Method and system for generating air action plan based on reasoning about opponent's intention

Abstract

This application discloses a method and system for generating an air action plan that combines reasoning about the opponent's intentions. The method includes: when executing a cluster confrontation, the threat perception and plan display component receives the real-time confrontation situation, evaluates the threat level of the cluster target, and issues a threat alert to the intention reasoning model; the intention reasoning model identifies the opponent's possible intention type, updates the knowledge graph, queries the entity relationship in the knowledge graph, and obtains our alternative action plan; the effectiveness evaluation model evaluates the confrontation effectiveness between each possible intention of the opponent and our alternative action plan in a template matching manner; the game solving model forms a game matrix of various confrontation situations and confrontation effectiveness, transforms it into a target optimization problem, and calculates our optimal action plan and execution probability. This application infers the opponent's intentions by analyzing the real-time confrontation situation and generates our air action plan, which has the advantages of high dynamics, high reliability, and good real-time performance.

Description

Technical Field

[0001] The present application relates to the field of distributed collaborative technology, and in particular to a method and system for generating an air action plan in combination with reasoning about the other party's intention. Background Art

[0002] In recent years, the scenarios and forms of swarm confrontations have become increasingly diverse, placing higher demands on the quality and efficiency of air action plan generation. Air action plans provide strategic guidance and resource allocation during swarm confrontations. High-quality air action plans can create an advantage in swarm confrontations, continuously improving our effectiveness and becoming a key factor in determining the outcome of swarm confrontations. However, existing air action plan generation systems often overlook issues such as the uncertainty and rapid changes in the enemy's intentions, resulting in a lack of dynamic, real-time, and reliable air action plans. Summary of the Invention

[0003] In view of this, the present application provides a method and system for generating an air action plan that combines reasoning about the other party's intention.

[0004] This application discloses a method for generating an air action plan in combination with opponent intention reasoning, which includes:

[0005] When executing swarm confrontation, the threat perception and planning display component receives real-time confrontation status, assesses the threat level of the swarm target, and issues threat alerts to the intent reasoning model. The real-time confrontation status includes the platform's motion status, the number of remaining devices, the target type, and the target's trajectory. Swarm confrontation is the use of distributed and collaborative intelligent clusters to conduct adversarial offense and defense.

[0006] The intention inference model identifies the other party's possible intention type, updates the knowledge graph, queries the entity relationships in the knowledge graph, and obtains our alternative action plan. The other party's possible intention includes the number of clusters, cluster movement trends, cluster action types, and cluster threat level. The knowledge graph includes entities and entity relationships.

[0007] The effectiveness evaluation model uses template matching to evaluate the effectiveness of each possible opponent's intention against our alternative action plan. Our alternative action plan includes our current number of clusters, cluster types, cluster real-time locations, cluster perception range, cluster attack range, and cluster confrontation resources consumed.

[0008] The game-solving model transforms various confrontation situations and confrontation effectiveness into a game matrix, converts it into a target optimization problem, and calculates our optimal action plan and execution probability.

[0009] Furthermore, the assessing of the threat level of the cluster target includes:

[0010] Cluster Target Depend on Single target composition, Threat level of a single target for:

[0011]

[0012] in, 、 Target Air Superiority Target confrontation capability The weight value of , ;

[0013] Cluster Target Threat level for:

[0014]

[0015] when When the notification intention inference model starts running, Threat level alarm threshold.

[0016] Furthermore, the assessing of the threat level of the cluster target includes:

[0017] The threat level of a single target is related to the target's air superiority; the target's air superiority is related to the confrontation angle, confrontation distance and speed of both sides. Expressed as:

[0018]

[0019] in, 、 、 They are the weights of the confrontation angle, confrontation distance and speed respectively; is the angle between the target and our course; This is our maximum detection distance. The distance between the detected target and our side; is the opponent's maximum speed, is the speed difference between the target and our side; 、 、 The value range of is the same, both greater than 0 and less than 1.

[0020] Furthermore, the assessing of the threat level of the cluster target includes:

[0021] The threat level of a single target is related to the target confrontation capability; the target confrontation capability is related to the maneuverability, strike capability, and detection capability of both platforms. Expressed as:

[0022]

[0023] in, The difference in mobility between the two platforms, To combat the gap in capabilities between the two sides, is the platform detection capability gap, and its value varies with the platform types of both parties; is the weight value of the difference in mobility between the two sides, is the weight value of the difference in attack capability between the two sides, Detect capability gaps between the two platforms.

