Near space multi-target large-scale constellation collaborative observation method

Abstract

The invention relates to a near space multi-target large-scale constellation collaborative observation method. All-weather tracking and detection are carried out on multiple targets flying at hypersonic velocity in near space based on a space-based observation system on-orbit satellite; comprising the following steps: S1, in the whole process of a near space multi-target large-scale constellation collaborative observation task, distributing and scheduling the near space multi-target large-scale constellation observation task by using a staged multi-source triggering collaborative observation task planning method; and S2, in a near space multi-target low-uncertainty stage, resource scheduling is carried out through a learning-based low-uncertainty multi-target collaborative observation task planning method. The method has the advantages that large-scale observation constellation resources are effectively distributed, the multi-target capacity is high, the target loss rate is low, and the resource utilization efficiency is high.

Description

[Technical field]

[0001] The present invention relates to the field of satellite aerospace technology, and in particular to a near-space multi-target large-scale constellation collaborative observation method. [Background technology]

[0002] Near space refers to the space area with an altitude of 20 to 100 km above the ground. The altitude is between the upper limit of the operating altitude of traditional aircraft and the lower limit of the operating altitude of spacecraft, covering most areas of the stratosphere, the mesosphere and parts of the thermosphere. Near Space Hypersonic Vehicle (NSHV) refers to an aircraft that operates in near space, has a flight speed higher than Mach 5, and can use its own power or aerodynamic force to perform large-scale maneuvers. Faced with the high speed, near space airspace operating altitude and strong maneuverability of near hypersonic targets, traditional land / sea-based radars are affected by factors such as line of sight and ground obstruction, making it difficult to effectively detect targets at long distances. At the same time, near hypersonic targets can avoid the effective areas of multi-point distributed radars with their strong maneuverability, and fixed radar systems find it difficult to relay detection of them.

[0003] The space-based observation system uses satellites in orbit as detection resources, which can effectively reduce the impact of geographical and environmental factors to achieve all-weather tracking and detection of targets. In order to achieve the rapid response requirements for all-weather monitoring tasks around the world, firstly, the number of available satellites in the space-based observation system must be able to meet global multiple coverage at any time; secondly, in order to cope with the high-speed and highly maneuverable movement characteristics of near-space hypersonic targets, the observation satellites need to be agile satellites with multiple degrees of freedom and rapid maneuverability. Therefore, in order to achieve the observation mission of near-space hypersonic targets, the space-based monitoring system is a large-scale constellation system composed of a large number of agile satellites.

[0004] Agile satellites refer to satellites with large-angle attitude maneuverability and high stability. They can adjust their attitude in a short time to complete various tasks.

[0005] The present invention aims at the technical problem of all-weather tracking and detection of nearby high-altitude targets and makes technical improvements to the large-scale constellation collaborative observation method of multiple targets in near space. [Summary of the invention]

[0006] The purpose of the present invention is to propose a near-space multi-target large-scale constellation collaborative observation method with effective allocation of large-scale observation constellation resources, strong multi-target capability, low target loss rate and high resource utilization efficiency.

[0007] To achieve the above-mentioned purpose, the technical solution adopted by the present invention is a large-scale constellation collaborative observation method for multiple targets in near-space, which performs all-weather tracking and detection of multiple targets flying at hypersonic speeds in near-space based on on-orbit satellites of a space-based observation system; the method comprises the following steps:

[0008] S1. During the whole process of near-space multi-target large-scale constellation collaborative observation mission, a phased multi-source triggered collaborative observation mission planning method is used to allocate and schedule near-space multi-target large-scale constellation observation mission;

[0009] S2. In the near-space multi-target low-uncertainty stage, resource scheduling is performed through a learning-based low-uncertainty multi-target collaborative observation task planning method.

