CN118605609A - A system and method for constructing drone swarm attack situation based on knowledge graph - Google Patents

Abstract

A knowledge-graph-based unmanned aerial vehicle swarm attack situation construction system and method belong to the technical field of data processing. In the system, a data selection module selects 1 target and N unmanned aerial vehicles from a dynamic knowledge graph according to a combat task at a time t, the N unmanned aerial vehicles form an unmanned aerial vehicle bee group, the characteristic vector of each unmanned aerial vehicle forming different bee groups is respectively input into each graph attention mechanism model of a graph attention mechanism model layer to generate a damage vector to the target, and a neural network generates a combat index according to the damage vector; and sequencing the combat indexes of different swarms, and selecting the optimal unmanned aerial vehicle swarm as the unmanned aerial vehicle swarm of the final attack target. The system and the method provided by the invention can select the best unmanned plane bee matched with the system to attack the target.

Description

A system and method for constructing drone swarm attack situation based on knowledge graph

Technology Leader

The present invention relates to a knowledge graph-based UAV swarm attack situation construction system and method, and belongs to the field of electrical data processing technology.

Background Art

A Chinese invention patent application with publication number CN115544801A discloses a method for demonstrating battlefield situation of drone swarm combat, the method comprising: before simulation, the situation demonstration platform retrieves the combat configuration module to generate a number of battlefield environment configuration files and simulation model configuration files according to combat requirements; and displays the situation of the corresponding battlefield facility model and the corresponding simulation model according to the battlefield environment configuration file and the simulation model configuration file; during the simulation process, the demonstration control module controls the reception of battlefield information data and simulation data of one or more simulation models, and sends the received data to the situation demonstration platform; the situation demonstration platform performs a dynamic situation display on the received battlefield information data; and performs a dynamic situation display on the operating status of the corresponding simulation model according to the received simulation data.

However, it did not report how to select the best drone bees for coordination.

Summary of the invention

The technical problem to be solved by the present invention is to propose a system and method for constructing a drone swarm attack situation based on a knowledge graph, which can select the best drone swarm for coordination.

To achieve the above-mentioned purpose of the invention, the present invention provides a knowledge graph-based drone swarm attack situation construction system, which includes a dynamic knowledge graph generation module and an attack situation generation module, wherein the attack situation construction module includes a data selection module, N combat index item calculation modules and a sorting module, wherein the data selection module selects 1 target and N drones from the dynamic knowledge graph according to the combat mission at time t, and the N drones form a drone swarm, and inputs the feature vector of each drone attack target of the first drone swarm into the graph attention mechanism model layer of the first calculation module; the data selection module also selects another drone from the dynamic knowledge graph according to the combat mission at time t to replace one drone in the first drone swarm to form a second drone swarm, and inputs the feature vector of each drone attack target of the second drone swarm into The graph attention mechanism model layer of the second computing module; the data selection module also selects one more drone from the dynamic knowledge graph according to the combat mission at time t to replace one drone in the second drone swarm that has not replaced other drones to form the third drone swarm, and inputs the feature vector of each drone attack target of the third drone swarm into the graph attention mechanism model layer of the third computing module. By analogy, the data selection module selects one drone from the dynamic knowledge graph according to the combat mission at time t to replace the drone in the N-1th drone swarm that has not replaced other drones to form the Nth drone swarm, and inputs the feature vector of each drone attack target of the Nth drone swarm into the graph attention mechanism model layer of the Nth computing module. The drone members of the N drone swarms are all different, and N is a positive integer greater than or equal to 2;

Each combat index item calculation module includes a graph attention mechanism model layer and a neural network layer. The graph attention mechanism model layer includes N graph attention mechanism models. The nth graph attention mechanism model outputs a damage vector to the target at time t as c ⁿ _t , so that the output vector of the graph attention mechanism model layer is C=[c ¹ _t ,…,c ⁿ _t ,…,c ^N _t ]; the neural network generates K combat index items at time t according to the input vector C at time t, where K is a positive integer greater than or equal to 2;

The sorting module conducts a comprehensive analysis on the K index items output by the N computing modules, sorts them according to the comprehensive analysis results, and selects the drone swarm with the best comprehensive results as the optimal coordinated drone swarm for the final attack target.

