A cooperative game method and system for cross-domain heterogeneous unmanned cluster

Abstract

The application relates to the technical field of war game deduction, and discloses a cross-domain heterogeneous unmanned cluster cooperative game war game deduction method and system, which aims to solve the problems of poor reality of deduction, timeliness of decision-making and reliability of task execution of an existing method, and mainly comprises the following steps: a multi-domain heterogeneous cooperative combat model is constructed, a multi-granularity real-time situation assessment is performed based on the model to generate a global strategic situation and a local tactical situation; a dynamic game problem is constructed based on the global strategic situation, an opponent strategy evolution trend is predicted by using a long short-term memory network to correct a strategy network output, an optimal cooperative countermeasures strategy is generated, and a self-adaptive task allocation mechanism is introduced based on the local tactical situation to generate a task allocation instruction set; a deduction process is driven based on the above strategy and instruction, and state change data of each unmanned platform is fed back to the situation assessment link in real time. The application improves the systematicness, dynamic adaptability and decision reliability of cross-domain heterogeneous unmanned cluster war game deduction.

Description

Technical Field

[0001] This invention relates to the field of wargaming technology, specifically to a collaborative game wargaming method and system for cross-domain heterogeneous unmanned clusters. Background Technology

[0002] With the rapid development of unmanned systems technology and artificial intelligence algorithms, future combat is accelerating its evolution towards unmanned, swarm-based, and intelligent operations. Cross-domain heterogeneous unmanned swarms, organic combat groups composed of various types of unmanned platforms such as airborne (drones), ground-based (unmanned vehicles, robots), and surface / underwater (unmanned surface vessels, underwater vehicles), can achieve combat effectiveness far exceeding that of a single platform through multi-domain collaboration and functional complementarity. Wargaming, as a crucial means of testing operational concepts, evaluating operational effectiveness, and supporting decision-making, urgently requires the development of new wargaming methods and systems capable of supporting collaborative operations by cross-domain heterogeneous unmanned swarms.

[0003] However, existing wargaming methods have significant shortcomings when dealing with cross-domain heterogeneous unmanned clusters:

[0004] First, traditional wargaming systems are mostly based on centralized server architectures with fixed rules and poor adaptability. They are difficult to represent the complex cross-domain interaction relationships and nonlinear collaborative effects between unmanned platforms in heterogeneous unmanned clusters. Furthermore, there is a lack of a unified data transmission and state synchronization mechanism between model construction and subsequent decision-making processes, resulting in insufficient realism and credibility of the simulations.

[0005] Second, existing intelligent decision-making algorithms are mostly designed for single-domain or homogeneous clusters, lacking a hierarchical decision-making architecture for cross-domain heterogeneous unmanned clusters. Existing methods typically employ a single-granularity situation assessment approach, mixing global strategic decisions with local tactical decisions, resulting in low decision-making efficiency. Furthermore, they lack effective means to address the dynamic evolution of adversary strategies faced by cross-domain heterogeneous clusters in dynamic battlefield environments. Because adversary strategies continuously change during red-blue force-on-force confrontations, the environment exhibits non-stationary characteristics for our decision-makers. Traditional methods struggle to effectively predict the changing trends of adversary strategies, leading to decision-making lagging behind changes in the battlefield situation.

[0006] Third, current task allocation mechanisms are mostly static or semi-static, with allocation only performed at the initial stage of the simulation. This fails to dynamically adjust the task weights and groupings of each unmanned platform during execution based on real-time battlefield conditions, leading to low coordination efficiency. Furthermore, it cannot quickly reallocate tasks in unforeseen circumstances such as damage at critical nodes, communication link interruptions, or the emergence of sudden threats, making it difficult to guarantee the continuity and reliability of task execution. In addition, existing methods lack closed-loop feedback from unmanned platform status change data during the simulation to the situation assessment stage, resulting in an open-loop decision-making process that cannot dynamically optimize subsequent decisions based on real-time data during the simulation.

[0007] In summary, existing wargaming methods, when facing cross-domain heterogeneous unmanned swarms, suffer from several shortcomings. These include a lack of a unified data transfer mechanism between model building and decision-making, a lack of hierarchical differentiation between global strategy and local tactics in situation assessment, a lack of effective means to predict the evolution of the opponent's strategies in adversarial decision-making, a lack of dynamic adjustment capabilities for task allocation based on real-time situation, and a lack of closed-loop feedback from state change data to the decision-making process. Consequently, the realism of the simulations, the timeliness of the decisions, and the reliability of the task execution are insufficient to meet the requirements of collaborative operations between cross-domain heterogeneous unmanned swarms. Summary of the Invention

[0008] This invention aims to address the problems of poor simulation realism, decision-making timeliness, and task execution reliability in existing cross-domain heterogeneous unmanned swarm wargaming methods, and proposes a collaborative game wargaming method and system for cross-domain heterogeneous unmanned swarms.

[0009] The technical solution adopted by the present invention to solve the above-mentioned technical problems is as follows:

[0010] In a first aspect, the present invention provides a method for collaborative game wargaming simulation of cross-domain heterogeneous unmanned clusters, the method comprising:

[0011] Step 1: Construct a multi-domain heterogeneous collaborative combat model. The multi-domain heterogeneous collaborative combat model is used to uniformly represent the attributes, cross-domain interaction rules and dynamic battlefield environment of each unmanned platform in a cross-domain heterogeneous unmanned cluster.

[0012] Step 2: Based on the multi-domain heterogeneous collaborative combat model, perform multi-granularity real-time situation assessment to generate global strategic situation and local tactical situation respectively;

[0013] Step 3: Based on the global strategic situation, the confrontation process between the red and blue sides is modeled as a dynamic game problem. The optimal cooperative confrontation strategy is solved using dynamic game theory. Before solving, the evolution trend of the opponent's strategy on both sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results. The strategy network model is used to solve the dynamic game problem and output the optimal cooperative confrontation strategy. At the same time, based on the local tactical situation, an adaptive task allocation mechanism is introduced to dynamically adjust the task weights and groups of each unmanned platform and generate a set of task allocation instructions.

[0014] Step 4: Based on the optimal cooperative confrontation strategy and task allocation instruction set, drive the war game simulation process, and feed back the status change data of each unmanned platform to the situation assessment stage in Step 2 in real time during the simulation. Finally, output the simulation results and debriefing analysis report.

[0015] Furthermore, in step 1, constructing a multi-domain heterogeneous collaborative combat model specifically includes:

[0016] The geographical environment, meteorological conditions, electromagnetic spectrum and communication network in the dynamic battlefield environment are parametrically modeled to form a unified battlefield situation space. The battlefield situation space is a unified digital representation that includes static battlefield environment data and real-time status data storage structure of each unmanned platform. The real-time status data of each unmanned platform is continuously updated in time sequence according to the simulation process.

