Event prediction method based on satellite orbit threat domain knowledge graph

A computer-readable storage medium, wherein the storage medium stores a computer program, and the computer program, when executed by one or more processors, implements a method comprising:

• acquiring a database of a satellite orbit threat domain;
• by respectively using entities stored in the database as nodes and using an incidence relationship between each entity as an edge, constructing a knowledge graph of the satellite orbit threat domain;
• wherein the entities comprise a satellite, a satellite event, and a motion orbit, and the incidence relationship comprises an event category and a motion state;
• extracting meta-paths from the knowledge graph;
• and for each of the meta-paths, determining, according to each entity linked by each edge contained in the meta-path, a meta-path having a same starting entity and/or a same ending entity as the meta-path from the meta-paths as a reference meta-path corresponding to the meta-path;
• determining, according to each entity linked by each edge contained in the meta-path and each entity linked by each edge contained in the reference meta-path corresponding to the meta-path, a plurality of entities having a related relationship and not connected by an edge;
• and constructing, according to the determined plurality of entities, a hyperedge;
• determining an entity feature corresponding to each entity through a feature extraction network in a to-be-trained prediction model;
• for each entity, constructing, according to respective entities that are linked by respective edges contained in a meta-path that an edge connected to the entity belongs to and respective entities contained in the hyperedge that the entity belongs to, a target tree structure with the entity as a root node, and aggregating entity features of respective entities contained in the target tree structure corresponding to the entity to determine a target feature of the entity;
• forming, according to any one satellite entity and any one satellite event entity in the entities, an entity-pair;
• determining, for each entity-pair, at least according to a target feature of a satellite entity in the entity-pair and a target feature of a satellite event entity in the entity-pair, an input feature of the entity-pair;
• obtaining, by inputting the input feature of the entity-pair into a prediction network in a to-be-trained prediction model, a predicted incidence relationship between the satellite entity and the satellite event entity in the entity-pair;
• and
• training, according to a difference between the incidence relationship between the satellite entity and the satellite event entity in the entity-pair in the knowledge graph and the predicted incidence relationship, the to-be-trained prediction model;
• and
• in response to a prediction request, determining a predicted incidence relationship between a to-be-predicted satellite and a to-be-predicted satellite event through the trained prediction model.

An electronic device, comprising a memory, one or more processors, and a computer program stored in the memory and executable on the one or more processors, wherein the one or more processors, when executing the program, implement a method comprising:

• acquiring a database of a satellite orbit threat domain;
• by respectively using entities stored in the database as nodes and using an incidence relationship between each entity as an edge, constructing a knowledge graph of the satellite orbit threat domain;
• wherein the entities comprise a satellite, a satellite event, and a motion orbit, and the incidence relationship comprises an event category and a motion state;
• extracting meta-paths from the knowledge graph;
• and for each of the meta-paths, determining, according to each entity linked by each edge contained in the meta-path, a meta-path having a same starting entity and/or a same ending entity as the meta-path from the meta-paths as a reference meta-path corresponding to the meta-path;
• determining, according to each entity linked by each edge contained in the meta-path and each entity linked by each edge contained in the reference meta-path corresponding to the meta-path, a plurality of entities having a related relationship and not connected by an edge;
• and constructing, according to the determined plurality of entities, a hyperedge;
• determining an entity feature corresponding to each entity through a feature extraction network in a to-be-trained prediction model;
• for each entity, constructing, according to respective entities that are linked by respective edges contained in a meta-path that an edge connected to the entity belongs to and respective entities contained in the hyperedge that the entity belongs to, a target tree structure with the entity as a root node, and aggregating entity features of respective entities contained in the target tree structure corresponding to the entity to determine a target feature of the entity;
• forming, according to any one satellite entity and any one satellite event entity in the entities, an entity-pair;
• determining, for each entity-pair, at least according to a target feature of a satellite entity in the entity-pair and a target feature of a satellite event entity in the entity-pair, an input feature of the entity-pair;
• obtaining, by inputting the input feature of the entity-pair into a prediction network in a to-be-trained prediction model, a predicted incidence relationship between the satellite entity and the satellite event entity in the entity-pair;
• and training, according to a difference between the incidence relationship between the satellite entity and the satellite event entity in the entity-pair in the knowledge graph and the predicted incidence relationship, the to-be-trained prediction model;
• and
• in response to a prediction request, determining a predicted incidence relationship between a to-be-predicted satellite and a to-be-predicted satellite event through the trained prediction model.