To achieve the above-mentioned purpose, the present invention also provides a method for constructing a drone swarm attack situation based on a knowledge graph, which comprises the following steps:

Step 1: The data selection module selects 1 target and N drones from the dynamic knowledge graph according to the combat mission at time t. The N drones form a drone swarm, and the feature vector of each drone attack target in the first drone swarm is input into the graph attention mechanism model layer of the first computing module.

Step 2: The data selection module also selects another drone from the dynamic knowledge graph according to the combat mission at time t to replace one drone in the first drone swarm to form the second drone swarm, and inputs the feature vector of each drone attack target of the second drone swarm into the graph attention mechanism model layer of the second computing module;

Step 3: The data selection module also selects another drone from the dynamic knowledge graph at time t according to the combat mission to replace one drone in the second drone swarm that has not replaced other drones to form the third drone swarm, and inputs the feature vector of each drone attack target of the third drone swarm into the graph attention mechanism model layer of the third computing module.

Step 4: By analogy, at time t, the data selection module selects one drone from the dynamic knowledge graph according to the combat mission to replace the drones in the N-1th drone swarm that have not replaced other drones to form the Nth drone swarm, and inputs the feature vector of the attack target of each drone in the Nth drone swarm into the graph attention mechanism model layer of the Nth computing module. The drone members of the N drone swarms are all different, and N is a positive integer greater than or equal to 2;

Each combat index calculation module includes a graph attention mechanism model layer and a neural network layer. The graph attention mechanism model layer includes N graph attention mechanism models. The damage vector output by the nth graph attention mechanism model at time t is c ⁿ _t , so that the output vector of the graph attention mechanism model layer is C=[c ¹ _t ,…,c ⁿ _t ,…,c ^N _t ];

Step 5: Generate K combat index items at time t according to the input vector C at time t through a neural network, where K is a positive integer greater than or equal to 2;

Step 6: The sorting module conducts a comprehensive analysis on the K index items output by the N computing modules, sorts them according to the comprehensive analysis results, and selects the drone swarm with the best comprehensive results as the optimal coordinated drone swarm for the final attack target.

Beneficial Effects

Compared with the prior art, the system and method for constructing the drone swarm attack situation based on the knowledge graph provided by the present invention have the following beneficial effects:

The present invention can select the best coordinated drone bees to attack the target.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a block diagram of a system for constructing a drone swarm attack situation based on a knowledge graph according to a first embodiment of the present invention.

FIG2 is a block diagram of the composition of the dynamic knowledge graph generation module provided in the first embodiment of the present invention.

FIG. 3 is a block diagram of the attack situation generating module provided in the first embodiment of the present invention.

DETAILED DESCRIPTION

In order to make the technical problems, technical solutions and beneficial effects to be solved by the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

First embodiment

Figure 1 is a block diagram of the composition of the drone swarm attack situation construction system based on the knowledge graph provided by the first embodiment of the present invention. As shown in Figure 1, the drone swarm attack situation construction system based on the knowledge graph provided by the first embodiment of the present invention includes a dynamic knowledge graph generation module and an attack situation generation module. The dynamic knowledge graph construction module constructs a dynamic knowledge graph including multiple entities, and the entities include all drones participating in the battle, the targets that the drones need to attack, and the obstacles between the drones and the targets that need to be attacked, and the obstacles include interception equipment; the situation construction module selects 1 target and N drone swarms that attack the target from the constructed dynamic graph dynamic knowledge graph, evaluates the combat index items of the N drone swarms, and selects the best coordinated drone swarm as the swarm of the attack target according to the evaluation results, and N is a positive integer.