[0017] Digital modeling is performed on each unmanned platform, which is distributed in multiple combat domains including air, ground, surface and underwater domains. The attributes of the unmanned platforms include kinematic model, dynamic constraints, sensor detection range, weapon payload and communication capability, resulting in digital models and attributes of each unmanned platform.

[0018] Define cross-domain interaction rules between various unmanned platforms, including the collaborative relationships, information exchange protocols, and operational constraints between unmanned platforms in different operational domains;

[0019] The battlefield situation space, the digital models and attributes of each unmanned platform, and the cross-domain interaction rules are bound and integrated to generate the multi-domain heterogeneous collaborative combat model.

[0020] Furthermore, in step 2, a multi-granularity real-time situation assessment is performed, specifically including:

[0021] The battlefield situation space and real-time status data of each unmanned platform are obtained from the multi-domain heterogeneous collaborative combat model.

[0022] Based on a global perspective, the overall battlefield situation is aggregated and analyzed to generate a global strategic situation, which includes the troop deployment situation of both sides, the control situation of key areas, and the evolution trend of the overall battlefield situation.

[0023] Based on a local viewpoint, fine-grained analysis is performed on each unmanned platform within a preset spatial range to generate a local tactical situation. The local tactical situation includes the distribution of each unmanned platform within the local area, the real-time threat level, and the local mission completion progress.

[0024] Furthermore, in step 3, the optimal cooperative adversarial strategy is solved using dynamic game theory, specifically including:

[0025] The collaborative confrontation process of cross-domain heterogeneous unmanned swarms is formalized as a stochastic dynamic game problem. The game participants are defined as the red and blue heterogeneous unmanned swarms, and the game state is defined as the joint state space composed of the current battlefield situation snapshot and the state space of each unmanned platform.

[0026] The stochastic dynamic game problem is defined as a tuple. ,in, Represents the joint state space, This represents the action space of the Red Team's heterogeneous unmanned swarm. This represents the action space of the blue team's heterogeneous unmanned swarm. Represents the state transition probability. Let represent the utility function of the red team's heterogeneous unmanned cluster. Represents the utility function of the blue team's heterogeneous unmanned cluster;

[0027] By solving the Nash equilibrium or approximate Nash equilibrium of the stochastic dynamic game problem, the optimal cooperative confrontation strategy for both the red and blue sides under different situations can be obtained.

[0028] Furthermore, in step 3, the evolution trend of the opponent's strategy between the red and blue sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results, specifically including:

[0029] Introducing a Long Short-Term Memory (LSTM) network to capture snapshots of the battlefield situation at multiple consecutive historical moments. Given the input, obtain the opponent's strategy for the next time step. The predicted feature vector, where, express A snapshot of the battlefield situation at any given moment. Indicates the number of historical moments. express The predicted feature vector of the opponent's strategy at any given moment; the battlefield situation snapshot is obtained by instantiating and filling the battlefield situation space at the corresponding moment.

[0030] Will Snapshot of the battlefield situation at any moment The predicted opponent's strategy is input into the policy network model, which solves the stochastic dynamic game problem to obtain the optimal cooperative adversarial strategy under the current situation. and ,in, express The optimal cooperative adversarial strategy for the red team's heterogeneous unmanned swarm at all times. express The optimal cooperative adversarial strategy for the heterogeneous unmanned cluster of the Blue Team at any time.

[0031] Furthermore, in step 3, solving the approximate Nash equilibrium of the stochastic dynamic game problem specifically includes:

[0032] Using mean-field game theory, the interaction between each unmanned platform in the cross-domain heterogeneous unmanned cluster and the overall state of the cluster is approximated as the interaction between each unmanned platform and the average state of the cluster. For an unmanned platform in our cluster, its influence on all other unmanned platforms in the cross-domain heterogeneous unmanned cluster is aggregated into a mean-field term. Individual value function and the mean field Their evolution is mutually coupled; among them, Indicates the first The individual states of each unmanned platform, i.e., the joint state space. The value function corresponds to a portion of the state components of the unmanned platform. The evolution is described by the Hamilton-Jacobi-Bellman equation, the mean field The evolution is described by the Fokker-Planck equations, and the coupling relationship between the two is achieved through joint solution;

[0033] An approximate Nash equilibrium strategy is obtained by solving the coupled Hamilton-Jacobi-Bellman equation and Fokker-Planck equation.

[0034] Furthermore, in step 3, an adaptive task allocation mechanism is introduced, specifically including:

[0035] The local tactical situation and real-time status data of each unmanned platform are acquired. The real-time status data includes the current position, remaining energy, payload type, and mission execution status.

[0036] Define task set and unmanned platform integration ,in, Indicates the first One task, Indicates the total number of tasks. Indicates the first An unmanned platform Indicates the total number of unmanned platforms; each task Each unmanned platform carries the mission type, required capability vectors, location, and time window attributes. Carry information including current location, remaining energy, payload type, and current mission status;

[0037] Constructing a dynamic utility matrix The dynamic utility matrix Indicates unmanned platform Execute the task The expected return is calculated using the following formula:

[0038] ;

[0039] in, Indicates task Real-time priority weights Indicates unmanned platform Abilities and Tasks Degree of matching requirements Indicates unmanned platform Move to task Resource consumption at the location , , These represent the weighting coefficients for priority weight, capability matching degree, and resource consumption, respectively.

[0040] Based on the dynamic utility matrix, online task reallocation is performed, generating dynamic grouping and task weight adjustment instructions.

[0041] Furthermore, online task reallocation based on the dynamic utility matrix specifically includes:

[0042] When the first preset condition is met, a distributed auction algorithm is used for task allocation: each unmanned platform acts as a bidder, each task acts as an auction item, and each unmanned platform uses the dynamic utility matrix mentioned above. The platform uses the utility value of each task as the basis for bidding, and after multiple rounds of iteration, it obtains an allocation scheme that satisfies individual rationality and budget balance.

[0043] When the second preset condition is met, a rapid reallocation mechanism is triggered: the affected tasks and surviving platforms are identified, and the dynamic utility matrix is ​​used to determine the dynamic utility matrix. The utility value submatrix corresponding to the affected tasks and surviving platforms is used as input. The centralized Hungarian algorithm is executed to redistribute the affected subset within a preset time threshold, generate a new grouping structure and task weight instructions and issue them to each unmanned platform.

[0044] The first preset condition is that the rapid redistribution mechanism is not triggered, and the second preset condition is that at least one of the following is detected: the platform damage ratio in the cross-domain heterogeneous unmanned cluster reaches a preset damage threshold, the critical communication link is interrupted, a sudden threat occurs, or the mission timeliness level reaches a preset emergency level.

[0045] Furthermore, step 4 specifically includes:

[0046] The optimal cooperative countermeasure strategy and the task allocation instruction set are parsed into a sequence of underlying control instructions for each unmanned platform. The sequence of underlying control instructions includes waypoint instructions, sensor on / off instructions, and weapon launch instructions.

