Comprehensive capability evaluation method for monitoring system for space target monitoring

Abstract

The invention discloses a comprehensive capability evaluation method for a monitoring system for space target monitoring, and relates to the technical field of space target monitoring, and the method combines a physical model and a long-short term memory (LSTM) network deep learning model, captures the nonlinear change of an orbit, effectively deals with the rapid change and abrupt change of the orbit, and improves the comprehensive capability of the monitoring system. High-precision prediction is still kept in a complex dynamic orbit environment, LSTM and extended Kalman filtering EKF are introduced, prediction can be performed based on historical orbit data, real-time observation data can be dynamically corrected, the response of the system to target orbit change is more timely, the self-adaptability of the system in the face of orbit change is further optimized through reinforcement learning, and the prediction accuracy of the system is improved. Compared with the prior art, the method has the advantages of reducing the response time, improving the real-time performance, dynamically adjusting the tracking strategy through reinforcement learning, and adopting an interference processing mechanism, effectively solving the problem of target trajectory overlapping or dynamic interference, and ensuring that the target is not lost or mistakenly tracked even in a high-density and complex-interference environment.

Description

Technical Field

[0001] The present invention relates to the technical field of space target monitoring, in particular to a method for evaluating the comprehensive capability of a monitoring system for space target monitoring. Background Art

[0002] Space target monitoring systems are responsible for monitoring and tracking a variety of high-speed moving space objects, such as satellites, space debris, and other spacecraft, in environments such as low-Earth orbit, medium-Earth orbit, and deep space. The trajectories of these objects are complex and dynamic. The monitoring system needs to have efficient real-time tracking capabilities, multi-sensor information fusion capabilities, and task processing capabilities to cope with multi-target scenarios. Accurate assessment of the comprehensive capabilities of the space target monitoring system directly affects the effectiveness of the system in actual missions.

[0003] Most targets in space are in orbit and are affected by the earth's gravity and solar radiation pressure. The orbital changes are nonlinear and difficult to predict. The current spatiotemporal dynamic evaluation method uses a two-body or three-body problem model to approximate the orbital changes, and uses linear interpolation to evaluate the dynamic response capability of the system. It relies on preset orbits and static monitoring models to evaluate the detection coverage and response time of the system. The change of the orbit will cause a large deviation between the static evaluation model and the actual monitoring scenario, and it is impossible to accurately evaluate the ability of the system in the face of real-time dynamic changes. In nonlinear orbital dynamics, the static model is difficult to capture target mutations or rapid accelerations. The evaluation results lack accuracy in adaptability to the system's real-time tracking. In addition, Most space monitoring targets consist of more than one, and the targets generally move on different orbits, and there may even be overlapping and interfering trajectories. Current evaluation methods usually use joint probabilistic data association (JPDA) or multiple hypothesis tracking (MHT) methods to track each target independently and use target filtering mechanisms to deal with trajectory confusion. However, in target-dense areas, the above evaluation methods cannot effectively deal with dynamic interference between targets. When dealing with multiple targets, target loss or mistracking may occur. This problem is more significant when the target trajectories are close to or overlap, making it difficult to accurately evaluate the actual capabilities of the system. Therefore, a comprehensive capability evaluation method for space target monitoring systems is urgently needed to solve such problems. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a comprehensive capability evaluation method for a monitoring system for space target monitoring to solve the problems of the current space target monitoring system evaluation method in spatiotemporal dynamic evaluation and multi-target tracking, such as simple models, overly ideal scenario assumptions, and difficulty in handling dynamic changes.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] The present invention provides a method for evaluating the comprehensive capability of a monitoring system for space target monitoring, which includes:

[0008] Step S1: Collaborative collection and fusion of multi-source data.

[0009] Integrate data sources, obtain initial monitoring data from the space monitoring system, fuse data from different monitoring sources, and output high-precision, temporally and spatially consistent target monitoring information, namely fused data;

[0010] Step S2, spatiotemporal dynamic evaluation,

[0011] Predict nonlinear track changes, perform dynamic evaluation based on the fusion data output in step S1, and output track prediction data;

[0012] Step S3, multi-target tracking and interference processing,

[0013] Perform multi-target identification and tracking. Use the orbit prediction data in step S2 to detect and track multiple targets in real time. The real-time detection and tracking methods include:

[0014] Perform dynamic trajectory tracking optimization, adopt reinforcement learning methods, dynamically adjust tracking strategies, and continuously optimize tracking paths and strategies through continuous environmental interaction, thereby reducing computational overhead and improving the real-time performance and accuracy of target tracking.

[0015] Multi-target dynamic interference processing, combining trajectory prediction and reinforcement learning, can effectively deal with dynamic interference and trajectory overlap of multiple targets, avoiding target loss or mistracking;

[0016] Step S4, system comprehensive capability assessment,

[0017] Perform dimensional evaluation, based on the tracking results of step S3, and comprehensively evaluate the performance of the system under multi-target and dynamically changing conditions.

[0018] Furthermore, the monitoring data sources include those from space monitoring systems,

[0019] Space monitoring systems, namely, target monitoring data obtained by space telescopes, space radars, space debris monitoring satellites and the Global Positioning System (GNSS).