FIG2 is a block diagram of a dynamic knowledge graph generation module provided in the first embodiment of the present invention. As shown in FIG2 , the dynamic knowledge graph generation module provided in the first embodiment of the present invention implements the following process:

S01: clustering the existing entities into multiple groups to obtain a group set Cr = [Cr ₁ , ..., Cr _a , ..., Cr _A ], where A is a positive integer greater than or equal to 3, and the groups include a target group, a drone swarm, and an obstacle group;

S02: In each group Cr _a , a knowledge graph of the group Cr _a is constructed with entities in the group as nodes and relationships between nodes as edges;

S03: Connect related entities in two adjacent groups with edges to form an initial knowledge graph;

S04: Calculate the repulsion between the newly discovered entity v _new and each group Cr _a in the existing group set Cr. If it is repulsive with each group Cr _a in the group set Cr, execute S06; if there are non-repulsive groups, the non-repulsive group set is recorded as Cr _absorb = [Cr ₁ , ..., Cr _b , ..., Cr _B ], B is a positive integer greater than or equal to 1, then calculate the similarity Sim (v _NEW , Cr _b ) between the newly discovered entity v _new and each non-repulsive group in the non-repulsive group set Cr _absorb ;

S05 If the similarity Sim(v _NEW ,Cr _b ) is less than the threshold, execute S06; if the similarity Sim(v _NEW ,Cr _b ) is greater than or equal to the threshold, place the newly discovered entity V _NEW in the group with the largest similarity Sim(v _NEW ,Cr _b ), calculate the topological association between the newly discovered entity v _new and the existing entities in the group, and connect them through edges;

S06: Create a new group Cr _NEW according to the newly discovered entity v _NEW , and add the group Cr _NEW to the group set Cr;

S07: Repeat S04, S05 and S06 until all newly discovered entities v _NEW are clustered to form a dynamic knowledge graph.

The first embodiment of the present invention realizes updating with newly discovered entities through the above technical solution without requiring a large amount of computing power overhead and saving the construction time of the dynamic knowledge graph.

In the first embodiment, the sensing module includes a measurement module, an evaluation module and a quantification module. The measurement module is used to measure the distance between one entity and another entity in real time, measure the remaining ammunition of the drone entity, the remaining energy of the drone operation (such as oil volume, power volume, fuel volume, etc.), the operating status of the drone (such as speed, posture, etc.), the location of obstacles, the operating speed of obstacles, the model of obstacles, etc.; the evaluation module is used to evaluate the status of the drone operator, communication capabilities, etc. in real time; the quantification module quantifies the information provided by the measurement module and the evaluation module to facilitate data processing.

Figure 3 is a block diagram of the composition of the attack situation generation module provided by the first embodiment of the present invention; as shown in Figure 3, the attack situation construction module includes a data selection module, N combat index item calculation modules and a sorting module, wherein the data selection module selects 1 target and N drones from the dynamic knowledge graph according to the combat mission at time t, and the N drones form a drone swarm, and inputs the feature vector of each drone attack target of the first drone swarm into the graph attention mechanism model layer of the first calculation module; the data selection module also selects another drone from the dynamic knowledge graph according to the combat mission at time t to replace one drone in the first drone swarm to form a second drone swarm, and inputs the feature vector of each drone attack target of the second drone swarm into the graph attention mechanism model layer of the second calculation module. The data selection module also selects one more drone from the dynamic knowledge graph at time t according to the combat mission to replace one drone in the second drone swarm that has not replaced other drones to form the third drone swarm, and inputs the feature vector of each drone attack target of the third drone swarm into the graph attention mechanism model layer of the third computing module. By analogy, the data selection module selects one drone from the dynamic knowledge graph at time t according to the combat mission to replace the drone in the N-1th drone swarm that has not replaced other drones to form the Nth drone swarm, and inputs the feature vector of each drone attack target of the Nth drone swarm into the graph attention mechanism model layer of the Nth computing module. The drone members of the N drone swarms are all different, and N is a positive integer greater than or equal to 2.