[0047] Parallel computing is used to divide the complete simulation process into multiple sub-region simulation threads. The calculation tasks of different unmanned platforms in each sub-region are assigned to multiple computing nodes. Each computing node independently calculates the state update of each unmanned platform under its responsibility within a unit simulation step. Data synchronization between computing nodes ensures that the simulation time step is consistent with the actual combat time step.

[0048] During the simulation, the state change data of each unmanned platform is collected in real time, and the state change data is fed back to the situation assessment stage in step 2 to update the input data for the next round of situation assessment.

[0049] Record all decision points and state change data during the simulation process to form a complete simulation log and debriefing data package, and finally output the simulation results and debriefing analysis report.

[0050] Secondly, the present invention provides a collaborative game wargaming system for cross-domain heterogeneous unmanned clusters, used to implement the collaborative game wargaming method for cross-domain heterogeneous unmanned clusters as described in the first aspect, the system comprising:

[0051] The model building unit is used to build a multi-domain heterogeneous collaborative combat model, which is used to uniformly represent the attributes, cross-domain interaction rules and dynamic battlefield environment of each unmanned platform in a cross-domain heterogeneous unmanned cluster.

[0052] The collaborative game decision-making unit is used to perform multi-granularity real-time situation assessment based on the multi-domain heterogeneous collaborative combat model, and generate global strategic situation and local tactical situation respectively.

[0053] Based on the aforementioned global strategic situation, the confrontation process between the red and blue sides is modeled as a dynamic game problem. Dynamic game theory is used to solve for the optimal cooperative confrontation strategy. Before solving, the evolution trend of the opponent's strategy on both sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results. This strategy network model is used to solve the dynamic game problem and output the optimal cooperative confrontation strategy. Simultaneously, based on the aforementioned local tactical situation, an adaptive task allocation mechanism is introduced to dynamically adjust the task weights and groupings of each unmanned platform, generating a set of task allocation instructions.

[0054] The simulation control unit is used to drive the wargame simulation process based on the optimal cooperative confrontation strategy and task allocation instruction set, and to feed back the status change data of each unmanned platform to the situation assessment stage of the cooperative game decision unit in real time during the simulation, and finally output the simulation results and debriefing analysis report.

[0055] The beneficial effects of this invention are:

[0056] This invention constructs a multi-domain heterogeneous collaborative combat model, providing a unified data foundation and state synchronization mechanism for subsequent situation assessment and decision-making. Through multi-granularity real-time situation assessment, it generates global strategic and local tactical situations, providing a basis for the hierarchical separation of strategic and tactical decisions. By introducing an adversary strategy evolution trend prediction mechanism based on the global strategic situation and adjusting the output of the strategy network accordingly, it improves the adaptability of adversarial decision-making to non-stationary environments. Simultaneously, by introducing an adaptive task allocation mechanism based on the local tactical situation, it achieves real-time dynamic adjustment of the task weights and groupings of each unmanned platform. By feeding back the state change data of each unmanned platform during the simulation to the situation assessment stage in real time, a complete decision-making closed loop is formed, effectively improving the realism of cross-domain heterogeneous unmanned swarm wargame simulations, the timeliness of decision-making, and the reliability of task execution. Attached Figure Description

[0057] Figure 1 A flowchart illustrating the collaborative game wargaming simulation method for cross-domain heterogeneous unmanned clusters provided in this embodiment;

[0058] Figure 2 This is a schematic diagram of the structure of a collaborative game wargaming simulation system for cross-domain heterogeneous unmanned clusters provided in an embodiment. Detailed Implementation

[0059] This invention proposes a collaborative game wargaming method and system for cross-domain heterogeneous unmanned swarms. By constructing a unified multi-domain heterogeneous collaborative combat model, it provides a consistent data foundation for subsequent assessment and decision-making. This allows situation assessment and task allocation to directly access the attribute and environmental data of each unmanned platform, avoiding representational distortion or data inconsistency caused by the separation of the model and decision-making processes. Based on this model, multi-granularity real-time situation assessment can be performed, generating a global strategic situation from a global perspective and a local tactical situation from a fine-grained analysis from a local perspective. This ensures that strategic and tactical decisions obtain the necessary information, avoiding decision-making errors caused by single-granularity assessments. A mechanism for predicting the evolution trend of adversary strategies is introduced based on the global strategic situation. This mechanism predicts the direction of adversary strategy changes through historical situation sequences and corrects the strategy network output accordingly, effectively compensating for the negative impact of dynamic evolution of adversary strategies on decision-making in non-stationary environments. Simultaneously, an adaptive task allocation mechanism is introduced based on the local tactical situation. By constructing a dynamic utility matrix from real-time state data for online task reallocation, the grouping and weights of each unmanned platform can be dynamically adjusted according to changes in the battlefield situation. By feeding back the state change data of each unmanned platform during the simulation process to the situation assessment stage in real time, the situation information on which each subsequent round of decision-making is based is based on the latest platform state, rather than the initial model or a static snapshot of the historical state. The resulting closed-loop iterative mechanism can continuously improve the quality of decision-making, thereby ensuring the dynamic adaptability and decision reliability of cross-domain heterogeneous unmanned swarm wargame simulation.

[0060] The technical solutions in this embodiment will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0061] Figure 1 A flowchart illustrating a collaborative game wargaming simulation method for cross-domain heterogeneous unmanned clusters is shown. Please refer to [link / reference]. Figure 1 The method includes steps 1 to 4:

[0062] Step 1: Construct a multi-domain heterogeneous collaborative combat model. The multi-domain heterogeneous collaborative combat model is used to uniformly represent the attributes, cross-domain interaction rules and dynamic battlefield environment of each unmanned platform in a cross-domain heterogeneous unmanned cluster.

[0063] In this embodiment, the construction of a multi-domain heterogeneous collaborative combat model specifically includes steps 11 to 14:

[0064] Step 11: Perform parametric modeling of the geographical environment, meteorological conditions, electromagnetic spectrum and communication network in the dynamic battlefield environment to form a unified battlefield situation space. The battlefield situation space is a unified digital representation that includes static battlefield environment data and real-time status data storage structure of each unmanned platform. The real-time status data of each unmanned platform is continuously updated in time sequence according to the simulation process.

[0065] Specifically, the dynamic battlefield environment includes geographical environment, meteorological conditions, electromagnetic spectrum, and communication network. The geographical environment mainly includes terrain, elevation, and road network information, represented by a digital elevation model; meteorological conditions mainly include visibility, wind speed, and precipitation data, described by a meteorological parameter field; the electromagnetic spectrum mainly includes interference levels and frequency band occupancy, represented by an electromagnetic situation map; and the communication network mainly includes bandwidth, latency, and connectivity parameters, characterized by a network topology model. After parametrically modeling these four environmental elements, they are superimposed and merged to form a unified battlefield situation space. This battlefield situation space is a unified digital representation containing static battlefield environment data and real-time status data storage structures for each unmanned platform. The static battlefield environment data includes terrain data, elevation data, and road network data—environmental information that does not change or changes slowly over time. The real-time status data of each unmanned platform is continuously updated according to the simulation progress, for example, updating the position, velocity, attitude, and energy status of each unmanned platform every 0.1 seconds.