[0020] Space-based monitoring system, namely the high-orbit target monitoring data obtained by geosynchronous orbit GEO monitoring satellites and infrared monitoring satellites,

[0021] And ground-based monitoring systems, namely data obtained by large ground-based optical telescopes, laser ranging systems and ground-based radar systems for long-range target monitoring.

[0022] Furthermore, in step S1, multi-source data from space telescopes, space radars, space debris monitoring satellites, global positioning system (GNSS), geosynchronous orbit (GEO) monitoring satellites, infrared monitoring satellites, large ground-based optical telescopes, laser ranging systems, and ground-based radar systems are fused, and the extended Kalman filter (EKF) is used as the core framework to output target state estimation, i.e., fused data.

[0023] Define the state vector, the target state vector x k Represents the spatial position and velocity at time k:

[0024] Among them, p x,k ,p y,k ,p z,k represents the three-dimensional spatial coordinates of the target at time k, v x,k ,v y,k ,v z,k represents the velocity component of the target at time k;

[0025] State transition equation: x k+1 =f(x k )+w k , where f(x k ) is a nonlinear state transfer function, combining the effects of gravitational perturbation and solar radiation pressure on the target orbit, w k is process noise, which obeys Gaussian distribution where Q k is the process noise covariance matrix;

[0026] The nonlinear state transfer function f(x k ) consists of:

[0027] Gravitational perturbation is mainly caused by the non-uniform gravitational field of the earth. The gravitational perturbation model has an acceleration a gravity Expressed as:

[0028] Where μ is the Earth's gravitational constant, is the distance between the target and the center of the earth, J2 is the second-order perturbation term of the earth's gravitational field, R e is the radius of the Earth, is the geocentric latitude of the target;

[0029] Solar radiation pressure model, its acceleration a srp Expressed as:

[0030] Among them, P srp is the solar radiation pressure constant, A is the cross-sectional area of ​​the target, C ris the reflection coefficient of the target, m is the mass of the target, is the unit direction vector pointing to the sun;

[0031] Total acceleration a k From the superposition of gravitational perturbation and solar radiation pressure, we can get:

[0032] a k =a gravity +a srp ;

[0033] Based on the current state x k and the total acceleration a k , the state update equation of the target at the next moment is expressed as:

[0034] Where Δt is the time step, a x,k ,a y,k ,a z,k is the total acceleration a k , combining the gravitational perturbation and solar radiation pressure.

[0035] Furthermore, the target state estimation method output in step S1 further includes:

[0036] Calculate the Jacobian matrix F of the state transfer equation k , the Jacobian matrix F k By the state vector x k Taking partial derivatives, we get Jacobian matrix F k Used to linearize the state transfer equation;

[0037] According to the state transition equation, predict the state at the next moment in, represents the predicted state vector at time k+1, represents the state transition function;

[0038] According to the Jacobian matrix F k and process noise Q k , the covariance matrix of the predicted state estimate:

[0039] Among them, P k+1|k represents the covariance matrix of the predicted state, F k The Jacobian matrix representing the state transition equation is defined as P k Represents the state covariance matrix at the current moment, Q k represents the process noise covariance matrix, T represents the transpose operation;

[0040] Calculate the Kalman gain, Kalman gain Kk+1 The calculation formula is:

[0041] Among them, K k+1 represents the Kalman gain, H k+1 The Jacobian matrix representing the observation equation is defined as R k+1 represents the observation noise covariance matrix;

[0042] Combined observations z k+1 Update the state estimate with the predicted value:

[0043] in, represents the updated state estimate,

[0044] z k+1 represents the observation data, Represents a nonlinear observation model that transforms the predicted state into the observation space;

[0045] Update the covariance matrix, P k+1 =(IK k+1 H k+1 )P k+1|k , where P k+1 represents the updated covariance matrix, and I represents the identity matrix.

[0046] Furthermore, in step S1, the target state is predicted and updated by the prediction phase and update of the extended Kalman filter to generate a state estimate, where the target state estimate includes the three-dimensional position and velocity information of the target;

[0047] In step S2, the target trajectory change is predicted based on the fused data, i.e., the output of step S1.

[0048] Furthermore, the track nonlinear change prediction step in step S2 includes:

[0049] Based on the fusion data of step S1, that is, the current state estimation, a nonlinear orbit prediction model is used to capture the target orbit change and predict the future orbit state. Assume that at time k, the historical state data is {x k-n ,x k-n+1 ,…,x k}, n is the window length of historical data, the historical data is input into the long short-term memory network LSTM to predict nonlinear track changes, and the historical state data is recorded as:

[0050] Represents the historical trajectory state sequence of the target, including the state information of the past n moments;

[0051] Use LSTM for track prediction through historical track data Use the LSTM network to learn the nonlinear changes of the trajectory and output the predicted state of the target at the next moment k+1

[0052] in, is the orbital state of the target at time k+1;

[0053] Perform dynamic corrections, combine LSTM predictions with gravitational perturbations and solar radiation pressure models, and perform weighted averaging of orbital predictions:

[0054] in, Represents the corrected orbit prediction data, combining the predictions of the physical model and the LSTM model, represents the track state predicted by the LSTM model, represents the state predicted based on the physical model, α represents the balance coefficient, and orbit prediction data As the output of step S2;

[0055] In step S2, nonlinear orbit prediction is performed using an LSTM model based on historical orbit data, and dynamic correction is performed in combination with the physical model to output orbit prediction data, effectively responding to nonlinear changes in the orbit and improving the system's prediction capability in a dynamic orbit environment.