[0024] Furthermore, the intention inference model infers the other party's possible intention based on the target information; the intention inference model consists of a neural network module and a knowledge inference module; the neural network module consists of a long short-term memory network (LSTM), a fully connected network, and a softmax layer. Each LSTM layer has multiple hidden layers, and each fully connected layer has multiple hidden layers. The ReLU excitation function is used, and the softmax layer logically classifies and outputs the intention type; the intention types include coordinated attack, circling, avoidance, and formation flight;

[0025] The neural network module collects historical track information of cluster targets, identifies the current cluster target's intention type, and simultaneously updates the relevant information of the threat target to the knowledge reasoning module; the relevant information includes the component platform, type, and intention category;

[0026] The knowledge reasoning module includes a knowledge graph and a knowledge rule set. The entities in the knowledge graph are divided into platforms, intent types, confrontation strategy types, clusters, action plans, and targets; the entity relationships in the knowledge graph are divided into plan execution, cluster management, attack targets, sensor loading, equipment, and command and control; the knowledge rule set includes multiple rules, each of which describes the conditions for the establishment of an entity relationship. If the entity relationship is established, the corresponding entity relationship is queried from the knowledge graph based on the entity relationship to obtain our alternative action plan.

[0027] Furthermore, the effectiveness evaluation model pre-emptively deduces the confrontation process between various possible intentions of the other party and our action plan, and uses three types of indicators, namely cost, benefit, and loss, as effectiveness evaluation results to form a template library; the template library includes multiple templates, each template including the relative situation of the two parties, our action plan, the intention of the other party, and the effectiveness evaluation result;

[0028] The effectiveness evaluation model matches the template library with the current confrontation situation, combines the evaluation effectiveness value of the template with the highest similarity with the similarity weight, and calculates the evaluation effectiveness of the current confrontation situation; the current confrontation situation includes the relative situation of the two sides, our alternative action plan, and the possible intentions of the other party.

[0029] Furthermore, let the current relative situation between the two sides be , our alternative action plan , the other party may intend , then the template in the template library that is most similar to the current adversarial situation The following conditions are met:

[0030]

[0031]

[0032] in, is the total number of templates in the template library, For the The relative situation in the template, For the Our action plan as a template, For the The other party's intention of the template, The current relative situation between the two sides With the Relative situation in templates The difference weight value of Alternative action plans for us With the Our action plan in the template The difference weight value of The other party's possible intention With the The other party's intention in the template The difference weight value of

[0033] For the The smaller the difference, the more similar the template is to the current confrontation situation.

[0034] Furthermore, the effectiveness evaluation model evaluates the effectiveness of the current confrontation situation, including:

[0035] Based on the template that is most similar to the current confrontation situation The degree of difference , normalized to generate similarity weight , then the cost of our side in the current confrontation situation is 、Our benefits in the current confrontation situation 、Our losses in the current confrontation situation , and the overall evaluation performance value They are:

[0036]

[0037]

[0038]

[0039] in, For template The pre-calculated adversarial cost in for The calculation weight of For template The pre-calculated adversarial payoff in for The calculation weight of For template The adversarial loss pre-computed in for The calculation weight of is the normalization function.

[0040] Furthermore, the game solving model forms a game matrix with various confrontation situations and evaluation effectiveness; the leftmost column of the game matrix constructed by the game solving model is our alternative action plan, the top row is the opponent's possible intention, and each matrix value is our alternative action plan Possible intentions of the other party Evaluation value of the combat effectiveness .

[0041] Furthermore, the game solving model transforms the game matrix into a target optimization problem and calculates the optimal action plan and execution probability, including:

[0042] The game solving model is to maximize the difference between our gains and losses as the optimization goal and construct the optimization function , the optimization variable is ,in For our Alternative action plans, For the The execution probability of alternative action plans, the optimization function The particle swarm optimization algorithm (PSO) is used to solve the problem and obtain the optimal action plan and execution probability. The core idea of ​​the game solving model is to find the optimal combination of our alternative action plan and execution probability. We will compete with the opponent according to this optimal combination, and the sum of all the results of the competition is optimal. The target optimization solution of the game solving model is as follows:

[0043]

[0044]

[0045]

[0046]

[0047] in, is the maximum value function, Indicates the Our alternative action plan Possible intentions to gather with the other party The adversarial benefits, Indicates the Our alternative action plan Possible intentions to gather with the other party Minimize the loss, Indicates the use of Our alternative action plan With the other party Possible intention Our benefits in the current confrontation situation The weight value and our losses in the current confrontation situation The weight value of Indicates the The probability of executing an alternative action plan With the other party The execution probability of a possible intention The result of the operation, Indicates our Alternative Action Plans With the other party Possible intention The profit value of the confrontation, Indicates our Alternative Action Plans With the other party Possible intention The loss value of the confrontation, A collection of possible intentions of the other party The number of possible intentions of the other party, The cost to our side in the current confrontation situation.