[0010] Preferably, step S1 specifically includes the following sub-steps:

[0011] S11, near-space multi-target large-scale constellation collaborative observation mission stage classification, according to the target model TAR M The current observation stage set COPS is used to divide the target into task stages, and the target set TIS for the type identification stage, the target set VHMS for the very high uncertainty maintenance stage, the target set MTS for the high uncertainty maintenance stage and the target set HOS for the low uncertainty handover stage are obtained;

[0012] S12, according to the trigger source that affects the collaborative observation resource allocation, resource allocation and scheduling are performed for the targets in the target set TIS in the type identification phase;

[0013] S13, according to the trigger source affecting the collaborative observation resource allocation, the resource allocation and scheduling of the targets in the very high uncertainty maintenance phase target set VHMS;

[0014] S14, according to the trigger source affecting the collaborative observation resource allocation, the resource allocation and scheduling of the targets in the high uncertainty maintenance phase target set MTS;

[0015] S15. According to the trigger source that affects the allocation of collaborative observation resources, resources are allocated and scheduled for the targets in the target set HOS in the low uncertainty handover phase.

[0016] Preferably, the trigger sources affecting the collaborative observation resource allocation in step S1 include: the appearance of a new target, the completion of a target phase task, the triggering of a target cycle in a high uncertainty maintenance phase, and the change of a target maneuvering state.

[0017] Preferably, step S12 specifically includes the following sub-steps:

[0018] S121, dynamically prioritize the targets in the target set TIS in the type identification stage to obtain TIS sorted_p, and initialize the occupied resource set UAS;

[0019] S122, yes Each target in the algorithm is allocated observation resources based on the observation resource allocation algorithm in the single-step feedback type identification phase;

[0020] S123. Add the allocated observation resources to the occupied resource set Update t at the same time shce +1 time planning

[0021] S124, repeat steps S122 to S123 until TIS sorted_p All goals in the resource allocation are completed;

[0022] S125, if the target TIS j If the corresponding observation resource allocation and scheduling are insufficient, the target TIS j Add the very high uncertainty maintenance phase target set VHMS.

[0023] Preferably, step S13 specifically includes the following sub-steps:

[0024] S131, dynamically prioritize the targets in the very high uncertainty maintenance stage target set VHMS to obtain VHMS sorted_p , and initialize the occupied resource set UAS;

[0025] S132, yes Observation resources are allocated to each target using the optimal single-star observation strategy;

[0026] S133. Add the allocated observation resources to the occupied resource set Update t at the same time shce +1 time planning

[0027] S134, repeat steps S132 to S133 until VHMS sorted_p Complete resource allocation for all objectives.

[0028] Preferably, step S14 specifically includes the following sub-steps:

[0029] S141, determine whether the observation resources in the MTS observation planning scheme for the high uncertainty maintenance stage target set are consistent with t shce +1 time, the target set TIS of the type identification stage and the target set VHMS of the very high uncertainty maintenance stage conflict, and the observation period Δt of the high uncertainty maintenance stage M_MT Is it finished? If not, go to step S15;

[0030] S142, dynamically prioritize the targets in the high uncertainty maintenance stage target set MTS to obtain MTS sorted_p , and initialize the occupied resource set UAS;

[0031] S143, yes Each target in the algorithm is allocated observation resources by maintaining cooperative observation algorithm based on intention;

[0032] S144. Add all allocated observation resources in the observation period of the high uncertainty maintenance phase to the occupied resource set Update t at the same time shce +1 to t shce +Δt M_MT Time planning plan

[0033] S145, repeat steps S143 to S144 until VHMS sorted_p Complete resource allocation for all objectives.

[0034] Preferably, step S15 specifically includes the following sub-steps:

[0035] S151, dynamically prioritize the targets in the low uncertainty handover stage target set HOS to obtain HOS sorted_p , and initialize the occupied resource set UAS;

[0036] S152, yes Each target in the algorithm executes a learning-based low-uncertainty multi-target collaborative observation task planning algorithm to allocate observation resources;

[0037] S153. Add all allocated observation resources in the observation period of the low uncertainty handover phase to the occupied resource set Update t at the same time shce +1 to t shce +Δt M_MT Time planning plan

[0038] S154, repeat steps S153 to S154 until HOS sorted_p Complete resource allocation for all objectives.