Each combat index item calculation module includes a graph attention mechanism model layer and a neural network layer. The graph attention mechanism model layer includes N graph attention mechanism models. The nth graph attention mechanism model outputs a damage vector to the target at time t as c ⁿ _t , so that the output vector of the graph attention mechanism model layer is C=[c ¹ _t ,…,c ⁿ _t ,…,c ^N _t ]; the neural network generates K combat index items at time t according to the input vector C at time t, where K is a positive integer greater than or equal to 2;

The sorting module conducts a comprehensive analysis on the K index items output by the N computing modules, sorts them according to the comprehensive analysis results, and selects the drone swarm with the best comprehensive results as the optimal coordinated drone swarm for the final attack target.

The first embodiment of the present invention can select the best coordinated drone bees to attack the target through the above technical solution.

The destructive capability of each drone on its target is determined by the drone's own performance, the performance of its equipment, the skills of the operator, the distance from the target and the environment in which it is located. The destructive capability of a drone swarm on its target is also related to factors such as the communication capability and collaboration capability between drone swarms.

In the first embodiment, the data selection module includes a hard attention mechanism model, and the hard attention mechanism model selects drone entities from the dynamic knowledge graph according to the attack probability of the attack target, and sequentially selects drone entities with higher attack probability from the dynamic knowledge graph.

In the first embodiment, the characteristic vector of the drone includes the quantitative value of the drone performance, the quantitative value of the performance of the weapon equipped on the drone, the quantitative value of the skill of the personnel operating the drone, the quantitative value of the skill of the personnel operating the weapon, the quantitative value of the ammunition used for the weapon, the distance value from the target, the quantitative value of the personnel status, etc., that is, h _n = [a ¹ , a ² , …, a ^N ]; the damage vector of the attack target includes: the quantitative value of the ability to hit the target, the quantitative value of the ability to damage the target, the quantitative value of the ammunition consumption, the quantitative value of the time to complete the damage to the target, etc. The combat indicators include the hit rate, the target damage rate, the average ammunition consumption, the regional suppression rate and the assault mission completion time, etc., which are expressed as ].

In the first embodiment,

Where σ is the first activation function, is the degree of cooperation between the feature vector _hj of the j-th UAV attack target and the feature vector _hn of the n-th UAV attack target in the swarm; _enj is the feature vector of the edge connecting the j-th UAV and the n-th UAV; is the loss degree of the feature vector h _n of the n-th UAV attack target by the feature vector h _m of the m-th obstacle interception swarm among M obstacles; e _nm is the feature vector of the edge connecting the m-th obstacle and the n-th UAV; LeakyReLU is the second activation function; ρ is the parameter from the input layer to the hidden layer of the graph attention mechanism model; W _U , W _O , W _r , W _v represent parameter matrices; || means splicing them together.

The first embodiment takes into account the degree of cooperation of other drones in the swarm with the drone's attack target and the degree of damage caused by obstacles to the drone's attack target, thereby making the calculated combat index items more scientific.

In the first embodiment, the neural network includes an input layer, a hidden layer and an output layer. The input layer includes N neurons. The vector input to the nth neuron at time t is:

b _t ⁿ =w ⁿ c ⁿ _t

Where w ⁿ is the weighting matrix of the nth damage vector c ⁿ _t ;

The hidden layer includes a feedback layer and a comparison layer, wherein the feedback layer includes I neurons, I is a positive integer greater than or equal to N, and the output of the i-th neuron at time t is:

Where w ⁿⁱ is the weight between the nth neuron in the input layer and the i-th neuron in the feedback layer; _ut-1 ⁱ is the output of the i-th neuron at time t-1, α is the adjustment coefficient, and exp is the third activation function;

The comparison layer includes I neurons, and the output of the i-th neuron is:

In the formula, β is a constant;

The output layer includes K neurons, each of which outputs a combat indicator. The output of the kth neuron is:

Where w ^ki is the weight between the i-th neuron in the comparison layer and the k-th neuron in the output layer.

The first embodiment takes into account the impact of the drone's historical data on the current data through the above technical solution, so that the calculated combat index items are closer to the actual situation.

Second embodiment

The second embodiment of the present invention describes only the contents that are different from the first embodiment.

