Search

CN-122001454-A - Multi-mode constraint-based constellation planning method and system

CN122001454ACN 122001454 ACN122001454 ACN 122001454ACN-122001454-A

Abstract

The invention relates to a constellation planning method and system based on multi-mode constraint, belongs to the technical field of information, and solves the problems of multi-mode constraint fragmentation and insufficient balancing capability of complex constellation planning in the prior art. The method comprises the steps of obtaining multi-mode constraint data, planning target data and multiple sets of candidate constellation parameters, analyzing the multi-mode constraint data, obtaining constraint semantic vectors by utilizing a constraint semantic encoder, encoding to obtain a constraint embedded matrix, encoding the planning target data and each set of candidate constellation parameters to obtain a target vector and multiple parameter vectors, fusing the constraint embedded matrix, the target vector and each parameter vector to obtain multiple fusion vectors, and calculating corresponding candidate satisfaction scores according to all the fusion vectors to determine the target constellation parameter sets. The invention realizes multi-mode constraint depth fusion, balance and no conflict, has high planning efficiency and meets the planning requirement of a complex constellation system.

Inventors

  • XIE BAOGUO
  • LIU YA

Assignees

  • 空天信息大学(筹)

Dates

Publication Date
20260508
Application Date
20260331

Claims (10)

  1. 1. A constellation programming method based on multi-modal constraints, comprising: Acquiring multi-mode constraint data, planning target data and a plurality of groups of candidate constellation parameters; Analyzing the multi-mode constraint data, and obtaining corresponding constraint semantic vectors by utilizing a constraint semantic encoder; Encoding the constraint semantic vector to obtain a corresponding constraint embedded matrix; Encoding the planning target data and each group of candidate constellation parameters to obtain a corresponding target vector and a plurality of parameter vectors; fusing the constraint embedding matrix, the target vector and each parameter vector to obtain a plurality of fusion vectors; And calculating candidate satisfaction degree scores of each fusion vector based on all the fusion vectors, and determining a target constellation parameter set.
  2. 2. The method of claim 1, wherein the multimodal constraint data includes a plurality of original constraint sub-items; Analyzing the multi-mode constraint data, and obtaining corresponding constraint semantic vectors by utilizing a constraint semantic encoder, wherein the method comprises the following steps: Determining the constraint type of each original constraint sub-item by using a semantic classification model; determining corresponding constraint intensity according to the constraint type of each original constraint sub-item; Based on each original constraint sub-item, the corresponding constraint type and constraint intensity, the corresponding constraint semantic vector is obtained by utilizing the constraint semantic encoder.
  3. 3. The method of claim 2, wherein obtaining the corresponding constraint semantic vector using the constraint semantic encoder based on each of the original constraint sub-items and the corresponding constraint type, constraint strength, comprises: Identifying the constraint format of each original constraint sub-item, and determining a corresponding specification constraint sub-item according to a corresponding normalization strategy; Based on a constraint-parameter semantic mapping dictionary, acquiring associated parameters corresponding to each canonical constraint sub-item; Based on each specification constraint sub-item and corresponding constraint type, constraint intensity and association parameters, converting the specification constraint sub-items into corresponding standard constraint sub-items according to a constraint semantic standardization protocol; based on each standard constraint sub-item, generating a corresponding constraint semantic vector by using the constraint semantic encoder.
  4. 4. A method according to claim 3, wherein generating, based on each of the standard constraint sub-items, a corresponding constraint semantic vector using the constraint semantic encoder, comprises: performing conflict investigation on all the standard constraint sub-items; if the standard constraint sub-items with conflicts exist, a conflict prompt is generated; And if the standard constraint sub-items with the conflict do not exist, generating corresponding constraint semantic vectors by utilizing the constraint semantic encoder based on each standard constraint sub-item.
  5. 5. The method of claim 2, wherein encoding the constraint semantic vector results in a corresponding constraint embedding matrix, comprising: Adding a type identifier for each constraint semantic vector according to the constraint type of each constraint semantic vector; And sorting all constraint semantic vectors added with type identifiers according to the constraint intensity of all constraint semantic vectors, and constructing the constraint embedding matrix.
  6. 6. The method of claim 2, wherein fusing the constraint embedded matrix, the target vector, and each of the parameter vectors to obtain a plurality of fused vectors comprises: Determining corresponding constraint intensity factors according to the constraint intensity of each constraint semantic vector; calculating a target matching degree factor corresponding to each constraint semantic vector according to each constraint semantic vector and each target vector; Correcting the constraint embedded matrix according to the constraint intensity factor and/or the target matching degree factor; And performing attention fusion on the corrected constraint embedding matrix, the target vector and each parameter vector to obtain a plurality of fusion vectors.
  7. 7. The method of claim 1, wherein calculating a candidate satisfaction score for each of the fusion vectors based on all of the fusion vectors, determining a set of target constellation parameters, comprises: calculating the candidate satisfaction score corresponding to each fusion vector by using a scoring head based on the fusion vector; And determining a target fusion vector according to the candidate satisfaction scores of all the fusion vectors, wherein a single set of candidate constellation parameters corresponding to the target fusion vector are used as the target constellation parameter set.
  8. 8. The method of claim 7, wherein each set of the candidate constellation parameters comprises a plurality of constellation parameters; Determining a target fusion vector according to the candidate satisfaction scores of all the fusion vectors, including: selecting a standby fusion vector according to the candidate satisfaction score, wherein a single group of candidate constellation parameters corresponding to the standby fusion vector are used as a standby constellation parameter group; Determining a corresponding association constraint set for each of the constellation parameters of the backup set of constellation parameters; Calculating association constraint scores corresponding to the constellation parameters based on the association constraint sets, and judging whether the association constraint scores reach an association constraint threshold value or not; And if all the association constraint scores of the backup constellation parameter sets reach the corresponding association constraint thresholds, determining the corresponding backup fusion vector as the target fusion vector.
  9. 9. The method of claim 1, wherein the constrained semantic encoder is trained by: Acquiring a constellation planning constraint corpus, and converting the constellation planning constraint corpus into a corresponding standardized constraint sample set; Constructing a training sample pair according to the standardized constraint sample set, and performing contrast learning training on the constraint semantic encoder; acquiring constellation scene data, wherein the constellation scene data comprises sample constraint data and a known constellation planning scheme; and performing fine tuning training on the trained constraint semantic encoder according to the constellation scene data to obtain the trained constraint semantic encoder.
  10. 10. A multi-modal constraint-based constellation planning system comprising: An input layer for acquiring multi-modal constraint data, planning target data and a plurality of groups of candidate constellation parameters; the constraint processing layer is used for analyzing the multi-mode constraint data and obtaining corresponding constraint semantic vectors by utilizing a constraint semantic encoder; the coding layer is used for coding the constraint semantic vector to obtain a corresponding constraint embedded matrix; the coding layer is further configured to code the planning target data and each set of candidate constellation parameters to obtain a corresponding target vector and a plurality of parameter vectors; the fusion layer is used for fusing the constraint embedding matrix, the target vector and each parameter vector to obtain a plurality of fusion vectors; And the generation layer is used for calculating the corresponding candidate satisfaction degree scores according to all the fusion vectors and determining a target constellation parameter set.

