modal feature degradation mapping function and output the modal feature degradation mapping values. : ; Calculate trajectory pairs Modal feature degradation perception coefficient : ,in For trajectory Modal feature degradation mapping value, For trajectory Modal feature degradation mapping value; It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. Step S5: Construct a pseudo-trajectory consistency mismatch assessment model based on the trajectory crossover risk coefficient and modal feature degradation perception coefficient, output the pseudo-trajectory consistency mismatch assessment index, and assess the pseudo-trajectory consistency mismatch risk of the current system. A pseudo-trajectory consistency mismatch assessment model is constructed based on the trajectory intersection and overlap risk coefficient and the modal feature degradation perception coefficient, and the pseudo-trajectory consistency mismatch assessment index is output. The formula used in the pseudo-trajectory consistency mismatch assessment model is as follows: In the formula This is a pseudo-trajectory consistency mismatch assessment index. The risk coefficient for trajectory intersection and overlap. The modal feature degradation perception coefficient, These represent the preset proportional coefficients for the trajectory intersection and overlap risk coefficient and the modal feature degradation perception coefficient, respectively. All are greater than 0; It should be noted that the above formulas are all dimensionless calculations. Commonly used methods for removing dimensions include Min-Max normalization and Z-Score standardization, which will not be elaborated here. The settings should be tailored to the specific circumstances. For example, an expert-empowered approach could be adopted, where experts in relevant fields are invited to determine the pre-defined proportions for each indicator through professional opinion surveys and comprehensive evaluations. It can be 0.5 or 0.5; As shown in the above calculation expressions, the larger the trajectory overlap risk coefficient and the larger the modal feature degradation perception coefficient, the larger the pseudo-trajectory consistency mismatch evaluation index. This indicates that the target trajectories in the current system have a stronger tendency to overlap or fit together in spatial paths. At the same time, the modal information carried by each trajectory shows significant degradation, causing the traditional reliance on multimodal data such as physical position, velocity direction, and attitude features for trajectory identity determination to lose reliability. As a result, multiple trajectories that originally belong to different targets exhibit misleading structural features of "pseudo-consistency" from the model's perspective, leading to high-confidence mismatches and false trajectory fusion. Conversely, the smaller the trajectory overlap risk coefficient, the higher the risk coefficient. The smaller the modal feature degradation perception coefficient, the smaller the pseudo-trajectory consistency mismatch evaluation index, indicating that the target trajectories currently being processed by the system have good spatial structure discrimination, and there is no significant spatial intersection, path fitting or dynamic overlap behavior between the trajectories. At the same time, the multimodal features carried by each target trajectory (such as acceleration spectrum, attitude spectrum, thermal imaging features or spectral features, etc.) maintain high expression integrity and recognition discrimination in terms of temporal structure, spectral stability and directional consistency. This enables the PNTA method to extract effective target discrimination indexes in the nonlinear dynamic feature space when performing trajectory association analysis, thereby achieving high confidence and high accuracy trajectory matching. The pseudo-trajectory consistency mismatch assessment index is compared with the preset pseudo-trajectory consistency mismatch assessment index threshold to assess the pseudo-trajectory consistency mismatch risk of the current system, as follows: If the pseudo-trajectory consistency mismatch evaluation index is greater than the pseudo-trajectory consistency mismatch evaluation index threshold, it indicates that there is significant trajectory overlap and severe degradation of multi-target modal features in the current system. This leads to a high risk of structural confusion and feature ambiguity in the PNTA model during trajectory matching judgment, making it highly susceptible to triggering high-risk error behaviors such as trajectory identity misjudgment, association shift, or false fusion. At this time, the distinguishability between trajectory states is weakened, and the system exhibits strong pseudo-consistency in both the spatiotemporal and feature dimensions, indicating that the recognition mechanism is at a critical state of high false association. The pseudo-trajectory consistency mismatch risk of the current system is marked as high pseudo-trajectory consistency mismatch risk. If the pseudo-trajectory consistency mismatch assessment index is less than or equal to the pseudo-trajectory consistency mismatch assessment index threshold, it indicates that the current trajectory set still maintains good distinguishability in terms of spatial path layout and modal feature integrity, there is no serious overlap between the target trajectories, and the modal features have not degraded significantly. The PNTA method can still rely on nonlinear feature dynamics and structural trends to perform effective distinction and high-confidence matching. In this state, the system's identification and association of target trajectories is stable and reliable, the identification link is in a low-noise, low-interference normal operation mode, and the pseudo-trajectory consistency mismatch risk of the current system is marked as low pseudo-trajectory consistency mismatch risk. It should be noted that the threshold for the pseudo-trajectory consistency mismatch assessment index can be flexibly set according to specific application needs and task scenarios, such as trajectory complexity, recognition accuracy requirements, modal perception capabilities, and the dynamic nature of the orbital environment. It is not unique or fixed. This threshold is mainly used to qualitatively judge and classify the "degree of potential mismatch risk" during the automatic trajectory data association process, and is a key judgment threshold in the system's risk control mechanism.

