Frequency spectrum prediction method based on time sequence knowledge graph under satellite-ground fusion network

Abstract

The invention relates to a frequency spectrum prediction method based on a time sequence knowledge graph under a satellite-ground convergence network, and belongs to the field of wireless communication. The method comprises the following steps: S1, constructing a time sequence knowledge graph of the satellite-ground fusion network; s2, discretizing the time information to form a time interval set; s3, performing neighbor aggregation on nodes in the graph based on a structure attention mechanism to obtain structure-enhanced node embedding representation; s4, constructing a time hyperplane set, and calculating the weight of each time hyperplane based on a time attention mechanism; s5, performing weighted fusion on the plurality of time hyperplanes to generate a time attention weighted hyperplane; s6, projecting the structure enhanced node representation to a time attention weighted hyperplane to perform spectrum relation modeling; and S7, outputting a spectrum occupancy prediction result in the target time period. According to the method, the time sequence knowledge graph of the satellite-ground fusion network is constructed, and a structure attention mechanism and a time attention hyperplane mechanism are introduced, so that accurate prediction of the spectrum occupancy state is realized, a time interval with key influence on spectrum prediction can be highlighted, and the accuracy and reliability of spectrum prediction in the satellite-ground fusion network are improved; and technical support is provided for spectrum resource scheduling and sharing.

Description

Technical Field

[0001] This invention belongs to the field of wireless communication and relates to a spectrum prediction method based on time-series knowledge graph in a satellite-ground fusion network. Background Technology

[0002] With the rapid development of low-Earth orbit satellite communication systems and terrestrial cellular networks, space-ground converged networks are gradually becoming an important form of future communication networks. Space-ground converged networks can achieve global communication coverage through the collaborative work of satellite and terrestrial communication systems. However, the allocation and sharing of spectrum resources in space-ground converged networks face many challenges.

[0003] On the one hand, due to satellite orbital motion, dynamic changes in the distribution of ground users, and the diversity of service demands, spectrum occupancy status exhibits obvious temporal correlation and non-stationarity; on the other hand, there are complex spatial and service relationships among different satellites, ground stations, and frequency bands, and traditional methods based on statistical models or simple time series models are difficult to simultaneously characterize these complex relationships.

[0004] Existing spectrum prediction methods mostly focus on time series modeling, ignoring the structural relationships between entities in satellite-ground integrated networks; while knowledge graph-based methods can represent entity relationships, most fail to effectively incorporate temporal evolution characteristics or treat different times as equal, failing to highlight the importance of key time intervals for spectrum prediction.

[0005] Therefore, there is an urgent need for a spectrum prediction method that can simultaneously model the entity relationship structure and temporal evolution characteristics in the space-ground fusion network and highlight key temporal information, so as to improve the accuracy and reliability of spectrum prediction. Summary of the Invention

[0006] In view of the above problems, this invention proposes a spectrum prediction method based on temporal knowledge graph in a space-ground fusion network. This method constructs a temporal knowledge graph for the space-ground fusion network, introduces a structural attention mechanism to model the relationships between entities' neighbors, and further introduces a temporal attention hyperplane mechanism to assign different weights to different time intervals, thereby achieving high-precision prediction of spectrum occupancy status.

[0007] To achieve the above objectives, the present invention adopts the following technical solution.

[0008] A spectrum prediction method based on temporal knowledge graph in a space-ground fusion network is characterized by the following steps:

[0009] S1: Construct a temporal knowledge graph for the space-ground integrated network;

[0010] S2: Discretize the time information to form a set of time intervals;

[0011] S3: Based on the structural attention mechanism, the nodes in the graph are aggregated by their neighbors to obtain a structurally enhanced node embedding representation;

[0012] S4: Construct a set of temporal hyperplanes and calculate the weights of each temporal hyperplane based on the temporal attention mechanism;

[0013] S5: Weighted fusion of multiple temporal hyperplanes to generate a temporally attention-weighted hyperplane;

[0014] S6: Project the structure-enhanced node representation onto a time-attention-weighted hyperplane to model spectral relationships;

[0015] S7: Output the spectrum occupancy prediction results for the target time period.