The second embodiment of the present invention provides a method for constructing a drone swarm attack situation based on a knowledge graph, which comprises the following steps:

Step 1: The data selection module selects 1 target and N drones from the dynamic knowledge graph according to the combat mission at time t. The N drones form a drone swarm, and the feature vector of each drone attack target in the first drone swarm is input into the graph attention mechanism model layer of the first computing module.

Step 2: The data selection module also selects another drone from the dynamic knowledge graph according to the combat mission at time t to replace one drone in the first drone swarm to form the second drone swarm, and inputs the feature vector of each drone attack target of the second drone swarm into the graph attention mechanism model layer of the second computing module;

Step 3: The data selection module also selects another drone from the dynamic knowledge graph at time t according to the combat mission to replace one drone in the second drone swarm that has not replaced other drones to form the third drone swarm, and inputs the feature vector of each drone attack target of the third drone swarm into the graph attention mechanism model layer of the third computing module.

Step 4: By analogy, at time t, the data selection module selects one drone from the dynamic knowledge graph according to the combat mission to replace the drones in the N-1th drone swarm that have not replaced other drones to form the Nth drone swarm, and inputs the feature vector of the attack target of each drone in the Nth drone swarm into the graph attention mechanism model layer of the Nth computing module. The drone members of the N drone swarms are all different, and N is a positive integer greater than or equal to 2;

Each combat index calculation module includes a graph attention mechanism model layer and a neural network layer. The graph attention mechanism model layer includes N graph attention mechanism models. The damage vector output by the nth graph attention mechanism model at time t is c ⁿ _t , so that the output vector of the graph attention mechanism model layer is C=[c ¹ _t ,…,c ⁿ _t ,…,c ^N _t ];

Step 5: Generate K combat index items at time t according to the input vector C at time t through a neural network, where K is a positive integer greater than or equal to 2;

Step 6: The sorting module conducts a comprehensive analysis on the K index items output by the N computing modules, sorts them according to the comprehensive analysis results, and selects the drone swarm with the best comprehensive results as the optimal coordinated drone swarm for the final attack target.

The attack situation generation module provided in the first embodiment is used to select drones and attack targets from the dynamic knowledge graph to conduct exercises, and compared with manually selecting drones and attack targets from the dynamic knowledge graph to conduct exercises, the combat effectiveness is compared in terms of hit rate, target damage rate, ammunition consumption, regional suppression rate, assault mission completion time, etc., as shown in the following table:

Indicator Human selection of drone swarms The attack situation generation module selects the drone swarm Hit Rate 60% 83% Target damage rate 53% 70% Ammunition consumption 70% 45% Regional suppression rate 35% 50% Operation completion time About 100 minutes About 45 minutes

.

It can be seen from the comparative exercises that the combat indicators achieved by the attack situation generation module constructed by the present invention through selecting drones and attack targets from the dynamic knowledge graph are better than the combat indicators achieved by manual rule control. Therefore, the drone swarm attack situation construction system and method based on the knowledge graph provided by the present invention can enable drone swarms to cooperate with each other, so as to achieve autonomous decision-making to maximize the combat effectiveness of the single unit, and on the other hand, coordinate attacks to achieve the global optimal strategy.

Third embodiment

The third embodiment of the present invention further provides a program product, which includes a computer program code, and the computer program code can be called by a processor to execute the method described in the second embodiment.

The beneficial effects of the program product provided by the third embodiment are the same as those of the first embodiment or the second embodiment, and will not be described again here.

Fourth embodiment

A fourth embodiment of the present invention provides a computer storage system, wherein the storage system stores computer program code, and when the computer program code is called and executed by a processor, the method described in the second embodiment is implemented. When the computer storage system stores the computer program code, it creates a mapping associated with a plurality of block identifiers having a plurality of binary combinations of data of a block size; according to the mapping associated with the plurality of block identifiers having the plurality of binary combinations of data of the block size: the processor generates a block identifier list representing a list of data blocks in a storage device, the processor sends the block identifier list to a backup storage device in a cloud-based storage environment to copy the list of data blocks in the storage device, and stores the mapping associated with the plurality of block identifiers on a computer-readable storage medium, wherein the plurality of block identifiers have the plurality of binary combinations of data of the block size and the list of block identifiers in the storage device.