[0018] In practical engineering applications, the threshold for the pseudo-trajectory consistency mismatch evaluation index can be configured in the following ways: Empirical statistical method: Run the PNTA model on a historical trajectory recognition dataset to statistically analyze the distribution range of the pseudo-mismatch index for various trajectory combinations, and use a quantile value (such as the 95% confidence upper bound) as an initial threshold estimate. For example, in a typical low-Earth orbit constellation recognition task, the statistically derived reasonable threshold is 0.62~0.74.

[0019] Task accuracy-driven approach: When the system has a very low tolerance for mismatches in the identification task (such as reconnaissance satellites, multi-satellite synchronous mission coordination, etc.), a lower threshold should be set (such as 0.45 or 0.55); while in observation tasks with a strong ability to identify trajectory redundancy (such as meteorological satellites, resource remote sensing satellites, etc.), the threshold setting can be relaxed (such as 0.75 or 0.85).

[0020] Expert-empowered approach: This method involves evaluating and judging the system using expert systems or mission planners, and then comprehensively scoring it based on multiple indicators such as the complexity of the system's operating scenario, the probability of modal degradation, and the frequency of trajectory intersections to generate an optimal risk threshold. For example, in a high-density satellite constellation formation identification scenario, experts might recommend setting the threshold to 0.60 to ensure that the system can identify potential false fusion risks in advance.

[0021] Adaptive parameter tuning method: In trajectory recognition systems with online learning and dynamic feedback capabilities, the threshold can be dynamically adjusted by considering the historical false judgment rate, target recognition accuracy, and modal signal-to-noise ratio trends, thereby achieving real-time optimization of the pseudo-trajectory consistency mismatch evaluation threshold.

[0022] Step S6: Dynamically adjust the initial matching confidence based on the current system's pseudo-trajectory consistency mismatch risk; If the pseudo-trajectory consistency mismatch risk of the system is marked as high pseudo-trajectory consistency mismatch risk, a confidence risk feedback adjustment function is constructed based on the pseudo-trajectory consistency mismatch assessment index output by the pseudo-trajectory consistency mismatch assessment model: ,in This is the corrected match confidence score. The initial match confidence level is dynamically adjusted. This invention introduces a trajectory overlap risk coefficient to keenly capture the aggregation and intersection behavior of trajectories within a sliding window. Its value directly reflects the degree of structural confusion caused by the "apparent consistency" between trajectories. Secondly, the modal feature degradation perception coefficient dynamically measures the degree of feature degradation caused by occlusion, interference, or sensor failure of the multimodal information (such as attitude spectrum, infrared texture, radar echo, etc.) carried by each trajectory from two dimensions: temporal stability and spectral distribution. Through the joint evaluation of these two risk factors, confidence suppression and modal enhancement mechanisms can be proactively triggered when the system enters a high pseudo-consistency mismatch risk zone, avoiding the disruption of subsequent clustering and scheduling path deduction caused by false trajectory merging. It can also maintain the high efficiency and stability of the matching algorithm under low-risk conditions, thereby achieving refined control over the risk of misjudgment in the trajectory recognition link and significantly improving the accuracy, robustness, and system response capability of multi-target trajectory association in complex environments.

[0023] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0024] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0025] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A satellite spatiotemporal trajectory data correlation analysis method based on PNTA, characterized in that: Includes the following steps: Step S1: Collect multimodal information uploaded from different satellite platforms to construct a trajectory state vector sequence and predict the future state of the trajectory; Step S2: Calculate the initial matching confidence score by performing spatial similarity matching on any two predicted trajectories within the sliding window; Step S3: Obtain trajectory intersection and overlap information between trajectories within the sliding window and calculate the trajectory intersection and overlap risk coefficient; Step S4: Obtain modal feature degradation information between trajectories within the sliding window and calculate the modal feature degradation sensing coefficient; Step S5: Construct a pseudo-trajectory consistency mismatch assessment model based on the trajectory crossover risk coefficient and modal feature degradation perception coefficient, output the pseudo-trajectory consistency mismatch assessment index, and assess the pseudo-trajectory consistency mismatch risk of the current system. Step S6: Dynamically adjust the initial matching confidence based on the pseudo-trajectory consistency mismatch risk of the current system.

2. The satellite spatiotemporal trajectory data correlation analysis method based on PNTA according to claim 1, characterized in that: In step S1, multimodal observation information from different satellite platforms is collected and fused. For the collected multimodal observation information, a unified time reference is established, and interpolation alignment and sampling normalization are performed. Filters are applied to remove jump points and missing segments, and a trajectory state vector sequence is constructed for each trajectory object. A temporal trajectory representation module with a time attention mechanism is constructed based on the trajectory state vector sequence, and the output trajectory potential state representation sequence is generated. An embedding representation is constructed for the perturbation data. The future time step state is generated by using the trajectory historical state and the perturbation embedding to form a predicted trajectory sequence, thus completing the prediction of the future state of the trajectory.