[0039] Preferably, step S2 specifically includes the following sub-steps:

[0040] S21, in the task preprocessing phase, the stable task interval division is calculated according to the initial trajectory and constellation information of the target, and the output result is used as the input data of the loop scheduling phase;

[0041] S22. In the cyclic scheduling link, an observation pair decision network based on the proximal policy optimization reinforcement learning method is used to schedule resources for time-sensitive, low-uncertainty multi-target collaborative observation tasks through a time-driven decision-making process, and to allocate observation resources to all targets in each sub-task interval.

[0042] Preferably, step S21 specifically includes the following sub-steps:

[0043] S21, initial visibility calculation, based on the initial predicted input trajectory and constellation information of the target, considering the detection distance and angle constraints of each satellite in the constellation, calculates the visible time window of each target with all observation satellites during its life cycle;

[0044] S22. Calculate the maximum stable time period of each target resource according to the visible time window;

[0045] S23, further divide the task intervals, combine the observed stable interval set obtained in step S22 with the planned task start and end time to obtain a task interval partition set, perform interval length check on the task interval partition set, perform secondary division for those with lengths greater than mTD, and finally obtain a task interval set.

[0046] Preferably, step S22: the observation-to-decision network based on the proximal policy optimization reinforcement learning method includes an outer loop and an inner loop, the outer loop represents a set of task intervals obtained by task preprocessing, and the inner loop is the observation-to-decision network.

[0047] A near-space multi-target large-scale constellation collaborative observation method of the present invention has the following beneficial effects: through a learning-based collaborative observation task planning method and a staged multi-source triggered collaborative observation task planning method, the collaborative observation task planning problem caused by various typical explosion scenarios of multiple hypersonic aircraft in near space can be effectively dealt with; in the multi-target low uncertainty stage, resource scheduling is performed through a learning-based collaborative observation task planning method to solve the resource conflict problem caused by multiple targets in the low uncertainty stage; during the entire mission, a staged triggered collaborative observation task planning method is used to allocate and schedule large-scale constellation observation tasks; it has the advantages of strong multi-target capability, low target loss rate, high resource utilization efficiency, etc.

Brief Description of the Drawings

[0048] Figure 1 It is a step-by-step diagram of a method for collaborative observation of large-scale constellations of multiple targets in near-space.

[0049] Figure 2 It is a full-stage triggering flow chart of a near-space multi-target large-scale constellation collaborative observation method.

[0050] Figure 3It is a multi-target low-uncertainty phase flow chart of a near-space multi-target large-scale constellation collaborative observation method.

[0051] Figure 4 It is a schematic diagram of three near-space hypersonic target trajectories with the same initial position O and landing point in the key area B. [Specific implementation method]

[0052] The present invention will be further described below in conjunction with embodiments and with reference to the accompanying drawings.

[0053] Example

[0054] This embodiment implements a near-space multi-target large-scale constellation collaborative observation method.

[0055] Figure 1 This is a step-by-step diagram of a near-space multi-target large-scale constellation collaborative observation method. Figure 1 As shown, this embodiment discloses a large-scale constellation collaborative observation task planning method for multiple hypersonic aircraft in near space. Through the collaborative observation task planning method based on learning and the collaborative observation task planning method based on staged multi-source triggering, it can effectively deal with the collaborative observation task planning problems caused by various typical burst scenarios of multiple hypersonic aircraft in near space. In the multi-target low uncertainty stage, resource scheduling is performed through the collaborative observation task planning method based on learning to solve the resource conflict problem caused by multiple targets in the low uncertainty stage; the large-scale constellation observation tasks are allocated and scheduled using the collaborative observation task planning method based on staged triggering throughout the task. The large-scale constellation collaborative observation task planning method based on this embodiment has the advantages of strong multi-target capability, low target loss rate, and high resource utilization efficiency.