The beneficial effects of the computer storage medium provided by the fourth embodiment are the same as those of the first embodiment or the second embodiment, and will not be described again here.

In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "multiple" is two or more, unless otherwise clearly and specifically defined. The meaning of "several" is one or more, unless otherwise clearly and specifically defined.

The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments, and the above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, and these changes and improvements fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims (8)

1. A drone swarm attack situation construction system based on knowledge graph, which includes a dynamic knowledge graph generation module and an attack situation generation module, wherein the attack situation construction module includes a data selection module, N combat index item calculation modules and a sorting module, wherein the data selection module selects 1 target and N drones from the dynamic knowledge graph according to the combat mission at time t, the N drones form a drone swarm, and inputs the feature vector of each drone attack target of the first drone swarm into the graph attention mechanism model layer of the first calculation module; the data selection module also selects another drone from the dynamic knowledge graph according to the combat mission at time t to replace one drone in the first drone swarm to form a second drone swarm, and inputs the feature vector of each drone attack target of the second drone swarm into the second calculation module. The graph attention mechanism model layer of the block; the data selection module also selects one more drone from the dynamic knowledge graph according to the combat mission at time t to replace one drone in the second drone swarm that has not replaced other drones to form the third drone swarm, and inputs the feature vector of each drone attack target of the third drone swarm into the graph attention mechanism model layer of the third computing module. By analogy, the data selection module selects one drone from the dynamic knowledge graph according to the combat mission at time t to replace the drone in the N-1th drone swarm that has not replaced other drones to form the Nth drone swarm, and inputs the feature vector of each drone attack target of the Nth drone swarm into the graph attention mechanism model layer of the Nth computing module. The drone members of the N drone swarms are all different, and N is a positive integer greater than or equal to 2; Each combat index item calculation module includes a graph attention mechanism model layer and a neural network layer. The graph attention mechanism model layer includes N graph attention mechanism models. The nth graph attention mechanism model outputs a damage vector to the target at time t as c ⁿ _t , so that the output vector of the graph attention mechanism model layer is C=[c ¹ _t ,…,c ⁿ _t ,…,c ^N _t ]; the neural network generates K combat index items at time t according to the input vector C at time t, where K is a positive integer greater than or equal to 2; The sorting module conducts a comprehensive analysis on the K index items output by the N computing modules, sorts them according to the comprehensive analysis results, and selects the drone swarm with the best comprehensive results as the optimal coordinated drone swarm for the final attack target. 2. The knowledge graph-based drone swarm attack situation construction system according to claim 1, characterized in that the dynamic knowledge graph construction module is used to implement the following process: S01: clustering the existing entities into multiple groups to obtain a group set Cr = [Cr ₁ , ..., Cr _a , ..., Cr _A ], where A is a positive integer greater than or equal to 3, and the groups include a target group, a drone swarm, and an obstacle group; S02: In each group Cr _a , a knowledge graph of the group Cr _a is constructed with entities in the group as nodes and relationships between nodes as edges; S03: Connect related entities in two adjacent groups with edges to form an initial knowledge graph; S04: Calculate the repulsion between the newly discovered entity v _new and each group Cr _a in the existing group set Cr. If it is repulsive with each group Cr _a in the group set Cr, execute S06; if there are non-repulsive groups, the non-repulsive group set is recorded as Cr _absorb = [Cr ₁ , ..., Cr _b , ..., Cr _B ], B is a positive integer greater than or equal to 1, then calculate the similarity Sim (v _NEW , Cr _b ) between the newly discovered entity v _new and each non-repulsive group in the non-repulsive group set Cr _absorb ; S05 If the similarity Sim(v _NEW ,Cr _b ) is less than the threshold, execute S06; if the similarity Sim(v _NEW ,Cr _b ) is greater than or equal to the threshold, place the newly discovered entity V _NEW in the group with the largest similarity Sim(v _NEW ,Cr _b ), calculate the topological association between the newly discovered entity v _new and the existing entities in the group, and connect them through edges; S06: Create a new group Cr _NEW according to the newly discovered entity v _NEW , and add the group Cr _NEW to the group set Cr; S07: Repeat S04, S05 and S06 until all newly discovered entities v _NEW are clustered to form a dynamic knowledge graph. 