[0048] This application also discloses a system for generating an air action plan in combination with opponent intention reasoning, which implements the above-mentioned method for generating an air action plan in combination with opponent intention reasoning, and includes a threat perception and plan display component, an intention reasoning model, an effectiveness evaluation model, and a game solving model;

[0049] When executing swarm confrontation, the threat perception and plan display component is used to receive real-time confrontation situation and assess the threat level of swarm targets; the real-time confrontation situation includes the movement status of our platform, the number of remaining equipment, the target type, and the target track;

[0050] The intention reasoning model is used to identify the opponent's possible intention types and use knowledge reasoning to initially screen our alternative action plans. The opponent's possible intentions include the number of clusters, cluster movement trends, cluster action types, and cluster threat level.

[0051] The effectiveness evaluation model is used to evaluate the effectiveness of each possible opponent's intention against our alternative action plan through template matching. Our alternative action plan includes our current number of clusters, cluster types, cluster real-time locations, cluster perception range, cluster attack range, and cluster confrontation resources consumed.

[0052] The game-solving model is used to transform various confrontation situations and confrontation effectiveness into a game matrix, convert it into a target optimization problem, and calculate our optimal action plan and execution probability.

[0053] Due to the adoption of the above technical solution, this application has the following advantages:

[0054] This application combines dynamic game theory with air action plan generation to construct an air action plan generation system. By analyzing the threat level of enemy clusters, the system focuses limited air resources on the enemy's core targets. Furthermore, the system fully considers the various possible enemy intentions and the effectiveness of our action plan, ensuring optimal effectiveness. This system is highly dynamic, real-time, and reliable.

[0055] 2. The target confrontation capability of this application is related to the maneuverability, strike capability, and detection capability of both platforms. When analyzing the threat level of a single target, the threat perception and planning display component reads the various capability values ​​and weights of the platform of that type from the database and normalizes them into the various capabilities of the platform. The maneuverability of each type of platform in the database is composed of capability factors such as the platform's maximum speed, overload, turning angle, and range; the strike capability includes capability factors such as the maximum attack distance of the equipment, damage range, strike accuracy, and flight speed; the detection capability is composed of capability factors such as the sensor's detection range, false alarm rate, and frequency band range.

[0056] 3. The intention inference model of this application outputs the other party's intention type through a neural network module that has been trained offline. During offline training, the training data of the neural network module includes three types of features: target historical track, target platform type, and target threat level. The target historical track is composed of 30 historical waypoints sampled at intervals of 0.5 seconds. Each waypoint contains the target's longitude, latitude, altitude, heading angle, and speed. The data set has a total of 8,000 samples, which are divided into training and test sets in a ratio of 8:2. The time window of the long short-term memory network is set to 10s, and the number of training times is set to 200.

[0057] 4. The knowledge reasoning module of the intention reasoning model of this application obtains a variety of alternative action plans for our side based on the real-time situation, expert knowledge, and multiple knowledge rule sets (Semantic Web Rule Language) for the target intention. In the knowledge reasoning module, expert knowledge and real-time situation constitute a real-time updated knowledge graph, which refreshes the attributes and relationships of each entity in the knowledge graph by parsing each frame of real-time situation data. The entities in the knowledge graph are divided into 6 categories: platform, intent type, confrontation strategy type, cluster, action plan, and target; entity relationships can be divided into 7 categories: plan execution, strategy adoption, cluster management, attack target, sensor loading, equipment equipping, and command and control. The knowledge rule set (Semantic Web Rule Language) infers our alternative action plans by querying the entity status in the knowledge graph.

[0058] 5. This application's effectiveness evaluation model uses template matching to evaluate the effectiveness of the adversary's potential intentions against our alternative action plans. This effectiveness evaluation model internally constructs a template library containing a variety of adversarial templates. When generating each template offline, the model randomizes the initial adversarial situation, our action plan, the adversary's intentions, and other adversarial conditions, and then begins a simulation. After the simulation is complete, the adversarial effectiveness is calculated, and each adversarial condition and corresponding adversarial effectiveness are stored as a template in the template library.