[0056] Furthermore, step S3 uses reinforcement learning to optimize the target tracking strategy and handle dynamic interference between multiple targets based on the trajectory prediction data in step S2:

[0057] Suppose there are N targets, and the state of each target consists of a position and a velocity vector. For the trajectory prediction data x of the i-th target at time t, i (t) is expressed as:

[0058] Among them, p x,i (t),p y,i (t),p z,i (t) represents the three-dimensional spatial position of target i at time t, v x,i (t),v y,i (t),v z,i (t) represents the velocity component of target i at time t;

[0059] The total tracking error objective function is expressed as:

[0060] in, represents the total tracking error function, x i (t) represents the true trajectory state of the i-th target at time t, represents the predicted trajectory state of the i-th target at time t, |·| represents the Euclidean norm;

[0061] Define state, action and reward function to optimize multi-target tracking strategy, state s t Represents the state of the system, consisting of the current orbit information of all targets and observational data Composition, expressed as:

[0062] Among them, x i (t) represents the orbit prediction data of target i, z i (t) represents the sensor observation data of target i, which comes from multi-source monitoring;

[0063] Action a t Indicates the adjustment of the tracking path of each target. The adjusted path is:

[0064] in, represents the updated trajectory prediction of target i at time t+1, Δx i (t) represents the adjustment of trajectory prediction based on reinforcement learning strategy;

[0065] Reward function r t To minimize the tracking error, it is defined as the negative of the error:

[0066] Among them, r t represents the reward function, which encourages reducing the total tracking error, represents the total tracking error at the current time t;

[0067] Through the policy network π(s t ,a t ) determines the best action in a given state, minimizes the tracking error, and uses the policy gradient algorithm to update the policy network. The policy update rule is:

[0068] in, Represents the gradient of the policy, based on the current policy parameters θ, π θ (s t ,a t ) represents the policy network, that is, given the state s t Next take action a t The probability of r t Represents the reward function, which is used to measure the effect of the current action;

[0069] At each time step, reinforcement learning adjusts the policy parameters θ according to the feedback from the environment, i.e., the decrease or increase in the trajectory tracking error, to optimize the tracking strategy, reduce computational overhead and improve tracking accuracy.

[0070] Furthermore, in step S3, the multi-target dynamic interference processing method is:

[0071] When the trajectories of multiple targets overlap or there is dynamic interference, processing is performed to avoid target loss or mistracking. The interference measurement function D(t) is defined to measure the degree of trajectory overlap or interference between target i and target j:

[0072] Among them, D i,j (t) represents the interference measure between target i and target j at time t, |x i (t)-x j (t)| represents the difference between the trajectories of target i and target j, using the Euclidean distance metric, and σ represents the distance scale parameter to control interference;

[0073] When D i,j When (t) is close to 1, it means that the target trajectories are highly overlapped or interfered with. At this time, reinforcement learning adjusts the tracking strategy to separate the tracking paths of the two targets, reducing the risk of mistracking and loss.

[0074] When interference occurs on the target trajectory, reinforcement learning adds negative rewards for interference processing and adjusts the strategy to separate the target trajectory. The adjusted reward function is:

[0075] Among them, λ represents the weight parameter, which controls the impact of the interference measure on the total reward. represents the sum of interference measures among all targets;

[0076] Output multi-target tracking results, that is, trajectory prediction results after dynamic adjustment and interference processing by reinforcement learning: Finally, the tracking results of multiple targets at time t+1 are output

[0077] Dynamic trajectory tracking and optimization of multiple targets are performed through reinforcement learning, and the interference measurement function D is defined. i,j (t) It handles dynamic interference between targets, updates the tracking strategy through the policy gradient method, adjusts the tracking path of each target, and reduces mistracking and target loss. It can not only achieve real-time target trajectory tracking, but also continuously adjust the strategy through reinforcement learning to adapt to the complex dynamic trajectory environment of multiple targets.

[0078] Furthermore, the evaluation dimensions include,

[0079] Tracking accuracy, based on the degree of track matching between the actual target and the predicted target, quantifies the tracking accuracy of the system.

[0080] Response time, which evaluates the time from target discovery to tracking completion, ensures that the system can respond to sudden changes in trajectory under real-time conditions.

[0081] Coverage, combining space, space-based and ground-based monitoring systems to evaluate the detection and tracking coverage of the system,

[0082] Anti-interference capability, which evaluates the stability and reliability of the system when dealing with multiple target interferences.

[0083] Furthermore, in step S4, a multi-dimensional comprehensive evaluation is performed based on the multi-target tracking results of step S3 to evaluate the tracking accuracy, response time, coverage, and anti-interference capability of the system in a multi-target dynamic environment:

[0084] Tracking accuracy evaluation is quantified by the deviation between the actual target position and the predicted position. Based on the matching degree between the tracking result and the true trajectory, the error of the target position and velocity is statistically calculated to calculate the overall tracking error.

[0085] Response time evaluation: evaluates the time it takes for the system to successfully track the target after detecting it;

[0086] Coverage assessment, which evaluates the detection and tracking coverage of the system, assessing the visibility and trackability of multiple targets based on the system's monitoring capabilities, namely the coverage of space-based, space-based, and ground-based systems;

[0087] Anti-interference capability assessment: whether the target can be distinguished and correctly tracked when target trajectories overlap and interfere with each other.