[0056] 1. Phased multi-source triggered collaborative observation mission planning process

[0057] In the problem of uncertain multi-target collaborative observation task planning, since the multi-target itself has diverse burst forms, non-cooperative characteristics and uncertainty, and the observation system and the targets are both in high-speed motion when performing the observation task, it is a highly dynamic resource allocation problem. Therefore, it is necessary to first analyze the trigger sources of task replanning in the whole process of uncertain multi-target collaborative observation task planning.

[0058] Figure 2 It is a full-stage triggering flow chart of a near-space multi-target large-scale constellation collaborative observation method. Figure 2 As shown in Figure 2, in this problem, the trigger sources that affect the allocation of collaborative observation resources mainly include the following aspects:

[0059] 1. New targets emerge

[0060] The appearance of a new target means that after the task planning system completes resource allocation at a historical moment, the observation system receives a new target observation demand from external input during the process of executing the observation command according to the established plan. After the new target appears, the task needs to be re-planned immediately based on the initial information of the target and the occupancy of the observation resources in the corresponding observation coverage area. Since the uncertainty of the new target is at its highest at the current task planning moment, at least a single observation resource needs to be allocated to the target to ensure the observation system's perception of it. If two or more targets appear at the same time, it is first necessary to divide the observation area according to the initial position of the target, consider the resource occupancy in different observation areas, and complete the coordinated observation resource allocation.

[0061] 2. Target stage tasks completed

[0062] The completion of the target phase task refers to the end of the observation system's established observation task phase for the corresponding target. Since the observation task targets and observation modes at different target phases are not the same, task re-planning needs to be triggered when the target completes a certain phase of the observation task. For example, when the target in the collaborative observation type identification phase is tracked and observed and the trajectory type is identified in the collaborative observation mode, when the number of trajectory points that meet the accuracy reaches the type identification algorithm threshold, the collaborative tracking observation type identification phase task of the corresponding target is completed, the uncertainty of the corresponding target is reduced, and the target enters the high uncertainty maintenance phase, and the occupation of observation resources is reduced. The task planning system needs to choose to release the corresponding observation resources or maintain the tracking observation state of the target based on the target information and other multi-target states; or when the high uncertainty maintenance phase task of the target ends and enters the low uncertainty handover phase, the task re-planning should be triggered immediately, the current target should be added to the low uncertainty handover phase target set, and the learning-based low uncertainty multi-target collaborative observation task planning algorithm should be used to allocate resources to all targets in the set.

[0063] 3. High uncertainty maintenance phase target cycle trigger

[0064] The target cycle triggering in the high uncertainty maintenance stage means that for the target to be observed in the high uncertainty stage, the corresponding observation task planning cycle is to estimate the potential distribution position according to the target intention and target mobility, and determine the planning cycle and generate the observation resource allocation plan according to the geometric relationship between the coverage capability of the observation system at the corresponding moment and the target potential estimated position distribution space. During this cycle, if there is no other trigger source to change the observation plan, and the observation system observes all targets in the high uncertainty maintenance stage set according to the inherent observation plan. When the current observation cycle reaches the end time, it is necessary to re-estimate the potential distribution position and the observation coverage calculation based on the updated target-related information to determine the duration of the next planning cycle and the observation resource allocation plan.

[0065] 4. Changes in target maneuvering status

[0066] The change of the target maneuvering state refers to the process in which the observation system estimates the maneuvering state of the target through tracking filtering during the coordinated tracking of the target according to the mission planning scheme. When the target is detected to enter the lateral maneuvering mode, the mission replanning needs to be triggered. This is because when the target enters the lateral maneuvering state, the mission planning scheme based on the target's historical information will become invalid in a short time due to the maneuvering of the target. In addition, the maneuvering process of the target is usually important information for determining its attack intention. By tracking its maneuvering process, the accuracy of its intention inference can be improved. In addition, the target entering the maneuvering state usually occurs after the target type identification task stage and before the handover task stage. In order to obtain the moment when the target completes the maneuver, the observation system is also required to perform mission replanning to ensure the tracking observation of the target.