[0066] Step 12: Perform digital modeling of each unmanned platform. The unmanned platforms are distributed in multiple combat domains, including air domain, ground domain, surface domain and underwater domain. The attributes of the unmanned platforms include kinematic model, dynamic constraints, sensor detection range, weapon payload and communication capability, to obtain the digital model and attributes of each unmanned platform.

[0067] Specifically, the unmanned platforms in a cross-domain heterogeneous unmanned swarm are distributed across multiple operational domains, including air, ground, surface, and underwater domains. Air-domain unmanned platforms include fixed-wing UAVs and rotary-wing UAVs; ground-domain unmanned platforms include unmanned vehicles and ground robots; and surface and underwater unmanned platforms include unmanned surface vessels and underwater vehicles. When digitally modeling the unmanned platforms in each of these operational domains, their attributes need to be defined. Attributes include kinematic models, dynamic constraints, sensor detection range, weapon payload, and communication capabilities. Taking a fixed-wing UAV in the air domain as an example, the kinematic model uses six-degree-of-freedom kinematic equations to describe the changes in its spatial position, velocity, attitude angle, and angular velocity; dynamic constraints include limits such as maximum overload, maximum angular velocity, and maximum acceleration; sensor detection range includes the detection distance and detection angle of sensors such as photoelectric, infrared, and radar; weapon payload includes the range and probability of damage of weapons such as missiles, bombs, and electronic jamming pods; and communication capabilities include the data rate and range of the communication link. For unmanned vehicles in the ground domain, in addition to the above attributes, attributes such as terrain passability and stealth coefficient also need to be considered. Each unmanned platform is instantiated as an independent digital object, and a data interface for interacting with external decision-making algorithms is reserved for it.

[0068] Step 13: Define the cross-domain interaction rules between various unmanned platforms. The cross-domain interaction rules include the collaborative relationship, information exchange protocol and operational constraints between unmanned platforms in different combat domains.

[0069] Specifically, cross-domain interaction rules refer to the collaborative relationships, information exchange protocols, and operational constraints between unmanned platforms in different operational domains. Collaborative relationships refer to the functional complementarity and mission coordination methods between unmanned platforms in different domains. For example, UAVs in the air domain can provide wide-area reconnaissance and target designation services to unmanned vehicles in the ground domain, while unmanned vehicles in the ground domain can provide UAVs in the air domain with precise target identification and positioning services within concealed areas. Information exchange protocols refer to the formats and rules for data exchange between unmanned platforms. Unmanned platforms exchange information through collaborative operational message sets, which include message types such as target allocation messages, collaborative positioning messages, and situational awareness sharing messages. Each message type contains predefined data fields and transmission specifications. Operational constraints refer to the restrictions that each unmanned platform must adhere to during cross-domain collaboration, such as communication distance limitations, energy consumption limitations, and safety interval limitations.

[0070] Step 14: Bind and integrate the battlefield situation space, the digital models and attributes of each unmanned platform, and the cross-domain interaction rules to generate the multi-domain heterogeneous collaborative combat model.

[0071] Specifically, the battlefield situation space is used as a unified environmental data layer, the digital models and attributes of each unmanned platform are used as the entity data layer, and the cross-domain interaction rules are used as the rule data layer. These three layers of data are bound together through a unified association index to form a complete multi-domain heterogeneous collaborative combat model. This multi-domain heterogeneous collaborative combat model is used to uniformly represent the attributes, cross-domain interaction rules, and dynamic battlefield environment of each unmanned platform in a cross-domain heterogeneous unmanned swarm, providing a consistent and complete data foundation for subsequent situation assessment, game theory decision-making, and task allocation.

[0072] Step 2: Based on the multi-domain heterogeneous collaborative combat model, perform multi-granularity real-time situation assessment to generate global strategic situation and local tactical situation respectively.

[0073] In this embodiment, multi-granularity real-time situation assessment is performed, specifically including steps 21 to 23:

[0074] Step 21: Obtain the battlefield situation space and real-time status data of each unmanned platform from the multi-domain heterogeneous collaborative combat model.

[0075] Specifically, the battlefield situation space stores static battlefield environment data, including terrain data, elevation data, and road network data, as well as a storage structure for real-time status data of each unmanned platform. The real-time status data of each unmanned platform includes its current position, current velocity, current attitude, remaining energy, current payload status, and current mission execution status. This data is continuously updated in a time-series manner according to the simulation progress. After obtaining the above data from the multi-domain heterogeneous collaborative combat model, it is used as the input basis for situation assessment.

[0076] Step 22: Based on a global perspective, perform aggregate analysis of the overall battlefield situation to generate a global strategic situation, which includes the troop deployment situation of both sides, the control situation of key areas, and the evolution trend of the overall battlefield situation.

[0077] Specifically, a global perspective refers to focusing on macro-level information such as troop deployment, control of key areas, and trends in the overall battlefield. Aggregate analysis involves comprehensively summarizing the acquired battlefield situational data and real-time status data of various unmanned platforms to extract information reflecting the overall battlefield landscape. Specifically, by statistically analyzing the location distribution, quantity comparison, and type composition of unmanned platforms on both sides, the deployment status of both sides is obtained, such as the troop density and force distribution of the Red side in various areas; by assessing the occupation status and control strength of pre-defined key control areas (such as strategic locations, communication hubs, and supply nodes), the control status of key areas is obtained; and by comparing and judging the trends of situational data over multiple consecutive moments, the overall battlefield situation evolution trend is obtained, such as whether the center of gravity of the battlefield has shifted or whether the dominant side has changed. The above three types of information together constitute the global strategic situation.

[0078] Step 23: Perform fine-grained analysis on each unmanned platform within a preset spatial range based on the local field of view to generate a local tactical situation. The local tactical situation includes the distribution of each unmanned platform in the local area, the real-time threat level, and the local mission completion progress.

[0079] Specifically, local vision refers to narrowing the scope of attention to a predetermined spatial area, such as an area with a radius of several kilometers centered on a specific mission point, or the local combat area where a formation is located. Fine-grained analysis refers to conducting more detailed observation and evaluation of each unmanned platform within the predetermined spatial range. This includes analyzing the specific location distribution of each unmanned platform within the area, their relative positional relationships, and formation patterns to obtain distribution information of each unmanned platform within the local area; comprehensively judging the sensor detection range, weapon payload type, and communication link status of each unmanned platform within the area to assess the immediate threat level of both friendly and enemy forces in the local area to obtain the real-time threat level; and statistically evaluating the current mission progress of each unmanned platform within the area to determine whether each mission is progressing as planned to obtain the local mission completion progress.