[0059] 6. After Benshen's game matrix is ​​transformed into a target optimization problem, the optimal result is solved by the PSO optimization algorithm. When solving this problem with the PSO optimization algorithm, the optimization parameters are set as follows: the population size is 40, the number of iterations is 200, and the particle search space is , limiting the particle velocity range to , the spatial dimension is 2, the particle inertia weight is 0.9, the self-learning factor is 0.5, and the group learning factor is 0.8.

[0060] 7. The air action plan generation system constructed using this application is highly dynamic, real-time, and reliable, with a wide range of applications. It is highly applicable to distributed collaborative tasks, cluster game confrontation and other task scenarios.

[0061] 8. This application is applicable to fields such as distributed collaborative tasks, cluster game confrontation, situational awareness, etc., especially the field of distributed collaborative tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments recorded in the embodiments of the present application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0063] Figure 1 This is an architecture diagram of an air action plan generation system combined with adversary intention reasoning according to an embodiment of the present application;

[0064] Figure 2 Schematic diagram of threat assessment of the threat perception and plan display component of an embodiment of the present application;

[0065] Figure 3 This is a schematic diagram of the functional implementation of the intention reasoning model in an embodiment of the present application;

[0066] Figure 4 2 is a schematic diagram of the adversarial effectiveness evaluation of the effectiveness evaluation model of the embodiment of the present application;

[0067] Figure 5 It is a flow chart for solving the optimization problem of the game solving model in the embodiment of the present application. DETAILED DESCRIPTION

[0068] The present application is further described with reference to the accompanying drawings and embodiments. The embodiments described are only a part of the embodiments of the present application, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field should fall within the scope of protection of the embodiments of the present application.

[0069] See also Figure 1 The present application provides an embodiment of a method for generating an air action plan in combination with reasoning about the other party's intention, which includes:

[0070] When executing swarm confrontation, the threat perception and planning display component receives real-time confrontation status, assesses the threat level of the swarm target, and issues threat alerts to the intent reasoning model. The real-time confrontation status includes the platform's motion status, the number of remaining devices, the target type, and the target's trajectory. Swarm confrontation is the use of distributed and collaborative intelligent clusters to conduct adversarial offense and defense.

[0071] The intention inference model identifies the other party's possible intention type, updates the knowledge graph, queries the entity relationships in the knowledge graph, and obtains our alternative action plan. The other party's possible intention includes the number of clusters, cluster movement trends, cluster action types, and cluster threat level. The knowledge graph includes entities and entity relationships.

[0072] The effectiveness evaluation model uses template matching to evaluate the effectiveness of each possible opponent's intention against our alternative action plan. Our alternative action plan includes our current number of clusters, cluster types, cluster real-time locations, cluster perception range, cluster attack range, and cluster confrontation resources consumed.

[0073] The game-solving model transforms various confrontation situations and confrontation effectiveness into a game matrix, converts it into a target optimization problem, and calculates our optimal action plan and execution probability.

[0074] Optionally, evaluating the threat level of the cluster target includes:

[0075] Cluster Target Depend on Single target composition, Threat level of a single target for:

[0076]

[0077] in, 、 Target Air Superiority Target confrontation capability The weight value of , ;

[0078] Cluster Target Threat level for:

[0079]

[0080] when When the notification intention inference model starts running, Threat level alarm threshold.

[0081] Optionally, evaluating the threat level of the cluster target includes:

[0082] The threat level of a single target is related to the target's air superiority; the target's air superiority is related to the confrontation angle, confrontation distance and speed of both sides. Expressed as:

[0083]

[0084] in, 、 、 They are the weights of the confrontation angle, confrontation distance and speed respectively; is the angle between the target and our course; This is our maximum detection distance. The distance between the detected target and our side; is the opponent's maximum speed, is the speed difference between the target and our side; 、 、 The value range of is the same, both greater than 0 and less than 1.

[0085] Optionally, evaluating the threat level of the cluster target includes:

[0086] The threat level of a single target is related to the target confrontation capability; the target confrontation capability is related to the maneuverability, strike capability, and detection capability of both platforms. Expressed as:

[0087]

[0088] in, The difference in mobility between the two platforms, To combat the gap in capabilities between the two sides, is the platform detection capability gap, and its value varies with the platform types of both parties; is the weight value of the difference in mobility between the two sides, is the weight value of the difference in attack capability between the two sides, Detect capability gaps between the two platforms. 、 、 The value range of is the same, both greater than 0 and less than 1.