[0067] To solve the problem of overlapping observation mission stages caused by uncertain multiple targets in the single-point burst mode, a resource allocation strategy based on target mission stages is used to divide all targets within the coverage area of ​​the observation system into four sets according to the observation mission stages, namely: type identification stage target set, very high uncertainty maintenance stage target set, high uncertainty maintenance stage target set and low uncertainty handover stage target set.

[0068] The strategy flow of overlapping collaborative observation resource allocation in the uncertainty multi-target stage is shown as follows. The main strategy is to prioritize the allocation of observation resources to the type identification stage target, followed by the very high uncertainty maintenance stage target, the high uncertainty maintenance stage target, and finally the low uncertainty handover stage target. The specific process is as follows:

[0069] 1. Multi-objective task stage classification, according to the target model TAR M The current observation stage set COPS is used to divide the target into task stages, and TIS, VHMS, MTS, and HOS represent the target set in the type identification stage, the target set in the very high uncertainty maintenance stage, the target set in the high uncertainty maintenance stage, and the target set in the low uncertainty handover stage respectively;

[0070] 2. Allocate resources to goals within TIS

[0071] a) Dynamically prioritize the targets in TIS to obtain TIS sorted_p , and initialize the occupied resource set UAS;

[0072] b) Yes Each target in the algorithm executes the observation resource allocation algorithm based on the single-step feedback type identification phase;

[0073] c) Add the allocated observation resources to the occupied resource set Update t at the same time shce+1 time planning

[0074] d) Repeat steps 2.b to 2.c to TIS sorted_p All goals in the resource allocation are completed;

[0075] e) If the target TIS j If the corresponding observation resources are insufficient, they will be added to VHMS;

[0076] 3. Allocate resources to targets within VHMS

[0077] f) Dynamically prioritize the targets in VHMS to obtain VHMS sorted_p , and initialize the occupied resource set UAS;

[0078] g) Yes Observation resources are allocated to each target using the optimal single-star observation strategy;

[0079] h) Add the allocated observation resources to the occupied resource set Update at the same time

[0080] t shce +1 time planning

[0081] i) Repeat steps 3.b to 3.c for VHMS sorted_p All goals in the resource allocation are completed;

[0082] 4. Allocate resources to targets within MTS

[0083] j) Determine whether the observation resources related to the MTS observation planning scheme are consistent with t shce Is there a conflict or Δt between TIS and VHMS at time +1? M_MT End or not; if there is no conflict in the observation scheme and the observation period has not ended, go to step 5, otherwise execute steps 4.b to 4.e;

[0084] k) Dynamically prioritize the targets in MTS to obtain MTS sorted_p , and initialize the occupied resource set UAS;

[0085] 1) Yes Each target in the target maintains the cooperative observation algorithm based on the intention;

[0086] m) Add all allocated observation resources within the high uncertainty maintenance mission planning cycle to the occupied resource set Update t at the same time shce +1 to t shce +Δt M_MT Time planning plan

[0087] n) Repeat steps 4.c to 4.d for VHMS sorted_p All goals in the resource allocation are completed;

[0088] 5. Allocate resources to goals within HOS

[0089] o) Dynamically prioritize the targets in HOS to obtain HOS sorted_p , and initialize the occupied resource set UAS;

[0090] p) Yes Implement a learning-based low-uncertainty multi-target collaborative observation mission planning algorithm;

[0091] q) Add all allocated observation resources within the low uncertainty handover task planning cycle to the occupied resource set Update t at the same time shce +1 to t shce +Δt M_MT Time planning plan

[0092] 2. Learning-based low-uncertainty multi-target collaborative observation task planning method

[0093] Figure 3 It is a multi-target low uncertainty phase flow chart of a near-space multi-target large-scale constellation collaborative observation method. Figure 3 As shown in the figure, the low-uncertainty multi-target collaborative observation scheduling problem is highly related to time. The time-sensitive characteristics of the target will significantly increase the complexity of scheduling. Due to the introduction of time variables, even for a single-target observation task, the number of observation schemes is theoretically huge. In order to effectively solve the time-sensitive multi-target collaborative observation scheduling problem, a time-sensitive task observation planning method is proposed. It includes two main parts: task preprocessing and cyclic scheduling. Task preprocessing performs preliminary calculations based on the constellation and target information, and the output results are used as input data for the cyclic scheduling part. In the cyclic scheduling link, an observation pair decision network based on the proximal policy optimization (PPO) reinforcement learning method is proposed for the constellation scheduling problem with time-sensitive multi-target observations. A time-driven decision process is designed to make scheduling more efficient.

