CN-121980908-A - Space-sky target collaborative observation-oriented star group intelligent task planning method and system

CN121980908ACN 121980908 ACN121980908 ACN 121980908ACN-121980908-A

Abstract

Aiming at the challenges of high target moving speed, high environment uncertainty, multiple observation resources, high calling difficulty and the like of star group task planning of aerospace high-speed target collaborative observation, the invention provides an intelligent task planning method and system for aerospace high-speed targets. The invention realizes the efficient planning and scheduling of the large-scale remote sensing star group in the environment with high dynamic property and strong uncertainty.

Inventors

LIAO YUAN
CHENG KAN
LU ZHENG
JIN HAO
ZHAO YUTING
DONG SHAOPENG
YIN KAIFENG
XIAO FAN
Zou Yafang

Assignees

中国空间技术研究院

Dates

Publication Date: 20260505
Application Date: 20251212

Claims (8)

1. A space-sky-oriented high-speed target collaborative observation-oriented star group intelligent task planning method is characterized by comprising the following steps: Modeling a target observation task scene as a Markov decision process; training a strategy network and an evaluation network by using a near-end gradient descent algorithm; And carrying out reasoning decision by using the trained strategy network.
2. The space-sky-oriented high-speed target collaborative observation-oriented star group intelligent task planning method according to claim 1, wherein modeling a target observation task scene as a markov decision process comprises: Constructing a state space according to a task scene facing the target collaborative observation, wherein the state space comprises visible window information of a satellite to a target, a target observation priority, load maneuvering time, the satellite and a target number; setting an action to only output a planning result of one satellite to one target, and outputting a final multi-satellite and multi-target planning result by repeating the planning result for a plurality of times; setting a reward function, and calculating a reward value according to an index of a target observation priority for a single-step planning result output by a strategy network; After the strategy network outputs a single-step planning result each time, the state transfer function updates the visible window information with conflict in the current state according to the current state and the single-step planning result and outputs the updated state according to constraint conditions, wherein the constraint conditions comprise actual load occupation, observation time and load maneuvering time; and setting a discount factor, wherein the discount factor is used for controlling the importance degree of future rewards in the learning process.
3. The space-sky-oriented high-speed target collaborative observation-oriented star group intelligent task planning method according to claim 2, wherein the training of the task planning strategy network and the evaluation network by using the near-end gradient optimization algorithm comprises the following steps: And selecting the architecture of a strategy network and an evaluation network, and constructing the strategy network and the evaluation network by using a transducer neural network architecture, wherein the strategy network output end adopts a probability mask mechanism to restrict invalid action space, and the evaluation network constructs global value estimation through time sequence differential learning.
4. The space-sky-oriented high-speed target collaborative observation-oriented star group intelligent task planning method according to claim 3, wherein the utilizing the strategy network to make an inference decision comprises: After the training of the strategy network and the evaluation network is completed and convergence is realized, predicting the positions of the satellite and the target in a future set planning period according to the positioning orbit determination data of the satellite in the constellation and the position information of the target; according to the actual load parameters, calculating the visible window of each load of the satellite to the target; and encoding the visible window information and the target priority according to the input requirement of the strategy network, and inputting the strategy network to generate a task planning result.
5. A space-sky-oriented high-speed target collaborative observation-oriented star group intelligent task planning system is characterized by comprising: the first module is used for modeling the target observation task scene as a Markov decision process; the second module is used for training the strategy network and the evaluation network by utilizing a near-end gradient descent algorithm; and the third module is used for carrying out reasoning decision by utilizing the trained strategy network.
6. The space-sky-oriented high-speed target collaborative observation-oriented star group intelligent mission planning system of claim 5, wherein modeling the target observation mission scenario as a markov decision process comprises: Constructing a state space according to a task scene facing the target collaborative observation, wherein the state space comprises visible window information of a satellite to a target, a target observation priority, load maneuvering time, the satellite and a target number; setting an action to only output a planning result of one satellite to one target, and outputting a final multi-satellite and multi-target planning result by repeating the planning result for a plurality of times; setting a reward function, and calculating a reward value according to an index of a target observation priority for a single-step planning result output by a strategy network; After the strategy network outputs a single-step planning result each time, the state transfer function updates the visible window information with conflict in the current state according to the current state and the single-step planning result and outputs the updated state according to constraint conditions, wherein the constraint conditions comprise actual load occupation, observation time and load maneuvering time; and setting a discount factor, wherein the discount factor is used for controlling the importance degree of future rewards in the learning process.
7. The space-sky-oriented high-speed target collaborative observation-oriented star group intelligent task planning system according to claim 6, wherein the training of the task planning strategy network and the evaluation network by using the near-end gradient optimization algorithm comprises the following steps: And selecting the architecture of a strategy network and an evaluation network, and constructing the strategy network and the evaluation network by using a transducer neural network architecture, wherein the strategy network output end adopts a probability mask mechanism to restrict invalid action space, and the evaluation network constructs global value estimation through time sequence differential learning.
8. The space-sky-oriented high-speed target collaborative observation-oriented star group intelligent mission planning system as described in claim 7, wherein said utilizing a strategy network to make an inference decision comprises: After the training of the strategy network and the evaluation network is completed and convergence is realized, predicting the positions of the satellite and the target in a future set planning period according to the positioning orbit determination data of the satellite in the constellation and the position information of the target; according to the actual load parameters, calculating the visible window of each load of the satellite to the target; and encoding the visible window information and the target priority according to the input requirement of the strategy network, and inputting the strategy network to generate a task planning result.