[0016] Furthermore, in step S1, the spectrum sharing problem in the space-ground fusion network is modeled as a temporal knowledge graph, which is defined as follows: ,in Represents a set of entity nodes. Represents a set of relations. Represents a set of time information; a set of entity nodes. Including but not limited to satellite entity nodes, ground station entity nodes, and frequency band entity nodes; the relationships in the diagram are represented in quadruple form as follows: ,in The head entity represents the entity that initiates the spectrum relationship. The relation type (Relation) indicates spectrum occupancy, sharing, or interference relationships. The tail entity represents the receiving entity in the spectrum relationship. Let be the time interval in which the relationship is established. The start time, This is the end time.

[0017] Furthermore, in step S2, the original continuous timestamps are discretized; let the original timestamp set be: Count the frequency of occurrence of each timestamp: ,in Represents timestamp The frequency of occurrence It is an indicator function; passed through a frequency threshold. Group the timestamps to form a set of time intervals: ,in Indicates the first A discrete time interval This indicates the total number of time intervals.

[0018] Furthermore, in step S3, structural attention neighbor aggregation is performed on the entity nodes; let: For head entity The initial embedding vector, Tail entity The initial embedding vector, For relationship Embedded vector, Define the entity for the embedded vector dimension. The neighbor set is:

[0019] ;

[0020] Calculate the structure attention score:

[0021] ;

[0022] in It is a linear transformation matrix. For attention parameter vectors, This is a vector concatenation operation; normalization yields the structural attention weights:

[0023] ;

[0024] The structure-enhanced entity representation is defined as follows:

[0025] ;

[0026] in This represents a structure-enhanced entity embedding representation.

[0027] Furthermore, in step S4, for each time interval Construct a time hyperplane; define Time interval The corresponding hyperplane normal vector, and it satisfies the normalization constraint. Construct a joint representation of triples: Calculate the time attention score: ,in Represents the temporal attention mapping matrix; the temporal attention weights are calculated using the Softmax function:

[0028] ;

[0029] in Indicates time interval Attention weights =1.

[0030] Furthermore, in step S5, multiple temporal hyperplanes are weighted and fused to obtain a temporally attention-weighted hyperplane: Then normalize it: ,in This represents the time attention-weighted hyperplane normal vector ultimately used for projection.

[0031] Furthermore, in step S6, the structure-enhanced entity representation and relation representation are projected onto the attention-weighted hyperplane; the projection operator is defined: Calculate the projection representation of the head entity respectively Relational projection representation Tail real projection representation Construct a spectral relationship scoring function:

[0032] .

[0033] Furthermore, in step S7, the spectrum prediction result is output according to the scoring function; the score is mapped to probability using the Sigmoid function. ,in This represents the probability of spectrum occupancy within the target time period. This is the Sigmoid function.

[0034] Compared with existing technologies, the advantages of this invention are: it can simultaneously model the structural relationships and temporal evolution characteristics in satellite-ground fusion networks, thereby improving the accuracy of spectrum prediction; by introducing a time attention mechanism to highlight the importance of key time intervals, it is suitable for scenarios where spectrum occupancy is non-stationary; the structural attention mechanism improves the ability of entity embedding to express key neighbor relationships; and it is applicable to application scenarios such as spectrum sharing, interference prediction, and resource scheduling in satellite-ground fusion networks. Attached Figure Description

[0035] To make the objectives, technical solutions, and beneficial effects of this invention clearer, the following figures are provided for illustration:

[0036] Figure 1 This is an overall flowchart of the spectrum prediction method described in the embodiments of the present invention;

[0037] Figure 2 This is a diagram illustrating the model structure based on structural attention and temporal attention as described in an embodiment of the present invention.

[0038] Figure 3 This is a schematic diagram of the construction of the time attention-weighted hyperplane according to an embodiment of the present invention. Detailed Implementation

[0039] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0040] This invention addresses the challenges of complex spectrum occupancy states and significant temporal evolution in space-ground fusion networks by proposing a spectrum prediction method based on a temporal knowledge graph. First, a temporal knowledge graph of the space-ground fusion network is constructed using historical spectrum data. Second, a structural attention mechanism is introduced to model the neighbor relationships of entity nodes. Further, a time attention hyperplane mechanism is used to perform weighted modeling of different time intervals. Finally, spectrum relationships are predicted in a time-aware semantic space.

[0041] Through the above methods, the present invention can effectively improve the accuracy and stability of spectrum prediction in satellite-ground integrated networks, and provide technical support for the rational allocation and scheduling of spectrum resources.