[0089] Optionally, the intention inference model infers the other party's possible intention based on the target information; the intention inference model is composed of a neural network module and a knowledge inference module; the neural network module is composed of a long short-term memory network (LSTM), a fully connected network, and a softmax layer, each LSTM layer has multiple hidden layers, each fully connected layer has multiple hidden layers, and uses a ReLU excitation function. Finally, the softmax layer logically classifies and outputs the intention type; the intention types include coordinated attack, circling, avoidance, and formation flight;

[0090] The neural network module collects historical track information of cluster targets, identifies the current cluster target's intention type, and simultaneously updates the relevant information of the threat target to the knowledge reasoning module; the relevant information includes the component platform, type, and intention category;

[0091] The knowledge reasoning module includes a knowledge graph and a knowledge rule set. The entities in the knowledge graph are divided into platforms, intent types, confrontation strategy types, clusters, action plans, and targets; the entity relationships in the knowledge graph are divided into plan execution, cluster management, attack targets, sensor loading, equipment, and command and control; the knowledge rule set includes multiple rules, each of which describes the conditions for the establishment of an entity relationship. If the entity relationship is established, the corresponding entity relationship is queried from the knowledge graph based on the entity relationship to obtain our alternative action plan.

[0092] Optionally, the effectiveness evaluation model pre-emptively deduces offline the confrontation process between various possible intentions of the other party and our action plan, and uses three types of indicators, namely, cost, benefit, and loss, as effectiveness evaluation results to form a template library; the template library includes multiple templates, each template including the relative situation of both parties, our action plan, the intention of the other party, and the effectiveness evaluation result;

[0093] The effectiveness evaluation model matches the template library with the current confrontation situation, combines the evaluation effectiveness value of the template with the highest similarity with the similarity weight, and calculates the evaluation effectiveness of the current confrontation situation; the current confrontation situation includes the relative situation of the two sides, our alternative action plan, and the possible intentions of the other party.

[0094] Optionally, let the current relative situation between the two parties be , our alternative action plan , the other party may intend , then the template in the template library that is most similar to the current adversarial situation The following conditions are met:

[0095]

[0096]

[0097] in, is the total number of templates in the template library, For the The relative situation in the template, For the Our action plan as a template, For the The other party's intention of the template, The current relative situation between the two sides With the Relative situation in templates The difference weight value of Alternative action plans for us With the Our action plan in the template The difference weight value of The other party's possible intention With the The other party's intention in the template The difference weight value of

[0098] For the The smaller the difference, the more similar the template is to the current confrontation situation.

[0099] Optionally, the effectiveness evaluation model evaluates the effectiveness of the current confrontation situation, including:

[0100] Based on the template that is most similar to the current confrontation situation The degree of difference , normalized to generate similarity weight , then the cost of our side in the current confrontation situation is 、Our benefits in the current confrontation situation 、Our losses in the current confrontation situation , and the overall evaluation performance value They are:

[0101]

[0102]

[0103]

[0104] in, For template The pre-calculated adversarial cost in for The calculation weight of For template The pre-calculated adversarial payoff in for The calculation weight of For template The adversarial loss pre-computed in for The calculation weight of is the normalization function.

[0105] Optionally, the game solving model forms a game matrix with various confrontation situations and evaluation effectiveness; the leftmost column of the game matrix constructed by the game solving model is our alternative action plan, the top row is the opponent's possible intention, and each matrix value is our alternative action plan Possible intentions of the other party Evaluation value of the combat effectiveness .

[0106] Optionally, the game solving model converts the game matrix into a target optimization problem and calculates the optimal action plan and execution probability of our side, including:

[0107] The game solving model is to maximize the difference between our gains and losses as the optimization goal and construct the optimization function , the optimization variable is ,in For our Alternative action plans, For the The execution probability of alternative action plans, the optimization function The particle swarm optimization algorithm (PSO) is used to solve the problem and obtain the optimal action plan and execution probability. The core idea of ​​the game solving model is to find the optimal combination of our alternative action plan and execution probability. We will compete with the opponent according to this optimal combination, and the sum of all the results of the competition is optimal. The target optimization solution of the game solving model is as follows:

[0108]

[0109]

[0110]

[0111]

[0112] in, is the maximum value function, Indicates the Our alternative action plan Possible intentions to gather with the other party The adversarial benefits, Indicates the Our alternative action plan Possible intentions to gather with the other party Minimize the loss, Indicates the use of Our alternative action plan With the other party Possible intention Our benefits in the current confrontation situation The weight value and our losses in the current confrontation situation The weight value of Indicates the The probability of executing an alternative action plan With the other party The execution probability of a possible intention The result of the operation (for example, ), Indicates our Alternative Action Plans With the other party Possible intention The profit value of the confrontation, Indicates our Alternative Action Plans With the other party Possible intention The loss value of the confrontation, A collection of possible intentions of the other party The number of possible intentions of the other party, is our cost in the current confrontation situation. .