[0080] The overall strategic situation and the local tactical situation are used in two parallel decision-making stages, respectively. The overall strategic situation is output to the game strategy solution stage, providing a macro-level situational basis for both sides to formulate game strategy at the overall battlefield level. The local tactical situation is output to the task allocation stage, providing a local situational basis for the online task reallocation and grouping adjustment of various unmanned platforms within a pre-defined spatial range. This distinction in granularity ensures that strategic and tactical decisions can be processed in a hierarchical and parallel manner, avoiding the information mismatch problem caused by a single-granularity situational assessment.

[0081] Step 3: Based on the global strategic situation, the confrontation process between the red and blue sides is modeled as a dynamic game problem. The optimal cooperative confrontation strategy is solved using dynamic game theory. Before solving, the evolution trend of the opponent's strategy between the red and blue sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results. The strategy network model is used to solve the dynamic game problem and output the optimal cooperative confrontation strategy. At the same time, based on the local tactical situation, an adaptive task allocation mechanism is introduced to dynamically adjust the task weights and groups of each unmanned platform and generate a set of task allocation instructions.

[0082] Specifically, step 3, based on the global strategic situation generated in step 2, formalizes the confrontation process of the heterogeneous unmanned swarms of red and blue sides into a dynamic game problem. First, it takes snapshots of the battlefield situation at multiple consecutive historical moments as input, and uses a long short-term memory network to predict the evolution trend of the opponent's strategy on both sides. Then, it corrects the output of the strategy network model based on the prediction results to compensate for the non-stationary fluctuations caused by the dynamic evolution of the opponent's strategy. Then, it solves the Nash equilibrium or approximate Nash equilibrium of the dynamic game problem to obtain the optimal cooperative confrontation strategy of the red and blue sides under the current situation. At the same time, based on the local tactical situation generated in step 2, step 3 obtains the real-time status data of each unmanned platform within a local preset space. According to the adaptability of each unmanned platform to different tasks, it adjusts the task weight and grouping structure of each unmanned platform in real time to generate a set of task allocation instructions, thereby realizing task allocation in parallel with the global game strategy at the local level.

[0083] In this embodiment, the optimal cooperative adversarial strategy is solved using dynamic game theory, specifically including:

[0084] The collaborative confrontation process of cross-domain heterogeneous unmanned swarms is formalized as a stochastic dynamic game problem. The game participants are defined as the red and blue heterogeneous unmanned swarms, and the game state is defined as the joint state space composed of the current battlefield situation snapshot and the state space of each unmanned platform.

[0085] The stochastic dynamic game problem is defined as a tuple. ,in, Represents the joint state space, This represents the action space of the Red Team's heterogeneous unmanned swarm. This represents the action space of the blue team's heterogeneous unmanned swarm. Represents the state transition probability. Let represent the utility function of the red team's heterogeneous unmanned cluster. Represents the utility function of the blue team's heterogeneous unmanned cluster;

[0086] By solving the Nash equilibrium or approximate Nash equilibrium of the stochastic dynamic game problem, the optimal cooperative confrontation strategy for both the red and blue sides under different situations can be obtained.

[0087] Specifically, this embodiment formalizes the cooperative confrontation process between the red and blue teams in a cross-domain heterogeneous unmanned swarm as a stochastic dynamic game problem. The game participants are the red and blue heterogeneous unmanned swarms, and the game state is defined as the joint state space composed of the current battlefield situation snapshot and the state spaces of each unmanned platform. This stochastic dynamic game problem uses tuples... Mathematical representation is performed; based on this, by solving the Nash equilibrium or approximate Nash equilibrium of this stochastic dynamic game problem, the optimal cooperative confrontation strategy of the red and blue sides under different battlefield situations can be obtained.

[0088] In this embodiment, the evolution trend of the opponent's strategy between the red and blue sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results. Specifically, this includes:

[0089] Introducing a Long Short-Term Memory (LSTM) network to capture snapshots of the battlefield situation at multiple consecutive historical moments. Given the input, obtain the opponent's strategy for the next time step. The predicted feature vector, where, express A snapshot of the battlefield situation at any given moment. Indicates the number of historical moments. express The predicted feature vector of the opponent's strategy at any given moment; the battlefield situation snapshot is obtained by instantiating and filling the battlefield situation space at the corresponding moment.

[0090] Will Snapshot of the battlefield situation at any moment The predicted opponent's strategy is input into the policy network model, which solves the stochastic dynamic game problem to obtain the optimal cooperative adversarial strategy under the current situation. and ,in, express The optimal cooperative adversarial strategy for the red team's heterogeneous unmanned swarm at all times. express The optimal cooperative adversarial strategy for the heterogeneous unmanned cluster of the Blue Team at any time.

[0091] Specifically, firstly, this embodiment populates the battlefield situation space constructed in step 1 with real-time data instantiation at corresponding moments to obtain snapshots of the battlefield situation at each moment, ensuring that the situation information on which the prediction is based is consistent with the model environment data; secondly, a long short-term memory network is introduced to continuously... Using a sequence of snapshots of the battlefield situation at historical moments as input, the Long Short-Term Memory (LSTM) network learns the implicit temporal patterns in the historical situation sequence and outputs information about the next moment. The predicted feature vector of the opponent's strategy at each moment; based on this, Snapshot of the battlefield situation at any moment The predicted feature vector and the predicted feature vector are input into the policy network model. The predicted feature vector corrects the output of the policy network model so that the corrected output can predict the direction of the opponent's policy change. The policy network model solves the defined stochastic dynamic game problem and finally obtains the optimal cooperative confrontation strategy of the red and blue sides under the current situation.

[0092] In this embodiment, solving the approximate Nash equilibrium of the stochastic dynamic game problem specifically includes:

[0093] Using mean-field game theory, the interaction between each unmanned platform in the cross-domain heterogeneous unmanned cluster and the overall state of the cluster is approximated as the interaction between each unmanned platform and the average state of the cluster. For an unmanned platform in our cluster, its influence on all other unmanned platforms in the cross-domain heterogeneous unmanned cluster is aggregated into a mean-field term. Individual value function and the mean field Their evolution is mutually coupled; among them, Indicates the first The individual states of each unmanned platform, i.e., the joint state space. The value function corresponds to a portion of the state components of the unmanned platform. The evolution is described by the Hamilton-Jacobi-Bellman equation, the mean field The evolution is described by the Fokker-Planck equations, and the coupling relationship between the two is achieved through joint solution;

[0094] An approximate Nash equilibrium strategy is obtained by solving the coupled Hamilton-Jacobi-Bellman equation and Fokker-Planck equation.