[0088] The beneficial effects of the present invention are:

[0089] This invention, by combining physical models with the Long Short-Term Memory (LSTM) deep learning model, can capture nonlinear changes in the track, effectively respond to rapid changes and mutations in the track, and maintain high-precision predictions in complex dynamic track environments, thus solving the problem that traditional systems have difficulty in responding to nonlinear changes in the track.

[0090] The present invention introduces LSTM and extended Kalman filter (EKF), which can not only make predictions based on historical orbit data, but also dynamically correct real-time observation data, making the system respond more promptly to changes in the target orbit. Reinforcement learning further optimizes the system's adaptability in the face of orbit changes, reduces response time, and improves real-time performance.

[0091] The present invention dynamically adjusts the tracking strategy through reinforcement learning and adopts an interference processing mechanism to effectively solve the problem of target trajectory overlap or dynamic interference. By introducing an interference measurement function and strategy optimization based on reinforcement learning, it can distinguish multiple targets in real time, ensuring that the target will not be lost or mistracked even in high-density and complex interference environments.

[0092] The present invention acquires data from different monitoring platforms through multi-source data fusion (the collaborative work of space, space-based and ground-based monitoring systems), outputs high-precision state estimates that are consistent in time and space, and extends the Kalman filter to greatly improve the accuracy of data fusion, significantly improve the system's target tracking accuracy, and expand the system's monitoring coverage.

[0093] The present invention can quickly detect changes in the target orbit and make timely corrections through orbit prediction combined with real-time dynamic evaluation. The prediction and dynamic correction mechanism of orbital nonlinear changes enables the system to quickly adjust the tracking strategy when the target orbit undergoes sudden changes. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. 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.

[0095] Figure 1 Schematic diagram of the flow of the method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to the present invention. DETAILED DESCRIPTION

[0096] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0097] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0098] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0099] Example 1, with reference to Figure 1This embodiment provides a method for evaluating the comprehensive capability of a monitoring system for space target monitoring, comprising the following steps:

[0100] Step S1: Collaborative collection and fusion of multi-source data.

[0101] Integrate data sources, obtain initial monitoring data from the space monitoring system, fuse data from different monitoring sources, and output high-precision, temporally and spatially consistent target monitoring information, namely fused data;

[0102] The monitoring data sources include those from space monitoring systems,

[0103] Space monitoring systems, namely, target monitoring data obtained by space telescopes, space radars, space debris monitoring satellites and the Global Positioning System (GNSS).

[0104] Space-based monitoring system, namely the high-orbit target monitoring data obtained by geosynchronous orbit GEO monitoring satellites and infrared monitoring satellites,

[0105] and data obtained from ground-based monitoring systems, namely large ground-based optical telescopes, laser ranging systems, and ground-based radar systems for long-range target monitoring;

[0106] In step S1, multi-source data from space telescopes, space radars, space debris monitoring satellites, global positioning system (GNSS), geosynchronous orbit (GEO) monitoring satellites, infrared monitoring satellites, large ground-based optical telescopes, laser ranging systems, and ground-based radar systems are fused, and the extended Kalman filter (EKF) is used as the core framework to output target state estimates, i.e., fused data.

[0107] Define the state vector, the target state vector x k Represents the spatial position and velocity at time k:

[0108] Among them, p x,k ,p y,k ,p z,k represents the three-dimensional spatial coordinates of the target at time k, v x,k ,v y,k ,v z,k represents the velocity component of the target at time k;

[0109] State transition equation: x k+1 =f(x k )+w k , where f(x k ) is a nonlinear state transfer function, combining the effects of gravitational perturbation and solar radiation pressure on the target orbit, w k is process noise, which obeys Gaussian distribution where Q kis the process noise covariance matrix;

[0110] The nonlinear state transfer function f(x k ) consists of:

[0111] Gravitational perturbation is mainly caused by the non-uniform gravitational field of the earth. The gravitational perturbation model has an acceleration a gravity Expressed as:

[0112] Where μ is the Earth's gravitational constant, r = is the distance between the target and the center of the earth, J2 is the second-order perturbation term of the earth's gravitational field, R e is the radius of the Earth, is the geocentric latitude of the target;

[0113] Solar radiation pressure model, its acceleration a srp Expressed as:

[0114] Among them, P srp is the solar radiation pressure constant, A is the cross-sectional area of ​​the target, C r is the reflection coefficient of the target, m is the mass of the target, is the unit direction vector pointing to the sun;

[0115] Total acceleration a k From the superposition of gravitational perturbation and solar radiation pressure, we can get:

[0116] a k =a gravity +a srp ;

[0117] Based on the current state x k and the total acceleration a k , the state update equation of the target at the next moment is expressed as:

[0118] Where Δt is the time step, a x,k ,a y,k ,a z,k is the total acceleration a k The various components of , combined with gravitational perturbations and solar radiation pressure;

[0119] The target state estimation method output in step S1 also includes:

[0120] Calculate the Jacobian matrix F of the state transfer equation k , the Jacobian matrix F k By the state vector x k Taking partial derivatives, we get Jacobian matrix F kUsed to linearize the state transfer equation;