[0113] This application also provides an embodiment of an air action plan generation system that combines opponent intention reasoning, which implements the air action plan generation method that combines opponent intention reasoning described in the above embodiment, and includes a threat perception and plan display component, an intention reasoning model, an effectiveness evaluation model, and a game solving model;

[0114] When executing swarm confrontation, the threat perception and plan display component is used to receive real-time confrontation situation and assess the threat level of swarm targets; the real-time confrontation situation includes the movement status of our platform, the number of remaining equipment, the target type, and the target track;

[0115] The intention reasoning model is used to identify the opponent's possible intention types and use knowledge reasoning to initially screen our alternative action plans. The opponent's possible intentions include the number of clusters, cluster movement trends, cluster action types, and cluster threat level.

[0116] The effectiveness evaluation model is used to evaluate the effectiveness of each possible opponent's intention against our alternative action plan through template matching. Our alternative action plan includes our current number of clusters, cluster types, cluster real-time locations, cluster perception range, cluster attack range, and cluster confrontation resources consumed.

[0117] The game-solving model is used to transform various confrontation situations and confrontation effectiveness into a game matrix, convert it into a target optimization problem, and calculate our optimal action plan and execution probability.

[0118] For ease of understanding, this application provides a more specific embodiment:

[0119] See Figure 1The present application discloses an air action plan generation system that combines opponent intention reasoning and consists of a threat perception and plan display component, an intention reasoning model, an effectiveness evaluation model, and a game solving model. The threat perception and plan display component calculates the opponent's cluster threat level through real-time confrontation situation and platform inherent capability attributes, and issues a threat alert; the intention reasoning model identifies the opponent's possible intention type through the LSTM neural network, updates the knowledge graph based on expert knowledge in real time, and obtains our alternative action plan by querying the entity state in the knowledge graph through the knowledge rule set (Semantic Web rule language); the effectiveness evaluation model matches the template library built offline with the current confrontation situation, combines the evaluation effectiveness value of the template with the highest similarity with the similarity weight, and calculates the evaluation effectiveness of the current confrontation situation; the game solving model takes maximizing benefits and minimizing losses as optimization goals, constructs a target optimization function, and uses the PSO (particle swarm) algorithm to solve the optimal action plan and execution probability of our side.

[0120] See Figure 2 The operational process of the threat perception and planning display component of this application is as follows: After receiving the real-time confrontation situation, the threat perception and planning display component calculates the threat level of each individual target from the bottom up. Based on the relative situation of each individual target and the friendly side, the respective air superiority can be calculated. The target's confrontation capability can be calculated based on the platform type and inherent capabilities of the individual target. The threat level of each individual target is calculated through weighted calculation. Finally, the threat level of all platforms in a cluster is averaged as the threat level of the entire cluster. When the cluster's threat level exceeds the set threat threshold, the component issues a threat alert, notifying other models to begin processing.

[0121] See Figure 3 The functional implementation process of the intention reasoning model of this application is as follows: when the intention reasoning model receives threat target information, it collects the historical track information of the cluster target in the last 10 seconds, and uses the trained LSTM neural network to identify the intention category of the current cluster target. At the same time, the composition platform, type, intention category and other data of the threat target are synchronously updated to the knowledge graph. The status of the corresponding entity in the knowledge graph is queried by the prepared knowledge rule set to obtain our alternative action plan.

[0122] See Figure 4 The adversarial effectiveness evaluation process of the effectiveness evaluation model in this application is as follows: After receiving the real-time situation, our alternative action plans, and the opponent's possible intentions, the effectiveness evaluation model combines our alternative action plans with the opponent's possible intentions, and combines them with the real-time situation to form multiple adversarial scenarios. For each adversarial scenario, the model searches the offline template library to find the template with the highest similarity. The similarity between this template and the current adversarial scenario is normalized into a similarity weight. The effectiveness evaluation value of the current adversarial scenario is calculated based on the effectiveness evaluation value of this template and the similarity weight.

[0123] See Figure 5 The optimization problem-solving process of the game-solving model in this application is as follows: After receiving various adversarial scenarios and their effectiveness evaluations, the game-solving model constructs a game matrix and transforms it into a target optimization problem. The optimization objectives are determined to be maximum gain and minimum loss, and the optimization variables are the team's alternative action plans and execution probabilities. When solving this problem, parameters such as population size, dimension, number of iterations, weights, and learning factors are initialized. During each iteration, the positions of individual particles and the group are updated, and the optimal fitness values ​​of the individuals and the group are calculated. When the maximum number of iterations is reached, or the change between two fitness values ​​is less than a set threshold, the team's optimal action plan and action probability are output.