[0095] Specifically, in this embodiment, when solving the approximate Nash equilibrium of the stochastic dynamic game problem, mean-field game theory is used to reduce the solution complexity of large-scale clusters. Its core principle is to approximate the interaction between each unmanned platform and the overall state of the cluster in a cross-domain heterogeneous unmanned cluster as the interaction between each unmanned platform and the average state of the cluster, thereby simplifying the high-dimensional multi-player game problem into an interaction problem between a single individual and a group. Specifically, for one unmanned platform in our cluster, its comprehensive influence on all other unmanned platforms in the cluster is aggregated into a mean-field term. Meanwhile, the individual status of the unmanned platform itself is recorded as , representing the joint state space The values ​​in the diagram correspond to some state components of the unmanned platform; the decision-making gains of the unmanned platform are determined by the individual value function. The function's evolution is described as following the Hamilton-Jacobi-Bellman equations, while the mean field... The evolution over time follows the Fokker-Planck equations, which are coupled due to the mutual influence between individual states and the average state. By jointly solving these two coupled equations, an approximate Nash equilibrium strategy can be obtained, thereby reducing the complexity of the original problem from [previous problem name] while ensuring the rationality of the strategy. Reduced to This significantly improves the solution efficiency of large-scale cross-domain heterogeneous unmanned cluster game strategies.

[0096] In this embodiment, an adaptive task allocation mechanism is introduced, specifically including:

[0097] The local tactical situation and real-time status data of each unmanned platform are acquired. The real-time status data includes the current position, remaining energy, payload type, and mission execution status.

[0098] Define task set and unmanned platform integration ,in, Indicates the first One task, Indicates the total number of tasks. Indicates the first An unmanned platform Indicates the total number of unmanned platforms; each task Each unmanned platform carries the mission type, required capability vectors, location, and time window attributes. Carry information including current location, remaining energy, payload type, and current mission status;

[0099] Constructing a dynamic utility matrix The dynamic utility matrix Indicates unmanned platform Execute the task The expected return is calculated using the following formula:

[0100] ;

[0101] in, Indicates task Real-time priority weights Indicates unmanned platform Abilities and Tasks Degree of matching requirements Indicates unmanned platform Move to task Resource consumption at the location , , These represent the weighting coefficients for priority weight, capability matching degree, and resource consumption, respectively.

[0102] Based on the dynamic utility matrix, online task reallocation is performed, generating dynamic grouping and task weight adjustment instructions.

[0103] Specifically, the process of introducing the adaptive task allocation mechanism in this embodiment is as follows: First, the local tactical situation generated in step 2 and the real-time status data of each unmanned platform are acquired. This real-time status data includes the current position, remaining energy, payload type, and current task execution status of each unmanned platform; then, task sets are constructed respectively. and unmanned platform integration Each of the tasks Each unmanned platform carries mission type, required capability vector, location, and time window attributes. Each payload carries its current location, remaining energy, payload type, and current mission status; a dynamic utility matrix is ​​then constructed based on this information. Each element in the matrix represents an unmanned platform. Execute the task The expected benefits are determined; finally, based on the dynamic utility matrix, online task reallocation is carried out. The optimal allocation scheme is determined according to the expected benefits of each unmanned platform for each task. Dynamic grouping and task weight adjustment instructions are generated and sent to each unmanned platform to ensure that the task division always matches the current local tactical situation.

[0104] In this embodiment, online task reallocation based on the dynamic utility matrix specifically includes:

[0105] When the first preset condition is met, a distributed auction algorithm is used for task allocation: each unmanned platform acts as a bidder, each task acts as an auction item, and each unmanned platform uses the dynamic utility matrix mentioned above. The platform uses the utility value of each task as the basis for bidding, and after multiple rounds of iteration, it obtains an allocation scheme that satisfies individual rationality and budget balance.

[0106] When the second preset condition is met, a rapid reallocation mechanism is triggered: the affected tasks and surviving platforms are identified, and the dynamic utility matrix is ​​used to determine the dynamic utility matrix. The utility value submatrix corresponding to the affected tasks and surviving platforms is used as input. The centralized Hungarian algorithm is executed to redistribute the affected subset within a preset time threshold, generate a new grouping structure and task weight instructions and issue them to each unmanned platform.

[0107] The first preset condition is that the rapid redistribution mechanism is not triggered, and the second preset condition is that at least one of the following is detected: the platform damage ratio in the cross-domain heterogeneous unmanned cluster reaches a preset damage threshold, the critical communication link is interrupted, a sudden threat occurs, or the mission timeliness level reaches a preset emergency level.

[0108] Specifically, this embodiment employs two differentiated processing modes based on the stability of the battlefield situation for online task reallocation based on the dynamic utility matrix: When the simulation process is in a normal state, a distributed auction algorithm is used for task allocation, where each unmanned platform acts as a bidder and bids based on its expected returns for each task. Through multiple rounds of bidding iterations, an allocation scheme that satisfies individual rationality and budget balance is obtained. When abnormal situations are detected, such as the platform attrition ratio reaching a preset attrition threshold, critical communication link interruption, sudden threat occurrence, or task timeliness reaching a preset urgency level, a rapid reallocation mechanism is triggered. This mechanism identifies the affected tasks and surviving platforms, and uses the utility value submatrix corresponding to the affected tasks and surviving platforms in the dynamic utility matrix as input to execute a centralized Hungarian algorithm. Within a preset time threshold, the affected subset is reallocated, and new grouping structures and task weight instructions are generated and issued to each unmanned platform to ensure the continuity of task execution under sudden changes in local tactical situation.

[0109] Individual rationality refers to the rationality of each unmanned platform participating in task allocation as a bidder in the distributed auction algorithm. The expected utility a platform gains after executing its assigned task is no less than its utility level when it does not participate in the allocation; otherwise, the platform will lack the incentive to participate in task allocation, and the allocation scheme will not be able to execute stably. Specifically, in this application, individual rationality is manifested in the fact that when each unmanned platform independently bids based on the expected returns in the dynamic utility matrix, its final allocated return should be at least no less than the benchmark return when the platform is idle or executing a default task.

[0110] Budget balancing refers to the requirement that the sum of the expected benefits of all tasks in a task allocation scheme should satisfy an overall balance condition with the sum of the resource costs consumed by all unmanned platforms in executing the tasks. In other words, the overall expected benefit of each unmanned platform in executing its assigned tasks should not be less than the overall resource cost consumed in executing these tasks. Specifically, in this application, budget balancing is manifested in maintaining a global match between the overall expected benefit and overall resource consumption of each unmanned platform after task reallocation based on a dynamic utility matrix. This avoids ineffective allocation situations where the overall benefit of the allocation scheme is lower than the cost, thereby ensuring that the overall efficiency of the cluster in executing tasks is positive.

[0111] Step 4: Based on the optimal cooperative confrontation strategy and task allocation instruction set, drive the war game simulation process, and feed back the status change data of each unmanned platform to the situation assessment stage in Step 2 in real time during the simulation. Finally, output the simulation results and debriefing analysis report.

[0112] In this embodiment, step 4 specifically includes steps 41 to 44:

[0113] Step 41: Parse the optimal cooperative countermeasure strategy and the task allocation instruction set into a sequence of underlying control instructions for each unmanned platform. The sequence of underlying control instructions includes waypoint instructions, sensor on / off instructions and weapon launch instructions.