[0094] The initial trajectory information of the target from external input and the information of the observation constellation system itself constitute the algorithm input. The task preprocessing part calculates the stable task interval division according to the initial trajectory and constellation information of the target. The cyclic scheduling part uses the observation pair decision network to allocate observation resources for all targets in each sub-task interval according to the calculation results of the previous part.

[0095] Task preprocessing includes three main steps. The first step is the initial visibility calculation. Based on the initial predicted input trajectory and constellation information of the target, the visible time window with all observation satellites during the life cycle of each target is calculated considering the detection distance and angle constraints of each satellite in the constellation. The second step is to calculate the maximum time period of resource stability for each target based on the visible time window. The set of candidate satellites corresponding to each target will not change during this period, avoiding resource conflicts and reducing the complexity of scheduling. This is done by calculating the time points when all target candidate satellite groups change. Then, these time points are merged to obtain a time set. These time sets are arranged in order from small to large, and the candidate satellite sets of all targets can remain stable within the time interval formed by every two time points. The third step is to further divide the task interval. The second step obtains a stable observation interval, but some observation intervals may last too long. When the interval is greater than mTD, a second division is required to improve the efficiency of the task interval and avoid wasting observation resources. The set of observation stability intervals obtained in the second step is combined with the start and end times of the planned task to obtain the task interval partition set. This set is checked for interval length, and a second division is performed for those with a length greater than mTD, and finally the task interval set is obtained.

[0096] The loop scheduling part contains a trained observation pair decision network and two layers of loops. The outer loop represents the set of mission intervals during preprocessing. The set of satellites available to each target remains unchanged in each interval. The inner loop is the observation pair decision network, which is used to allocate resources to each target in each mission interval. The network allocates observation resources based on tO and tD of the given target at each decision point. Since each interval lasts between mTO and mTD, the network will not cause any target to be lost by choosing not to allocate resources in any given interval. When the observation pair allocation decision is completed, the constellation information and target information are updated. When all targets have completed the observation pair decision, the inner loop ends. When all mission intervals are completed, the outer loop ends. Finally, the planning result is output.

[0097] Typical scenario: Aerodynamic parameter settings for near-space hypersonic aircraft, mass 907.185 kg, aerodynamic reference area 0.4839 m 2 , the maximum heat flux is 7000kW / m 2The initial position of the target is set to 145° east longitude, 15° north latitude, 80km altitude, 6000m / s initial speed, and 0° trajectory inclination. In the large-scale observation system, each satellite is at an altitude of 1000km, set to a circular orbit with an eccentricity of 0, an orbital inclination of 60°, a constellation configuration of Walker-σ, parameters of 256 / 16 / 0, a maximum effective detection range of 4000km, a half-viewing angle of 1°, and an angular measurement error of 100μrad. This method can be used to achieve effective full-range tracking and observation of the target.

[0098] The landing point is set as the key area G, and a library of 30 trajectories of nearby hypersonic targets is obtained. The 250s after intercepting its trajectory is used as the low-uncertainty handover stage for algorithm verification. In the algorithm performance comparison part, 5 trajectories are randomly selected from the trajectory library for testing using a random method.

[0099] Figure 4 It is a schematic diagram of three near-space hypersonic target trajectories with the same initial position O and landing point position in the key area B. Figure 4 As shown in the figure, with O as the same initial position and the landing point position set as the key area B, three trajectories of adjacent hypersonic targets are obtained. The initial appearance time of target 1 is taken as time 0, and the initial times of target 2 and target 3 are 20s and 30s respectively. The three targets move towards the landing point B with different initial headings and maximum maneuverability at different times.