[0124] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit them. Although the present application has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present application can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present application should be included in the scope of protection of the claims of the present application.

Claims

1. A method for generating an air action plan by combining reasoning with the opponent's intention, characterized in that: include: When executing cluster confrontation, the threat perception and planning display component receives real-time confrontation situation, evaluates the threat level of cluster targets, and issues threat alerts to the intention reasoning model; the real-time confrontation situation includes the motion status of our platform, the number of remaining devices, the target type, and the target track; Cluster confrontation is the use of distributed collaborative intelligent clusters to conduct adversarial attack and defense; The intention inference model identifies the other party's possible intention type, updates the knowledge graph, queries the entity relationships in the knowledge graph, and obtains our alternative action plan. The other party's possible intention includes the number of clusters, cluster movement trends, cluster action types, and cluster threat level. The knowledge graph includes entities and entity relationships. The effectiveness evaluation model uses template matching to evaluate the effectiveness of each possible opponent's intention against our alternative action plan. Our alternative action plan includes our current number of clusters, cluster types, cluster real-time locations, cluster perception range, cluster attack range, and cluster confrontation resources consumed. The game-solving model transforms various confrontation situations and confrontation effectiveness into a game matrix, transforms it into a target optimization problem, and calculates our optimal action plan and execution probability; The effectiveness evaluation model pre-emptively deduces the confrontation process between various possible intentions of the other party and our action plan, and uses three types of indicators, namely cost, benefit and loss, as the effectiveness evaluation results to form a template library. The template library includes multiple templates, each template includes the relative situation of the two parties, our action plan, the intention of the other party, and the effectiveness evaluation results. The effectiveness evaluation model matches the template library with the current confrontation situation, combines the evaluation effectiveness value of the template with the highest similarity with the similarity weight, and calculates the evaluation effectiveness of the current confrontation situation. The current confrontation situation includes the relative situation of both sides, our alternative action plan, and the possible intentions of the other side. Assume that the current relative situation between the two sides is , our alternative action plan , the other party may intend , then the template in the template library that is most similar to the current adversarial situation The following conditions are met: in, is the total number of templates in the template library, For the The relative situation in the template, For the Our action plan as a template, For the The other party's intention of the template, The current relative situation between the two sides With the Relative situation in templates The difference weight value of Alternative action plans for us With the Our action plan in the template The difference weight value of The other party's possible intention With the The other party's intention in the template The difference weight value of For the The smaller the difference, the more similar the template is to the current confrontation situation.

2. The method for generating an aerial action plan by combining inference of the other party's intention according to claim 1, characterized in that: The threat level of the cluster target is evaluated, including: Cluster Target Depend on Single target composition, Threat level of a single target for: in, 、 Target Air Superiority Target confrontation capability The weight value of , ; Cluster Target Threat level for: when When the notification intention inference model starts running, Threat level alarm threshold.

3. The method for generating an air action plan by combining inference of the other party's intention according to claim 2, characterized in that: The threat level of the cluster target is evaluated, including: The threat level of a single target is related to the target's air superiority; the target's air superiority is related to the confrontation angle, confrontation distance and speed of both sides. Expressed as: in, 、 、 They are the weights of the confrontation angle, confrontation distance and speed respectively; is the angle between the target and our course; This is our maximum detection distance. The distance between the detected target and our side; is the opponent's maximum speed, is the speed difference between the target and our side; 、 、 The value range of is the same, both greater than 0 and less than 1.

4. The method for generating an air action plan by combining inference of the other party's intention according to claim 2, characterized in that: The threat level of the cluster target is evaluated, including: The threat level of a single target is related to the target confrontation capability; the target confrontation capability is related to the maneuverability, strike capability, and detection capability of both platforms. Expressed as: in, The difference in mobility between the two platforms, To combat the gap in capabilities between the two sides, is the platform detection capability gap, and its value varies with the platform types of both parties; is the weight value of the difference in mobility between the two sides, is the weight value of the difference in attack capability between the two sides, Detect capability gaps between the two platforms.