[0121] According to the state transition equation, predict the state at the next moment in, represents the predicted state vector at time k+1, represents the state transition function;

[0122] According to the Jacobian matrix F k and process noise Q k , the covariance matrix of the predicted state estimate:

[0123] Among them, P k+1|k represents the covariance matrix of the predicted state, F k The Jacobian matrix representing the state transition equation is defined as P k Represents the state covariance matrix at the current moment, Q k represents the process noise covariance matrix, Represents a transpose operation;

[0124] Calculate the Kalman gain, Kalman gain K k+1 The calculation formula is:

[0125] Among them, K k+1 represents the Kalman gain, H k+1 The Jacobian matrix representing the observation equation is defined as R k+1 represents the observation noise covariance matrix;

[0126] Combined observations z k+1 Update the state estimate with the predicted value:

[0127] in, represents the updated state estimate,

[0128] z k+1 represents the observation data, Represents a nonlinear observation model that transforms the predicted state into the observation space;

[0129] Update the covariance matrix, P k+1 =(IK k+1 H k+1 )P k+1|k , where P k+1 represents the updated covariance matrix, and I represents the identity matrix;

[0130] Specifically, multi-source data fusion and extended Kalman filtering are used to integrate multi-source data from spatial monitoring, space-based monitoring and ground-based monitoring systems to provide high-precision, temporally and spatially consistent target state estimation. Multi-source collaborative fusion greatly reduces the limitations faced by a single observation system.

[0131] Step S2, spatiotemporal dynamic evaluation,

[0132] Predict nonlinear track changes, perform dynamic evaluation based on the fusion data output in step S1, and output track prediction data;

[0133] In step S1, the target state is predicted and updated through the prediction phase and update of the extended Kalman filter to generate a state estimate, which includes the three-dimensional position and velocity information of the target;

[0134] In step S2, the target trajectory change is predicted based on the fused data, i.e., the output of step S1;

[0135] The track nonlinear change prediction step in step S2 includes:

[0136] Based on the fusion data of step S1, that is, the current state estimation, a nonlinear orbit prediction model is used to capture the target orbit change and predict the future orbit state. Assume that at time k, the historical state data is {x k-n ,x k-n+1 ,…,x k}, n is the window length of historical data, the historical data is input into the long short-term memory network LSTM to predict nonlinear track changes, and the historical state data is recorded as:

[0137] Represents the historical trajectory state sequence of the target, including the state information of the past n moments;

[0138] Use LSTM for track prediction through historical track data Use the LSTM network to learn the nonlinear changes of the trajectory and output the predicted state of the target at the next moment k+1

[0139] in, is the orbital state of the target at time k+1;

[0140] Perform dynamic corrections, combine LSTM predictions with gravitational perturbations and solar radiation pressure models, and perform weighted averaging of orbital predictions:

[0141] in, Represents the corrected orbit prediction data, combining the predictions of the physical model and the LSTM model, represents the track state predicted by the LSTM model, represents the state predicted based on the physical model, α represents the balance coefficient, and orbit prediction data As the output of step S2;

[0142] In step S2, a nonlinear orbit prediction is performed using an LSTM model based on historical orbit data. Dynamic correction is performed in combination with the physical model to output orbit prediction data. This effectively addresses nonlinear orbit changes and improves the system's prediction capability in a dynamic orbit environment.

[0143] Specifically, physical models and deep learning are introduced to predict orbital changes based on the long short-term memory network (LSTM), capture dynamic orbital changes, and combine the influence of gravitational perturbations and external forces such as solar radiation pressure to achieve dynamic correction of orbital predictions, effectively address the problem of nonlinear orbital changes, and improve the accuracy and stability of predictions. In the case of rapid changes in the target orbit, the deep learning model can be used to respond and make corrections in a timely manner, thereby improving the system's adaptability in dynamic environments.

[0144] Step S3, multi-target tracking and interference processing,

[0145] Perform multi-target identification and tracking. Use the orbit prediction data in step S2 to detect and track multiple targets in real time. The real-time detection and tracking methods include:

[0146] Perform dynamic trajectory tracking optimization, adopt reinforcement learning methods, dynamically adjust tracking strategies, and continuously optimize tracking paths and strategies through continuous environmental interaction, thereby reducing computational overhead and improving the real-time performance and accuracy of target tracking.

[0147] Multi-target dynamic interference processing, combining trajectory prediction and reinforcement learning, can effectively deal with dynamic interference and trajectory overlap of multiple targets, avoiding target loss or mistracking;

[0148] Step S3 uses reinforcement learning to optimize the target tracking strategy and handle dynamic interference between multiple targets based on the trajectory prediction data in step S2:

[0149] Suppose there are N targets, and the state of each target consists of a position and a velocity vector. For the trajectory prediction data x of the i-th target at time t, i (t) is expressed as:

[0150] Among them, p x,i (t),p y,i (t),p z,i (t) represents the three-dimensional spatial position of target i at time t, v x,i (t),v y,i(t),v z,i (t) represents the velocity component of target i at time t;

[0151] The total tracking error objective function is expressed as:

[0152] in, represents the total tracking error function, x i (t) represents the true trajectory state of the i-th target at time t, represents the predicted trajectory state of the i-th target at time t, |·| represents the Euclidean norm;