[0114] Step 42: Using parallel computing, the complete simulation process is divided into multiple sub-region simulation threads. The calculation tasks of different unmanned platforms in each sub-region are assigned to multiple computing nodes. Each computing node independently calculates the state update of each unmanned platform under its responsibility within a unit simulation step. Data synchronization between computing nodes ensures that the simulation time step is consistent with the actual combat time step.

[0115] Step 43: During the simulation, real-time data on the status changes of each unmanned platform is collected, and the status change data is fed back to the situation assessment stage in Step 2 to update the input data for the next round of situation assessment.

[0116] Step 44: Record all decision points and state change data during the simulation process to form a complete simulation log and debriefing data package, and finally output the simulation results and debriefing analysis report.

[0117] Specifically, step 4 first parses the optimal cooperative countermeasure strategy and task allocation instruction set generated in step 3 into a sequence of underlying control instructions for each unmanned platform. This sequence includes waypoint instructions controlling the movement paths of each unmanned platform, sensor on / off instructions controlling the activation or deactivation of detection equipment on each unmanned platform, and weapon launch instructions controlling the execution of attack or jamming actions by each unmanned platform. Then, using parallel computing, the complete simulation process is divided into multiple sub-region simulation threads according to spatial regions or operational domains. The solution tasks for different unmanned platforms in each sub-region are assigned to multiple computing nodes. Each computing node independently calculates the state updates of the unmanned platforms it is responsible for within a unit simulation step. Simultaneously, data synchronization between computing nodes ensures... The simulation's time step is consistent with the actual combat time step, thus meeting the computational requirements of large-scale cluster real-time simulations. During the simulation process, real-time data on the position, velocity, attitude, remaining energy, and mission execution status of each unmanned platform are collected and fed back to the situation assessment stage in step 2 to update the input data for the next round of situation assessment, ensuring that the situation information used for subsequent decisions is continuously updated based on the latest platform status. Simultaneously, all decision points and status change data during the simulation are recorded to form a complete simulation log and debriefing data package. After the simulation, the final output includes key event annotations, can be replayed along a timeline, and allows for in-depth analysis of specific unmanned platforms or specific missions.

[0118] The following scenario uses a cross-domain heterogeneous unmanned swarm confrontation between red and blue teams as a model, combined with... Figure 1 The overall flowchart shown provides a detailed explanation of the method of this invention. The simulation scenario is set as follows: the red team consists of a cross-domain heterogeneous unmanned swarm composed of 20 reconnaissance and strike integrated fixed-wing UAVs, 50 small quadcopter UAVs, and 30 ground-based unmanned vehicles for follow-up combat; the blue team is the defending side, equipped with ground firepower units and electronic jamming equipment. The specific process includes the following steps:

[0119] First, execute step 1 to construct a multi-domain heterogeneous collaborative combat model. Parametric modeling of the dynamic battlefield environment is performed: a digital elevation model containing urban landmarks and hilly terrain is loaded as geographic environment data; initial weather conditions are set as cloudy with moderate visibility as meteorological parameters; a continuously operating interference source is simulated in the electromagnetic space as electromagnetic spectrum data; and a communication network topology with bandwidth and delay parameters is constructed as communication network data. These four types of data are superimposed and merged to form a unified battlefield situation space. Digital modeling of each unmanned platform is then performed: 20 fixed-wing UAVs, 50 quadcopter UAVs, and 30 ground unmanned vehicles are instantiated as independent digital objects. For the fixed-wing UAVs, a six-degree-of-freedom kinematic model, maximum overload constraints, electro-optical infrared sensor detection range, and air-to-ground missile weapon payload parameters are bound; for the quadcopter UAVs, a three-degree-of-freedom kinematic model, angular velocity constraints, and electronic jamming pod payload parameters are bound; and for the ground unmanned vehicles, a ground kinematic model, terrain passability coefficient, and reconnaissance equipment payload parameters are bound. Define cross-domain interaction rules between various unmanned platforms: stipulate that quadcopter UAVs are responsible for reconnaissance between low-altitude building clusters, and transmit target coordinates to high-altitude fixed-wing UAVs via tactical data links, with the fixed-wing UAVs carrying out the strikes; ground unmanned vehicles follow the quadcopter UAVs, and conduct ground assaults after receiving target instructions. The aforementioned battlefield situation space, the digital models and attributes of each unmanned platform, and the cross-domain interaction rules are bound and integrated to generate a multi-domain heterogeneous collaborative combat model.

[0120] Then, step 2 is executed, performing a multi-granularity real-time situation assessment based on the model. After the simulation begins, the battlefield situation is refreshed at a frequency of 20 Hz, that is, real-time status data of each unmanned platform is obtained from the multi-domain heterogeneous collaborative combat model every 0.05 seconds. The acquired data is processed in multiple granularities: under the global view, the position distribution and quantity comparison of all unmanned platforms of both the red and blue sides are aggregated and analyzed to generate a global strategic situation. This global strategic situation indicates that the red force is in an offensive state, and the blue force's defensive focus is located in the hilly area on the left. Under the local view, a fine-grained analysis is performed on the red unmanned platforms within a preset spatial range near the hills on the left, assessing the threat level and mission progress in this area to generate a local tactical situation. This local tactical situation shows that an attack formation is suppressed by blue force ground fire.

[0121] Next, step 3 is executed, which includes two parallel stages. In the game-theoretic strategy solution stage, the collaborative game decision-making unit acquires the global strategic situation, formalizing the confrontation process between the red and blue sides into a stochastic dynamic game problem. The game participants are defined as heterogeneous unmanned swarms of red and blue sides, and the game state is defined as a joint state space composed of the current battlefield situation snapshot and the state spaces of each unmanned platform. Before solving this game problem, the Long Short-Term Memory (LSTM) network takes five consecutive historical battlefield situation snapshots as input and outputs a predictive feature vector of the opponent's strategy for the blue side's firepower units' possible maneuver direction and firing frequency at the next moment. This predictive feature vector is used to correct the output of the strategy network model, enabling the strategy network model to predict the possibility that the blue side's firepower units will move to the hills to the north. The corrected strategy network model solves for the Nash equilibrium strategy, calculating that the red side's locally optimal collaborative confrontation strategy in this area is to release a smokescreen and then launch a flanking attack. Meanwhile, in the adaptive task allocation phase, the task allocation module, based on the local tactical situation, acquires real-time status data of each unmanned platform in the attack formation. It detects that the original forward reconnaissance mission is no longer feasible under fire threat. Therefore, it redefines the task set (including three sub-tasks: destroying enemy fire points, suppressing enemy sensors, and flanking maneuvers) and the unmanned platform set (including the remaining 4 ground unmanned vehicles and 7 quadcopter drones), constructing a dynamic utility matrix. Based on the expected benefits of each platform performing each task in this matrix, online task reallocation is performed. Two quadcopters carrying electronic jamming devices are assigned to the enemy sensor suppression sub-task, and two ground unmanned vehicles and three quadcopter drones are assigned to the flanking maneuver sub-task, generating a task allocation instruction set. This task allocation instruction is completed and output within 50 milliseconds.