[0042] Finally, 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 scope of protection of the present invention. Those skilled in the art can make various modifications and equivalent substitutions to the present invention without departing from its spirit and substance, and all such modifications and substitutions should fall within the scope of protection of the present invention.

Claims

1. A spectrum prediction method based on temporal knowledge graph in a space-ground fusion network, characterized in that: The method includes the following steps: S1: Construct a temporal knowledge graph for the space-ground integrated network; S2: Discretize the time information to form a set of time intervals; S3: Based on the structural attention mechanism, the nodes in the graph are aggregated by their neighbors to obtain a structurally enhanced node embedding representation; S4: Construct a set of temporal hyperplanes and calculate the weights of each temporal hyperplane based on the temporal attention mechanism; S5: Weighted fusion of multiple temporal hyperplanes to generate a temporally attention-weighted hyperplane; S6: Project the structure-enhanced node representation onto a time-attention-weighted hyperplane to model spectral relationships; S7: Output the spectrum occupancy prediction results for the target time period.

2. The spectrum prediction method based on temporal knowledge graph in a satellite-ground fusion network according to claim 1, characterized in that: In step S1, the spectrum sharing problem in the space-ground fusion network is modeled as a temporal knowledge graph, which is defined as follows: ,in Represents a set of entity nodes. Represents a set of relations. Represents a set of time information; a set of entity nodes. Including but not limited to satellite entity nodes, ground station entity nodes, and frequency band entity nodes; the relationships in the diagram are represented in quadruple form as follows: ,in The head entity represents the entity that initiates the spectrum relationship. The relation type (Relation) indicates spectrum occupancy, sharing, or interference relationships. The tail entity represents the receiving entity in the spectrum relationship. Let be the time interval in which the relationship is established. The start time, This is the end time.

3. The spectrum prediction method based on temporal knowledge graph in a satellite-ground fusion network according to claim 1, characterized in that: In step S2, the original continuous timestamps are discretized; let the original timestamp set be: Count the frequency of occurrence of each timestamp: ,in Represents timestamp The frequency of occurrence It is an indicator function; passed through a frequency threshold. Group the timestamps to form a set of time intervals: ,in Indicates the first A discrete time interval This indicates the total number of time intervals.

4. The spectrum prediction method based on temporal knowledge graph in a satellite-ground fusion network according to claim 1, characterized in that: In step S3, structural attention neighbor aggregation is performed on the entity nodes; set up : For head entity The initial embedding vector, Tail entity The initial embedding vector, For relationship Embedded vector, Define the entity for the embedded vector dimension. The neighbor set is: ； Calculate the structure attention score: ； in It is a linear transformation matrix. For attention parameter vectors, This is a vector concatenation operation; Normalization yields the structural attention weights: ； The structure-enhanced entity representation is defined as follows: ； in This represents a structure-enhanced entity embedding representation.

5. The spectrum prediction method based on temporal knowledge graph in a satellite-ground fusion network according to claim 1, characterized in that: In step S4, for each time interval Construct a time hyperplane; define Time interval The corresponding hyperplane normal vector, and it satisfies the normalization constraint. Construct a joint representation of triples: Calculate the time attention score: ,in Represents the temporal attention mapping matrix; the temporal attention weights are calculated using the Softmax function: ； in Indicates time interval Attention weights =1.

6. The spectrum prediction method based on temporal knowledge graph in a satellite-ground fusion network according to claim 1, characterized in that: In step S5, multiple temporal hyperplanes are weighted and fused to obtain a temporally attention-weighted hyperplane: Then normalize it: ,in This represents the time attention-weighted hyperplane normal vector ultimately used for projection.

7. The spectrum prediction method based on temporal knowledge graph in a satellite-ground fusion network according to claim 1, characterized in that: In step S6, the structure-enhanced entity representation and relation representation are projected onto the attention-weighted hyperplane; the projection operator is defined: Calculate the projection representation of the head entity respectively Relational projection representation Tail real projection representation Construct a spectral relationship scoring function: 。 8. The spectrum prediction method based on temporal knowledge graph in a satellite-ground fusion network according to claim 1, characterized in that: In step S7, the spectrum prediction result is output according to the scoring function; the score is mapped to probability using the Sigmoid function. ,in This represents the probability of spectrum occupancy within the target time period. This is the Sigmoid function.

Images