[0153] Define state, action and reward function to optimize multi-target tracking strategy, state s t Represents the state of the system, consisting of the current orbit information of all targets and observational data Composition, expressed as:

[0154] Among them, x i (t) represents the orbit prediction data of target i, z i (t) represents the sensor observation data of target i, which comes from multi-source monitoring;

[0155] Action a t Indicates the adjustment of the tracking path of each target. The adjusted path is:

[0156] in, represents the updated trajectory prediction of target i at time t+1, Δx i (t) represents the adjustment of trajectory prediction based on reinforcement learning strategy;

[0157] Reward function r t To minimize the tracking error, it is defined as the negative of the error:

[0158] Among them, r t represents the reward function, which encourages reducing the total tracking error, represents the total tracking error at the current time t;

[0159] Through the policy network π(s t ,a t ) determines the best action in a given state, minimizes the tracking error, and uses the policy gradient algorithm to update the policy network. The policy update rule is:

[0160] in, Represents the gradient of the policy, based on the current policy parameters θ, π θ (st ,a t ) represents the policy network, that is, given the state s t Next take action a t The probability of r t Represents the reward function, which is used to measure the effect of the current action;

[0161] At each time step, reinforcement learning adjusts the policy parameter θ based on the feedback from the environment, i.e., the decrease or increase in the track tracking error, to optimize the tracking strategy, reduce computational overhead and improve tracking accuracy.

[0162] In step S3, the multi-target dynamic interference processing method is:

[0163] When the trajectories of multiple targets overlap or there is dynamic interference, processing is performed to avoid target loss or mistracking. The interference measurement function D(t) is defined to measure the degree of trajectory overlap or interference between target i and target j:

[0164] Among them, D i,j (t) represents the interference measure between target i and target j at time t, |x i (t)-x j (t)| represents the difference between the trajectories of target i and target j, using the Euclidean distance metric, and σ represents the distance scale parameter to control interference;

[0165] When D i,j When (t) is close to 1, it means that the target trajectories are highly overlapped or interfered with. At this time, reinforcement learning adjusts the tracking strategy to separate the tracking paths of the two targets, reducing the risk of mistracking and loss.

[0166] When interference occurs on the target trajectory, reinforcement learning adds negative rewards for interference processing and adjusts the strategy to separate the target trajectory. The adjusted reward function is:

[0167] Among them, λ represents the weight parameter, which controls the impact of the interference measure on the total reward. represents the sum of interference measures among all targets;

[0168] Output multi-target tracking results, that is, trajectory prediction results after dynamic adjustment and interference processing by reinforcement learning: Finally, the tracking results of multiple targets at time t+1 are output

[0169] Dynamic trajectory tracking and optimization of multiple targets are performed through reinforcement learning, and the interference measurement function D is defined. i,j(t) Dealing with dynamic interference between targets, updating the tracking strategy through the policy gradient method, adjusting the tracking path of each target, and reducing mistracking and target loss. This not only enables real-time target trajectory tracking, but also continuously adjusts the strategy through reinforcement learning to adapt to the complex dynamic trajectory environment of multiple targets;

[0170] Specifically, in terms of multi-target tracking, reinforcement learning greatly optimizes the system's tracking strategy. Through reinforcement learning methods, the tracking path and strategy are continuously adjusted according to environmental feedback, which improves the real-time performance of tracking and reduces computational overhead. In addition, when there is interference or trajectory overlap between targets, reinforcement learning combined with trajectory prediction can effectively handle the interference of multiple targets, ensuring that the system can still accurately distinguish and track each target when there is multi-target interference.

[0171] Step S4, system comprehensive capability assessment,

[0172] Perform dimensional evaluation, based on the tracking results of step S3, to comprehensively evaluate the performance of the system under multi-target and dynamically changing conditions;

[0173] The evaluation dimensions include:

[0174] Tracking accuracy, based on the degree of track matching between the actual target and the predicted target, quantifies the tracking accuracy of the system.

[0175] Response time, which evaluates the time from target discovery to tracking completion, ensures that the system can respond to sudden changes in trajectory under real-time conditions.

[0176] Coverage, combining space, space-based and ground-based monitoring systems to evaluate the detection and tracking coverage of the system,

[0177] Anti-interference capability, which evaluates the stability and reliability of the system when dealing with multiple target interferences,

[0178] In step S4, a multi-dimensional comprehensive evaluation is performed based on the multi-target tracking results of step S3 to evaluate the tracking accuracy, response time, coverage, and anti-interference capability of the system in a multi-target dynamic environment:

[0179] Tracking accuracy evaluation is quantified by the deviation between the actual target position and the predicted position. Based on the matching degree between the tracking result and the true trajectory, the error of the target position and velocity is statistically calculated to calculate the overall tracking error.

[0180] Response time evaluation: evaluates the time it takes for the system to successfully track the target after detecting it;

[0181] Coverage assessment, which evaluates the detection and tracking coverage of the system, assessing the visibility and trackability of multiple targets based on the system's monitoring capabilities, namely the coverage of space-based, space-based, and ground-based systems;

[0182] Anti-interference capability assessment: detecting whether the target can be distinguished and correctly tracked when target trajectories overlap and interfere with each other;

[0183] Specifically, in terms of comprehensive capability evaluation, the system performance is quantitatively analyzed through four dimensions: tracking accuracy, response time, coverage, and anti-interference capability.