[0100] The average running time of learning-based task planning methods, Full-action, Random, GA, and H-GASA in 100 Monte Carlo experiments is statistically analyzed. The benchmark algorithm Random strategy has the shortest average running time because it does not require a lot of calculations. Next are the learning-based task planning method and Full-action strategy, with running times of 5.78 times and 5.87 times that of the benchmark algorithm, respectively. The running time of the GA algorithm is 80.1 times that of the benchmark algorithm. H-GASA has the longest running time, which is 7405.43 times that of the benchmark algorithm, because it combines two optimization methods. From the analysis of running time, compared with the benchmark algorithm, the learning-based task planning method has the shortest average running time. Considering the time-sensitive characteristics of nearby space targets, the shorter the average running time of the algorithm, the more timely the observation of the target can be, which can reduce the problem of target loss.

[0101] Those skilled in the art will appreciate that all or part of the steps to implement the above embodiments may be accomplished by hardware, or may be accomplished by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, wherein the storage medium may be a magnetic disk, an optical disk, a read-only memory (ROM) or a random access memory (RAM), etc.

[0102] The above is only a preferred embodiment of the present invention. It should be pointed out that ordinary technicians in this technical field can make several improvements and supplements without departing from the principle of the present invention. These improvements and supplements should also be regarded as the scope of protection of the present invention.

Claims

1. A large-scale constellation collaborative observation method for multiple targets in near-space, based on the on-orbit satellites of the space-based observation system to perform all-weather tracking and detection of multiple targets flying at hypersonic speeds in near-space; characterized in that The following steps are involved: S1. During the whole process of near-space multi-target large-scale constellation collaborative observation mission, a phased multi-source triggered collaborative observation mission planning method is used to allocate and schedule near-space multi-target large-scale constellation observation mission; S2. In the near-space multi-target low-uncertainty stage, resource scheduling is performed through a learning-based low-uncertainty multi-target collaborative observation task planning method.

2. The near-space multi-target large-scale constellation collaborative observation method according to claim 1 is characterized in that Step S1 specifically includes the following sub-steps: S11, near-space multi-target large-scale constellation collaborative observation mission stage classification, according to the target model TAR M The current observation stage set COPS is used to divide the target into task stages, and the target set TIS for the type identification stage, the target set VHMS for the very high uncertainty maintenance stage, the target set MTS for the high uncertainty maintenance stage, and the target set HOS for the low uncertainty handover stage are obtained; S12, according to the trigger source that affects the collaborative observation resource allocation, resource allocation and scheduling are performed for the targets in the target set TIS in the type identification phase; S13, according to the trigger source affecting the collaborative observation resource allocation, the resource allocation and scheduling of the targets in the very high uncertainty maintenance phase target set VHMS; S14, according to the trigger source affecting the collaborative observation resource allocation, the resource allocation and scheduling of the targets in the high uncertainty maintenance phase target set MTS; S15. According to the trigger source that affects the allocation of collaborative observation resources, resources are allocated and scheduled for the targets in the target set HOS in the low uncertainty handover phase.

3. The near-space multi-target large-scale constellation collaborative observation method according to claim 2 is characterized in that Step S1 The trigger sources that affect the allocation of collaborative observation resources include: the appearance of new targets, the completion of target phase tasks, the triggering of target cycles in the high uncertainty maintenance phase, and the change of target maneuvering state.

4. The near-space multi-target large-scale constellation collaborative observation method according to claim 3 is characterized in that Step S12 specifically includes the following sub-steps: S121, dynamically prioritize the targets in the target set TIS in the type identification stage to obtain TIS sorted_p , and initialize the occupied resource set UAS; S122, yes Each target in the algorithm is allocated observation resources based on the observation resource allocation algorithm in the single-step feedback type identification phase; S123. Add the allocated observation resources to the occupied resource set Update t at the same time shce +1 time planning S124, repeat steps S122 to S123 until TIS sorted_p All goals in the resource allocation are completed; S125, if the target TIS j If the corresponding observation resource allocation and scheduling are insufficient, the target TIS j Add the very high uncertainty maintenance phase target set VHMS.