5. The method for generating an aerial action plan by combining inference of the opponent's intention according to claim 1, characterized in that: The intention inference model infers the other party's possible intention based on the target information. The intention inference model consists of a neural network module and a knowledge inference module. The neural network module consists of a long short-term memory network (LSTM), a fully connected network, and a softmax layer. Each LSTM layer has multiple hidden layers, and each fully connected layer has multiple hidden layers. The ReLU excitation function is used, and the intent type is finally output through logical classification at the softmax layer. Intent types include coordinated attack, circling, evasion, and formation flying; The neural network module collects historical track information of cluster targets, identifies the current cluster target's intention type, and simultaneously updates the relevant information of the threat target to the knowledge reasoning module; the relevant information includes the component platform, type, and intention category; The knowledge reasoning module includes a knowledge graph and a knowledge rule set. The entities in the knowledge graph are divided into platforms, intent types, confrontation strategy types, clusters, action plans, and targets; the entity relationships in the knowledge graph are divided into plan execution, cluster management, attack targets, sensor loading, equipment, and command and control; the knowledge rule set includes multiple rules, each of which describes the conditions for the establishment of an entity relationship. If the entity relationship is established, the corresponding entity relationship is queried from the knowledge graph based on the entity relationship to obtain our alternative action plan.

6. The method for generating an air action plan by combining inference of the opponent's intention according to claim 1, characterized in that: The effectiveness evaluation model evaluates the effectiveness of the current confrontation situation, including: Based on the template that is most similar to the current confrontation situation The degree of difference , normalized to generate similarity weight , then the cost of our side in the current confrontation situation is 、Our benefits in the current confrontation situation 、Our losses in the current confrontation situation , and the overall evaluation performance value They are: in, For template The pre-calculated adversarial cost in for The calculation weight of For template The pre-calculated adversarial payoff in for The calculation weight of For template The adversarial loss pre-computed in for The calculation weight of is the normalization function.

7. The method for generating an aerial action plan by combining inference of the opponent's intention according to claim 6, characterized in that: The game solving model forms a game matrix with various confrontation situations and evaluation effectiveness. The leftmost column of the game matrix constructed by the game solving model is our alternative action plan, and the top row is the possible intention of the other party. Each matrix value is our alternative action plan. Possible intentions of the other party Evaluation value of the combat effectiveness .

8. The method for generating an air action plan by combining inference of the opponent's intention according to claim 6 or 7, characterized in that: The game-solving model transforms the game matrix into a target optimization problem and calculates the optimal action plan and execution probability, including: The game solving model is to maximize the difference between our gains and losses as the optimization goal and construct the optimization function , the optimization variable is ,in For our Alternative action plans, For the The execution probability of alternative action plans, the optimization function The particle swarm optimization algorithm (PSO) is used to solve the problem and obtain the optimal action plan and execution probability. The core idea of ​​the game solving model is to find the optimal combination of our alternative action plan and execution probability. We will compete with the opponent according to this optimal combination, and the sum of all the results of the competition is optimal. The target optimization solution of the game solving model is as follows: in, is the maximum value function, Indicates the Our alternative action plan Possible intentions to gather with the other party The adversarial benefits, Indicates the Our alternative action plan Possible intentions to gather with the other party Minimize the loss, Indicates the use of Our alternative action plan With the other party Possible intention Our benefits in the current confrontation situation The weight value of the current confrontation situation and our losses The weight value of Indicates the The probability of executing an alternative action plan With the other party The probability of executing a possible intention The result of the operation, Indicates our Alternative Action Plans With the other party Possible intention The profit value of the confrontation, Indicates our Alternative Action Plans With the other party Possible intention The loss value of the confrontation, A collection of possible intentions of the other party The number of possible intentions of the other party, The cost to our side in the current confrontation situation.

9. A system for generating an air action plan by combining reasoning with the opponent's intention, which implements the method for generating an air action plan by combining reasoning with the opponent's intention as described in any one of claims 1 to 8, characterized in that: Includes threat perception and plan display components, intention reasoning model, effectiveness evaluation model and game solving model; When executing swarm confrontation, the threat perception and plan display component is used to receive real-time confrontation situation and assess the threat level of swarm targets; the real-time confrontation situation includes the movement status of our platform, the number of remaining equipment, the target type, and the target track; The intention reasoning model is used to identify the opponent's possible intention types and use knowledge reasoning to initially screen our alternative action plans. The opponent's possible intentions include the number of clusters, cluster movement trends, cluster action types, and cluster threat level. The effectiveness evaluation model is used to evaluate the effectiveness of each possible opponent's intention against our alternative action plan through template matching. Our alternative action plan includes our current number of clusters, cluster types, cluster real-time locations, cluster perception range, cluster attack range, and cluster confrontation resources consumed. The game-solving model is used to transform various confrontation situations and confrontation effectiveness into a game matrix, convert it into a target optimization problem, and calculate our optimal action plan and execution probability.

Images