[0184] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A method for evaluating the comprehensive capability of a monitoring system for space target monitoring, characterized in that: include, Step S1, multi-source data collaborative collection and fusion, data source integration, obtaining initial monitoring data from the space monitoring system, fusion of data from different monitoring sources, and output of target monitoring information, i.e. fusion data; Step S2, spatiotemporal dynamic evaluation, prediction of orbital nonlinear changes, based on the fusion data output in step S1, dynamic evaluation is performed to output orbital prediction data; Step S3, multi-target tracking and interference processing, multi-target identification and tracking, using the orbit prediction data in step S2, real-time detection and tracking of multiple targets, the real-time detection and tracking method includes: Perform dynamic trajectory tracking optimization, use reinforcement learning methods, and dynamically adjust tracking strategies; Perform multi-target dynamic interference processing and combine orbit prediction and reinforcement learning; Step S4, system comprehensive capability evaluation, performs dimensional evaluation, and comprehensively evaluates the system's performance under multi-objective and dynamically changing conditions based on the tracking results of step S3.

2. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 1, characterized in that: The monitoring data sources include those from space monitoring systems, Space monitoring system, namely, target monitoring data obtained by space telescopes, space radars, space debris monitoring satellites and the Global Positioning System (GNSS). The space-based monitoring system, namely the high-orbit target monitoring data obtained by geosynchronous orbit GEO monitoring satellites and infrared monitoring satellites, And ground-based monitoring systems, namely large ground-based optical telescopes, laser ranging systems and ground-based radar systems for long-distance target monitoring.

3. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 2, characterized in that: In step S1, multi-source data from space telescopes, space radars, space debris monitoring satellites, global positioning system GNSS, geosynchronous orbit GEO monitoring satellites, infrared monitoring satellites, large ground-based optical telescopes, laser ranging systems and ground-based radar systems are fused, and the extended Kalman filter EKF is used as the core framework to output target state estimation, i.e., fused data; Define the state vector, the state vector x of the target k Represents the spatial position and velocity at time k: Among them, p x,k ,p y,k ,p z,k represents the three-dimensional spatial coordinates of the target at time k, v x,k ,v y,k ,v z,k represents the velocity component of the target at time k; State transition equation: x k+1 =f(x k )+w k , where f(x k ) is a nonlinear state transfer function, combining the influence of gravitational perturbation and solar radiation pressure on the target orbit, w k is the process noise, which obeys Gaussian distribution w k ~N(0,Q k ), where Q k is the process noise covariance matrix; The nonlinear state transfer function f(x k ) consists of: Gravitational perturbation model, its acceleration a gravity It is expressed as: Where μ is the Earth's gravitational constant, is the distance between the target and the center of the earth, J2 is the second-order perturbation term of the earth's gravitational field, R e is the radius of the Earth, is the geocentric latitude of the target; Solar radiation pressure model, its acceleration a srp It is expressed as: Among them, P srp is the solar radiation pressure constant, A is the cross-sectional area of ​​the target, C r is the reflection coefficient of the target, m is the mass of the target, is the unit direction vector pointing to the sun; Total acceleration a k From the superposition of gravitational perturbation and solar radiation pressure, we get: a k =a gravity +a srp ; Based on the current state x k and the total acceleration a k , the state update equation of the target at the next moment is expressed as: Among them, Δt is the time step, a x,k ,a y,k ,a z,k is the total acceleration a k , combining gravitational perturbations and solar radiation pressure.

4. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 3, characterized in that: The method of outputting the target state estimation in step S1 also includes: Calculate the Jacobian matrix F of the state transfer equation k , the Jacobian matrix F k From the state vector x k Taking partial derivatives, we get Jacobian matrix F k Used to linearize the state transfer equation; According to the state transfer equation, predict the state at the next moment in, represents the predicted state vector at time k+1, represents the state transition function; According to the Jacobian matrix F k and process noise Q k , the covariance matrix of the predicted state estimate: Among them, P k+1|k represents the covariance matrix of the predicted state, F k The Jacobian matrix representing the state transfer equation is defined as P k represents the state covariance matrix at the current moment, Q k represents the process noise covariance matrix, Represents a transpose operation; Calculate the Kalman gain, Kalman gain K k+1 The calculation formula is: Among them, K k+1 represents the Kalman gain, H k+1 The Jacobian matrix representing the observation equation is defined as R k+1 represents the observation noise covariance matrix; Combined observations z k+1 Update the state estimate with the predicted value: in, represents the updated state estimate, z k+1 represents the observed data, Represents a nonlinear observation model that transforms the predicted state into the observation space; Update the covariance matrix, P k+1 =(IK k+1 H k+1 ) k+1|k , where P k+1 represents the updated covariance matrix, and I represents the identity matrix.

5. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 4, characterized in that: In step S1, the target state is predicted and updated through the prediction phase and update of the extended Kalman filter to generate a state estimate, and the target state estimate includes the three-dimensional position and velocity information of the target; In step S2, the target track change is predicted based on the fused data, i.e., the output of step S1.

6. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 5, characterized in that: The track nonlinear change prediction step in step S2 includes: Based on the fusion data of step S1, i.e. the current state estimation, a nonlinear orbit prediction model is used to capture the target orbit change and predict the future orbit state. Assume that at time k, the historical state data is {x k-n ,x k-n+1 ,…,x k }, n is the window length of historical data, the historical data is input into the long short-term memory network LSTM to predict nonlinear track changes, and the historical state data is recorded as: Represents the historical trajectory state sequence of the target, including the state information of the past n moments; Using LSTM for orbit prediction, using historical orbit data Use the LSTM network to learn the nonlinear changes of the track and output the predicted state of the target at the next moment k+1 in, is the orbital state of the target at time k+1; Perform dynamic correction, combine LSTM prediction with gravitational perturbation and solar radiation pressure model, and perform weighted average of orbit prediction results: in, represents the corrected orbit prediction data, combining the predictions of the physical model and the LSTM model, represents the track state predicted by the LSTM model, represents the state predicted based on the physical model, α represents the balance coefficient, and orbit prediction data As the output of step S2.

7. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 6, characterized in that: Step S3 uses reinforcement learning to optimize the target tracking strategy and handle dynamic interference between multiple targets based on the trajectory prediction data in step S2: Suppose there are N targets, each target's state consists of a position and velocity vector. For the trajectory prediction data x of the i-th target at time t, i (t) is expressed as: Among them, p x,i (t),p y,i (t),p z,i (t) represents the three-dimensional spatial position of target i at time t, v x,i (t),v y,i (t),v z,i (t) represents the velocity component of target i at time t; The total tracking error objective function is expressed as: in, represents the total tracking error function, x i (t) represents the actual trajectory state of the i-th target at time t, represents the predicted trajectory state of the i-th target at time t, |·| represents the Euclidean norm; Define the state, action and reward function to optimize the multi-target tracking strategy. t Represents the state of the system, consisting of the current orbit information of all targets and observational data Composition, expressed as: Among them, x i (t) represents the orbit prediction data of target i, z i (t) represents the sensor observation data of target i, which comes from multi-source monitoring; Action a t Indicates the adjustment of the tracking path of each target. The adjusted path is: in, represents the updated trajectory prediction of target i at time t+1, Δx i (t) represents the adjustment of trajectory prediction based on reinforcement learning strategy; Reward function r t To minimize the tracking error, it is defined as the negative of the error: Among them, r t represents the reward function, which encourages reducing the total tracking error, represents the total tracking error at the current time t; Through the policy network π(s t ,a t ) determines the best action in a given state, minimizes the tracking error, and uses the policy gradient algorithm to update the policy network. The policy update rule is: in, Represents the gradient of the strategy, based on the current strategy parameters θ, π θ (s t ,a t ) represents the policy network, that is, given the state s t Take action a t The probability of t Represents the reward function, which is used to measure the effect of the current action; At each time step, reinforcement learning adjusts the policy parameters θ based on the feedback from the environment, i.e., the decrease or increase in the trajectory tracking error.

8. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 7, characterized in that: In step S3, the multi-target dynamic interference processing method is: When the trajectories of multiple targets overlap or there is dynamic interference, the interference measurement function D(t) is defined to measure the degree of trajectory overlap or interference between target i and target j: Among them, D i,j (t) represents the interference measure between target i and target j at time t, |x i (t)-x j (t)| represents the difference between the trajectories of target i and target j, using the Euclidean distance metric, and σ represents the distance scale parameter to control interference; When D i,j When (t) is close to 1, it means that the target trajectories are highly overlapped or interfered. At this time, reinforcement learning adjusts the tracking strategy to separate the tracking paths of the two targets. When interference occurs on the target trajectory, reinforcement learning increases the negative reward for interference processing and adjusts the strategy to separate the target trajectory. The adjusted reward function is: Among them, λ represents the weight parameter, which controls the impact of the interference measure on the total reward, represents the sum of interference measures among all targets; Output the multi-target tracking results, that is, the trajectory prediction results after dynamic adjustment and interference processing by reinforcement learning: Finally, the tracking results of multiple targets at time t+1 are output 9. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 8, characterized in that: The evaluation dimensions in step S4 include, Tracking accuracy, based on the degree of track matching between the actual target and the predicted target, quantifies the tracking accuracy of the system. Response time, which evaluates the time it takes for the system to detect a target and complete tracking. Coverage, combining space, space-based and ground-based monitoring systems to evaluate the detection and tracking coverage of the system, Anti-interference capability evaluates the stability and reliability of the system when dealing with multiple target interferences.

10. The method for evaluating the comprehensive capability of a monitoring system for space target monitoring according to claim 9, characterized in that: In step S4, a multi-dimensional comprehensive evaluation is performed based on the multi-target tracking results of step S3 to evaluate the tracking accuracy, response time, coverage and anti-interference ability of the system in a multi-target dynamic environment: Tracking accuracy evaluation, which is quantified by the deviation between the actual target position and the predicted position. According to the matching degree between the tracking result and the real trajectory, the error of the target position and velocity is counted to calculate the overall tracking error; Response time evaluation, which evaluates the time it takes for the system to successfully track the target after detecting it; Coverage assessment, which evaluates the detection and tracking coverage of the system, and assesses the visibility and trackability of multiple targets based on the system's monitoring capabilities, i.e., the coverage of space, space-based, and ground-based systems; Anti-interference capability assessment detects whether the target can be distinguished and correctly tracked when the target trajectories overlap and interfere with each other.

Images