5. The near-space multi-target large-scale constellation collaborative observation method according to claim 4 is characterized in that Step S13 specifically includes the following sub-steps: S131, dynamically prioritize the targets in the very high uncertainty maintenance stage target set VHMS to obtain VHMS sorted_p , and initialize the occupied resource set UAS; S132, yes Observation resources are allocated to each target using the optimal single-star observation strategy; S133. Add the allocated observation resources to the occupied resource set Update t at the same time shce +1 time planning S134, repeat steps S132 to S133 until VHMS sorted_p Complete resource allocation for all objectives.

6. A near-space multi-target large-scale constellation collaborative observation method according to claim 5, characterized in that Step S14 specifically includes the following sub-steps: S141, determine whether the observation resources in the MTS observation planning scheme for the high uncertainty maintenance stage target set are consistent with t shce +1 time, the target set TIS of the type identification stage and the target set VHMS of the very high uncertainty maintenance stage conflict, and the observation period Δt of the high uncertainty maintenance stage M_MT Is it finished? If not, go to step S15; S142, dynamically prioritize the targets in the high uncertainty maintenance stage target set MTS to obtain MTS sorted_p , and initialize the occupied resource set UAS; S143, yes Each target in the algorithm is allocated observation resources by maintaining cooperative observation algorithm based on intention; S144. Add all allocated observation resources in the observation period of the high uncertainty maintenance phase to the occupied resource set Update t at the same time shce +1 to t shce +Δt M_MT Time planning plan S145, repeat steps S143 to S144 until VHMS sorted_p Complete resource allocation for all objectives.

7. The near-space multi-target large-scale constellation collaborative observation method according to claim 6 is characterized in that Step S15 specifically includes the following sub-steps: S151, dynamically prioritize the targets in the low uncertainty handover stage target set HOS to obtain HOS sorted_p , and initialize the occupied resource set UAS; S152, yes Each target in the algorithm executes a learning-based low-uncertainty multi-target collaborative observation task planning algorithm to allocate observation resources; S153. Add all allocated observation resources in the observation period of the low uncertainty handover phase to the occupied resource set Update t at the same time shce +1 to t shce +Δt M_MT Time planning plan S154, repeat steps S153 to S154 until HOS sorted_p Complete resource allocation for all objectives.

8. The near-space multi-target large-scale constellation collaborative observation method according to claim 7, characterized in that Step S2 specifically includes the following sub-steps: S21, in the task preprocessing phase, the stable task interval division is calculated according to the initial trajectory and constellation information of the target, and the output result is used as the input data of the loop scheduling phase; S22. In the cyclic scheduling link, an observation pair decision network based on the proximal policy optimization reinforcement learning method is used to schedule resources for time-sensitive, low-uncertainty multi-target collaborative observation tasks through a time-driven decision-making process, and to allocate observation resources to all targets in each sub-task interval.

9. The near-space multi-target large-scale constellation collaborative observation method according to claim 8, characterized in that Step S21 specifically includes the following sub-steps: S21, initial visibility calculation, based on the initial predicted input trajectory and constellation information of the target, considering the detection distance and angle constraints of each satellite in the constellation, calculates the visible time window of each target with all observation satellites during its life cycle; S22. Calculate the maximum stable time period of each target resource according to the visible time window; S23, further divide the task intervals, combine the observed stable interval set obtained in step S22 with the planned task start and end time to obtain a task interval partition set, perform interval length check on the task interval partition set, perform secondary partition for those with lengths greater than mTD, and finally obtain a task interval set.

10. The near-space multi-target large-scale constellation collaborative observation method according to claim 9, characterized in that Step S22: The observation-to-decision network based on the proximal policy optimization reinforcement learning method includes an outer loop and an inner loop, wherein the outer loop represents a set of task intervals obtained by task preprocessing, and the inner loop is the observation-to-decision network.

Images