Description

Multi-mode constraint-based constellation planning method and system Technical Field The invention relates to the technical field of information, in particular to a constellation planning method and system based on multi-mode constraint. Background The constellation planning is a core link of satellite communication system construction, and multi-dimensional conditions such as orbit dynamics law, satellite hardware performance limit, business service quality requirement, space resource rule constraint and the like are comprehensively considered to generate a planning scheme with engineering feasibility, business suitability and compliance. With the development of complex constellation systems such as low-orbit broadband constellation, multi-orbit mixed constellation and the like, the existing constellation planning technology gradually exposes the following prominent defects: 1) Constraint modeling fragmentation, associated constraint verification missing. In the prior art, multi-mode constraints such as orbit dynamics, hardware physics, business requirements, resource rules and the like are dispersed in different algorithm modules, and a mode of 'independent modeling and later verification' is adopted, so that association relations among the constraints are split. For example, the orbit height parameter adjustment only considers the atmospheric resistance constraint, the associated constraint such as satellite power consumption, beam coverage and the like is not checked synchronously, the problem that single constraint is met but multiple associated constraint is violated easily occurs, the conflict can be corrected through 5-8 iterations, and the constraint satisfaction rate is only 65-75%. 2) The multi-objective balance capability is weak, and the engineering suitability is insufficient. In the prior art, a simple method such as weighted summation is adopted to solve the problem of multi-objective optimization, and multiple core objectives such as coverage efficiency, time delay, cost, compliance and the like are difficult to balance. For example, the number of satellites is increased to pursue coverage maximization, resulting in transmission cost exceeding budget, or the excessively low orbit height is selected to reduce time delay, violating satellite hardware life constraint, and the project availability of the final planning scheme is less than 80%, thus requiring additional investment of a large amount of resources for secondary correction. 3) The multi-mode data has poor suitability and semantic understanding, wherein the constellation planning involves multi-mode constraint data such as numerical type (such as upper limit of power consumption), text type (such as ITU rule), formula type (such as orbit perturbation equation), space geographic type (such as user density thermodynamic diagram) and the like, and the prior art lacks a unified semantic analysis and standardized conversion mechanism, so that the model cannot accurately understand the physical essence and priority of constraint, and parameter matching can be performed only based on surface characteristics, thereby further reducing the rationality and constraint satisfaction rate of the planning scheme. Therefore, a constellation planning method and system based on multi-mode constraint are needed, the multi-mode constraint depth fusion, balance and no conflict are achieved, the planning efficiency is high, and the planning requirement of a complex constellation system is met. Disclosure of Invention In view of the above analysis, the embodiment of the invention aims to provide a constellation planning method and system based on multi-mode constraint, which are used for solving the problems of multi-mode constraint fragmentation and insufficient balancing capability of complex constellation planning in the prior art. In one aspect, an embodiment of the present invention provides a constellation planning method based on multi-modal constraint, including: Acquiring multi-mode constraint data, planning target data and a plurality of groups of candidate constellation parameters; Analyzing the multi-mode constraint data, and obtaining corresponding constraint semantic vectors by utilizing a constraint semantic encoder; Encoding the constraint semantic vector to obtain a corresponding constraint embedded matrix; Encoding the planning target data and each group of candidate constellation parameters to obtain a corresponding target vector and a plurality of parameter vectors; fusing the constraint embedding matrix, the target vector and each parameter vector to obtain a plurality of fusion vectors; And calculating candidate satisfaction degree scores of each fusion vector based on all the fusion vectors, and determining a target constellation parameter set. Further, the multimodal constraint data includes a plurality of original constraint sub-items; Analyzing the multi-mode constraint data, and obtaining corresponding constraint semantic vectors