3. The satellite spatiotemporal trajectory data correlation analysis method based on PNTA according to claim 1, characterized in that: The logic for obtaining the trajectory intersection and overlap risk coefficient is as follows: For any pair of trajectories ,in For the i-th predicted trajectory, For the j-th predicted trajectory, construct its temporal compressed difference vector. ; At the preset distance threshold The following conditions are used to extract low-bifaction segments from continuous temporal sequences: The set of continuous time series low-branch segments that meet the extraction criteria is marked as: ,in As a unit of time, For predicting trajectories At the predicted time Trajectory state feature data vector, For predicting trajectories At the predicted time Trajectory state feature data vector; calculate cross response density coefficient Its calculation is expressed as follows: ,in This represents the number of time steps in the set of low-bifaction segments of a continuous time series that satisfy the extraction criteria. The temporal resolution for trajectory prediction, i.e., the time resolution between two adjacent points. The actual time interval, To predict the time step length; Construct a convergent attraction function and output the attraction potential field value between the two trajectories. : ,in The angle between the directions of the two predicted trajectories; The cross-response density coefficient and the attraction potential field value are linked to calculate the trajectory intersection and overlap risk coefficient. : .

4. The satellite spatiotemporal trajectory data correlation analysis method based on PNTA according to claim 1, characterized in that: The logic for obtaining the modal feature degradation perception coefficient is as follows: For each trajectory Within a sliding window, the predicted trajectory state feature data vector time series is extracted, and a manifold learning algorithm is used to map the high-dimensional modal features to a low-dimensional embedding subspace. , ,in This is the low-dimensional embedded trajectory state feature data vector representation form. This represents the embedded feature matrix. For dimensions; For each embedded subspace, calculate its state mutation rate sequence along the time dimension: , ,in The mutation rate of the state. As a unit of time, To predict the time step length; extract the extreme value gradient fluctuation ratio. : ,in This is a very small constant to prevent division by zero; For the original trajectory state feature data vector time series Perform wavelet packet transform to obtain its time-spectrum packet distribution in multiple frequency bands: , ,in This is the time-frequency representation of the trajectory state feature data vector. After wavelet packet transform OK, Column time-frequency characteristic matrix, Wavelet packet transform; Calculate the trajectory state characteristic instability distribution value based on the temporal spectrum packet distribution. : , ,in This represents all elements in the h-th column of the time-frequency characteristic matrix. Indicates all rows, This refers to the index number of a column in the time-frequency feature matrix. This represents the L2 norm distance between vectors; Construct a modal feature degradation mapping function and output the modal feature degradation mapping values. : ; Calculate trajectory pairs Modal feature degradation perception coefficient : ,in For trajectory Modal feature degradation mapping value, For trajectory The modal feature degradation mapping value.

5. The satellite spatiotemporal trajectory data correlation analysis method based on PNTA according to claim 1, characterized in that: A pseudo-trajectory consistency mismatch assessment model is constructed based on the trajectory intersection and overlap risk coefficient and the modal feature degradation perception coefficient, and the pseudo-trajectory consistency mismatch assessment index is output. The formula used in the pseudo-trajectory consistency mismatch assessment model is as follows: In the formula This is a pseudo-trajectory consistency mismatch assessment index. The risk coefficient for trajectory intersection and overlap. The modal feature degradation perception coefficient, These represent the preset proportional coefficients for the trajectory intersection and overlap risk coefficient and the modal feature degradation perception coefficient, respectively. All are greater than 0.

6. The satellite spatiotemporal trajectory data correlation analysis method based on PNTA according to claim 5, characterized in that: The pseudo-trajectory consistency mismatch assessment index is compared with the preset pseudo-trajectory consistency mismatch assessment index threshold to assess the pseudo-trajectory consistency mismatch risk of the current system, as follows: If the pseudo-trajectory consistency mismatch assessment index is greater than the pseudo-trajectory consistency mismatch assessment index threshold, then the pseudo-trajectory consistency mismatch risk of the current system is marked as high pseudo-trajectory consistency mismatch risk. If the pseudo-trajectory consistency mismatch assessment index is less than or equal to the pseudo-trajectory consistency mismatch assessment index threshold, then the pseudo-trajectory consistency mismatch risk of the current system is marked as low pseudo-trajectory consistency mismatch risk.

7. The satellite spatiotemporal trajectory data correlation analysis method based on PNTA according to claim 6, characterized in that: If the pseudo-trajectory consistency mismatch risk of the system is marked as high pseudo-trajectory consistency mismatch risk, a confidence risk feedback adjustment function is constructed based on the pseudo-trajectory consistency mismatch assessment index output by the pseudo-trajectory consistency mismatch assessment model: ,in This is the corrected match confidence score. The initial matching confidence level is dynamically adjusted.

Images