3. The knowledge graph-based drone swarm attack situation construction system according to claim 2 is characterized in that: , Where σ is the first activation function, is the degree of cooperation between the feature vector _hj of the j-th UAV attack target and the feature vector _hn of the n-th UAV attack target in the swarm; _enj is the feature vector of the edge connecting the j-th UAV and the n-th UAV; is the loss of the feature vector h _n of the n-th UAV attack target by the feature vector h _m of the m-th obstacle interception swarm among M obstacles; e _nm is the feature vector of the edge connecting the m-th obstacle and the n-th UAV; LeakyReLU is the second activation function; ρ is the parameter from the input layer to the hidden layer of the graph attention mechanism model; W _U , W _O , W _r , W _v represent parameter matrices; || means splicing them together. 4. The knowledge graph-based drone swarm attack situation construction system according to claim 3 is characterized in that the neural network includes an input layer, a hidden layer and an output layer, the input layer includes N neurons, and the vector input to the nth neuron at time t is: b _t ⁿ =w ⁿ c ⁿ _t Where w ⁿ is the weighting matrix of the nth damage vector c ⁿ _t ; The hidden layer includes a feedback layer and a comparison layer, wherein the feedback layer includes I neurons, I is a positive integer greater than or equal to N, and the output of the i-th neuron at time t is: , Where w ⁿⁱ is the weight between the nth neuron in the input layer and the i-th neuron in the feedback layer; _ut-1 ⁱ is the output of the i-th neuron at time t-1, α is the adjustment coefficient, and exp is the third activation function; The comparison layer includes I neurons, and the output of the i-th neuron is: , In the formula, β is a constant; The output layer includes K neurons, each of which outputs a combat indicator. The output of the kth neuron is: , Where w ^ki is the weight between the i-th neuron in the comparison layer and the k-th neuron in the output layer. 5. A method for constructing air-ground collaborative situation of unmanned aerial vehicles and unmanned vehicles based on graph computing, characterized by comprising the following steps: Step 1: The data selection module selects 1 target and N drones from the dynamic knowledge graph according to the combat mission at time t. The N drones form a drone swarm, and the feature vector of each drone attack target in the first drone swarm is input into the graph attention mechanism model layer of the first computing module. Step 2: The data selection module also selects another drone from the dynamic knowledge graph according to the combat mission at time t to replace one drone in the first drone swarm to form the second drone swarm, and inputs the feature vector of each drone attack target of the second drone swarm into the graph attention mechanism model layer of the second computing module; Step 3: The data selection module also selects another drone from the dynamic knowledge graph at time t according to the combat mission to replace one drone in the second drone swarm that has not replaced other drones to form the third drone swarm, and inputs the feature vector of each drone attack target of the third drone swarm into the graph attention mechanism model layer of the third computing module. Step 4: By analogy, at time t, the data selection module selects one drone from the dynamic knowledge graph according to the combat mission to replace the drones in the N-1th drone swarm that have not replaced other drones to form the Nth drone swarm, and inputs the feature vector of the attack target of each drone in the Nth drone swarm into the graph attention mechanism model layer of the Nth computing module. The drone members of the N drone swarms are all different, and N is a positive integer greater than or equal to 2; Each combat index calculation module includes a graph attention mechanism model layer and a neural network layer. The graph attention mechanism model layer