Description

Space-sky target collaborative observation-oriented star group intelligent task planning method and system Technical Field The invention relates to an aerospace high-speed target collaborative observation-oriented star group intelligent task planning method and system, and belongs to the technical field of remote sensing satellites. Background The significant reduction in satellite manufacturing costs has prompted the continued expansion of constellation sizes, with the number of satellites that need to be coordinated in a single mission having grown in a cross-magnitude. Meanwhile, modern satellites are commonly loaded with multi-mode loads and have autonomous mobility, so that observation tasks are independently executed from single equipment to multi-satellite cooperative networking evolution is realized. The traditional task planning method is based on a single-satellite independent decision mechanism, and is difficult to effectively process complex space-time constraint relations among satellite groups, and is characterized in that the heterogeneous satellite resource matching efficiency is low, the task conflict resolution capability is insufficient, and the dynamic environment response is lagged. Especially when dealing with sudden observation demands, the existing planning system is easy to have the problems of collaborative observation broken links, repeated consumption of resources and the like, and severely restricts the overall efficiency of the star group. This is an urgent need to construct intelligent planning systems with autonomous collaboration capabilities. On the other hand, unlike the traditional observation of ground targets, the aerospace high-speed moving targets comprise ballistic missiles, hypersonic weapons and the like, have strong maneuverability and high movement speed, and need multi-satellite cooperative observation for positioning and identification. Once the target is not effectively observed and tracked, the target motion state monitoring result diverges from the real motion state. The conventional satellite task planning method mostly adopts a periodic static scheduling strategy, and the planning timeliness and the re-planning frequency of the periodic static scheduling strategy are difficult to match with the dynamic change rhythm of a high-speed target. In addition, the multi-star relay observation requires each node to complete observation right switching and situation synchronization in a very short time, and the precision requirement of far exceeding the earth observation scene is provided for inter-star cooperative control. From the task background, the traditional star task planning method is concentrated on observing static targets on the ground, and allocates observation resources to point targets and regional targets to be observed, and the core task is to periodically allocate the observation resources to the point targets or the regional targets with known geographic coordinates. In contrast, the observation of an aerospace high-speed target needs to construct a parallax angle through double-star collaborative observation to calculate the elevation information in real time, and meanwhile, because the target is fast in moving speed and strong in maneuverability, a multi-star relay tracking mechanism needs to be established to maintain the continuity of a state observation chain. In terms of a planning method, the conventional star task planning method is mainly based on a centralized heuristic method, such as a task planning method based on a genetic algorithm, a particle swarm optimization algorithm and a simulated annealing algorithm, and the method is essentially that a suboptimal solution is found in a discrete solution space through a preset rule, so that the timeliness requirement on high-speed target observation cannot be met. Disclosure of Invention The invention aims at solving the technical problems that the space-sky high-speed target collaborative observation star group task planning faces the challenges of high target moving speed, strong environment uncertainty, multiple observation resources, large calling difficulty and the like. The technical scheme adopted by the invention is that the space-sky high-speed target collaborative observation-oriented star group intelligent task planning method comprises the following steps: Modeling a target observation task scene as a Markov decision process; training a strategy network and an evaluation network by using a near-end gradient descent algorithm; And carrying out reasoning decision by using the trained strategy network. Further, the modeling the target observation task scene as a markov decision process includes: Constructing a state space according to a task scene facing the target collaborative observation, wherein the state space comprises visible window information of a satellite to a target, a target observation priority, load maneuvering time, the satellite and a target number; setting an actio