includes N graph attention mechanism models. The damage vector output by the nth graph attention mechanism model at time t is c ⁿ _t , so that the output vector of the graph attention mechanism model layer is C=[c ¹ _t ,…,c ⁿ _t ,…,c ^N _t ]; Step 5: Generate K combat index items at time t according to the input vector C at time t through a neural network, where K is a positive integer greater than or equal to 2; Step 6: The sorting module conducts a comprehensive analysis on the K index items output by the N computing modules, sorts them according to the comprehensive analysis results, and selects the drone swarm with the best comprehensive results as the optimal coordinated drone swarm for the final attack target. 6. The method for constructing a drone swarm attack situation based on a knowledge graph according to claim 5, wherein the dynamic knowledge graph construction module is used to implement the following process: S01: clustering the existing entities into multiple groups to obtain a group set Cr = [Cr ₁ , ..., Cr _a , ..., Cr _A ], where A is a positive integer greater than or equal to 3, and the groups include a target group, a drone swarm, and an obstacle group; S02: In each group Cr _a , a knowledge graph of the group Cr _a is constructed with entities in the group as nodes and relationships between nodes as edges; S03: Connect related entities in two adjacent groups with edges to form an initial knowledge graph; S04: Calculate the repulsion between the newly discovered entity v _new and each group Cr _a in the existing group set Cr. If it is repulsive with each group Cr _a in the group set Cr, execute S06; if there are non-repulsive groups, the non-repulsive group set is recorded as Cr _absorb = [Cr ₁ , ..., Cr _b , ..., Cr _B ], B is a positive integer greater than or equal to 1, then calculate the similarity Sim (v _NEW , Cr _b ) between the newly discovered entity v _new and each non-repulsive group in the non-repulsive group set Cr _absorb ; S05: If the similarity Sim(v _NEW ,Cr _b ) is less than the threshold, execute S06; if the similarity Sim(v _NEW ,Cr _b ) is greater than or equal to the threshold, place the newly discovered entity V _NEW in the group with the largest similarity Sim(v _NEW ,Cr _b ), calculate the topological association between the newly discovered entity v _new and the existing entities in the group, and connect them through edges; S06: Create a new group Cr _NEW according to the newly discovered entity v _NEW , and add the group Cr _NEW to the group set Cr; S07: Repeat S04, S05 and S06 until all newly discovered entities v _NEW are clustered to form a dynamic knowledge graph. 7. The method for constructing a drone swarm attack situation based on a knowledge graph according to claim 6, characterized in that: , Where σ is the first activation function, is the degree of cooperation between the feature vector _hj of the j-th UAV attack target and the feature vector _hn of the n-th UAV attack target in the swarm; _enj is the feature vector of the edge connecting the j-th UAV and the n-th UAV; is the loss degree of the feature vector h _n of the n-th UAV attack target by the feature vector h _m of the m-th obstacle interception swarm among M obstacles; e _nm is the feature vector of the edge connecting the m-th obstacle and the n-th UAV; LeakyReLU is the second activation function; ρ is the parameter from the input layer to the hidden layer of the graph attention mechanism model; W _U , W _O , W _r , W _v represent parameter matrices; || means splicing them together. 8. The knowledge graph-based drone swarm attack situation construction system according to claim 7, characterized in that the neural network includes an input layer, a hidden layer and an output layer, the input layer includes N neurons, and the vector input to the nth neuron at time t is: b _t ⁿ =w ⁿ c ⁿ _t Where w ⁿ is the weighting matrix of the nth damage vector c ⁿ _t ; The hidden layer includes a feedback layer and a comparison layer, wherein the feedback layer includes I neurons, I is a positive integer greater than or equal to N, and the output of the i-th neuron at time t is: , Where w ⁿⁱ is the weight between the nth neuron in the input layer and the i-th neuron in the feedback layer; _ut-1 ⁱ is the output of the i-th neuron at time t-1, α is the adjustment coefficient, and exp is the third activation function; The comparison layer includes I neurons, and the output of the i-th neuron is: , In the formula, β is a constant; The output layer includes K neurons, each of which outputs a combat indicator. The output of the kth neuron is: , Where w ^ki is the weight between the i-th neuron in the comparison layer and the k-th neuron in the output layer.