[0122] Finally, step 4 is executed to drive the wargaming simulation process and form a closed-loop iteration. The simulation control and calculation unit receives the optimal cooperative confrontation strategy and task allocation instruction set generated in step 3, and parses it into a sequence of low-level control instructions for each unmanned platform, including waypoint instructions for controlling the quadcopter's forward reconnaissance, sensor switch instructions for controlling the activation of the electronic jamming pod, and weapon launch instructions for controlling the fixed-wing UAV's standby strike. Parallel computing is used to divide the complete simulation process into multiple sub-region simulation threads according to the region. Each computing node independently calculates the state update of the unmanned platform it is responsible for within a unit simulation step, and data synchronization between computing nodes ensures that the simulation time step is consistent with the actual combat time step. During the simulation process, the position and energy state change data of each unmanned platform are collected in real time, and these state change data are fed back to the situation assessment stage in step 2 to update the input data for the next round of situation assessment. The data recording module synchronously records the key decision point (discovery of fire points, strategy selection of flanking attack after releasing smoke screen, task reassignment scheme) and the state snapshots of each relevant unmanned platform, and stores them in the review database. The above process is repeated until the preset combat objectives are achieved or time resources are exhausted, and finally the simulation results and debriefing analysis report are output.

[0123] In summary, the collaborative game wargaming method for cross-domain heterogeneous unmanned swarms provided in this embodiment, by constructing a unified multi-domain heterogeneous collaborative combat model, provides a structured data foundation and state synchronization mechanism for situation assessment and decision-making, ensuring information continuity between model construction and subsequent stages. By executing multi-granularity real-time situation assessments, it generates global strategic and local tactical situations respectively, enabling macro-level game decisions and micro-level task allocation to obtain the necessary situational information support, avoiding information mismatch caused by single-granularity assessments. In game adversarial decision-making, a long short-term memory network is introduced to extract temporal features from historical situation sequences, predict the evolution trend of the opponent's strategy, and correct the strategy network output accordingly, effectively compensating for the dynamic adversarial environment. The negative impact of non-stationary changes in the opponent's strategy on decision-making effectiveness is addressed. In task allocation, a dynamic utility matrix is ​​constructed to quantitatively evaluate the expected returns of each unmanned platform performing each task in real time, enabling dynamic adjustment of task weights and grouping structures. A differentiated synergy between distributed auction algorithms and centralized Hungarian algorithms balances allocation efficiency under normal conditions with rapid reallocation capabilities under emergency conditions. Simultaneously, the state change data of each unmanned platform during the simulation process is fed back to the situation assessment stage in real time, forming a closed-loop iterative mechanism that ensures that the situation information used for subsequent decisions is continuously updated based on the latest platform states. A parallel computing architecture is employed to decompose the simulation process into multiple sub-regions and perform multi-node collaborative computation, ensuring the real-time performance of large-scale heterogeneous cluster simulations. These mechanisms work together to improve the systematic nature, dynamic adaptability, and decision reliability of cross-domain heterogeneous unmanned cluster wargame simulations throughout the entire process from model representation, situation awareness, strategy generation, task allocation to real-time feedback.

[0124] Based on the above technical solutions, this embodiment also proposes a collaborative game wargaming system for cross-domain heterogeneous unmanned clusters, used to implement the collaborative game wargaming method for cross-domain heterogeneous unmanned clusters as described in the embodiment. Please refer to [link to relevant documentation]. Figure 2 The system includes:

[0125] The model building unit is used to build a multi-domain heterogeneous collaborative combat model, which is used to uniformly represent the attributes, cross-domain interaction rules and dynamic battlefield environment of each unmanned platform in a cross-domain heterogeneous unmanned cluster.

[0126] The collaborative game decision-making unit is used to perform multi-granularity real-time situation assessment based on the multi-domain heterogeneous collaborative combat model, and generate global strategic situation and local tactical situation respectively.

[0127] Based on the aforementioned global strategic situation, the confrontation process between the red and blue sides is modeled as a dynamic game problem. Dynamic game theory is used to solve for the optimal cooperative confrontation strategy. Before solving, the evolution trend of the opponent's strategy on both sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results. This strategy network model is used to solve the dynamic game problem and output the optimal cooperative confrontation strategy. Simultaneously, based on the aforementioned local tactical situation, an adaptive task allocation mechanism is introduced to dynamically adjust the task weights and groupings of each unmanned platform, generating a set of task allocation instructions.

[0128] The simulation control unit is used to drive the wargame simulation process based on the optimal cooperative confrontation strategy and task allocation instruction set, and to feed back the status change data of each unmanned platform to the situation assessment stage of the cooperative game decision unit in real time during the simulation, and finally output the simulation results and debriefing analysis report.

[0129] It is understood that since the cross-domain heterogeneous unmanned cluster collaborative game wargaming simulation system described in this embodiment is a system for implementing the cross-domain heterogeneous unmanned cluster collaborative game wargaming simulation method described in the embodiment, the system disclosed in the embodiment is relatively simple to describe because it corresponds to the method disclosed in the embodiment. For relevant parts, please refer to the description of the method, and it will not be repeated here.

Claims

1. A method for collaborative game wargaming simulation of cross-domain heterogeneous unmanned clusters, characterized in that, The method includes: Step 1: Construct a multi-domain heterogeneous collaborative combat model. The multi-domain heterogeneous collaborative combat model is used to uniformly represent the attributes, cross-domain interaction rules and dynamic battlefield environment of each unmanned platform in a cross-domain heterogeneous unmanned cluster. Step 2: Based on the multi-domain heterogeneous collaborative combat model, perform multi-granularity real-time situation assessment to generate global strategic situation and local tactical situation respectively; Step 3: Based on the global strategic situation, the confrontation process between the red and blue sides is modeled as a dynamic game problem. The optimal cooperative confrontation strategy is solved using dynamic game theory. Before solving, the evolution trend of the opponent's strategy on both sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results. The strategy network model is used to solve the dynamic game problem and output the optimal cooperative confrontation strategy. At the same time, based on the local tactical situation, an adaptive task allocation mechanism is introduced to dynamically adjust the task weights and groups of each unmanned platform and generate a set of task allocation instructions. Step 4: Based on the optimal cooperative confrontation strategy and task allocation instruction set, drive the war game simulation process, and feed back the status change data of each unmanned platform to the situation assessment stage in Step 2 in real time during the simulation. Finally, output the simulation results and debriefing analysis report. Step 3 involves using dynamic game theory to solve for the optimal cooperative adversarial strategy, specifically including: The collaborative confrontation process of cross-domain heterogeneous unmanned swarms is formalized as a stochastic dynamic game problem. The game participants are defined as the red and blue heterogeneous unmanned swarms, and the game state is defined as the joint state space composed of the current battlefield situation snapshot and the state space of each unmanned platform. The stochastic dynamic game problem is defined as a tuple. ,in, Represents the joint state space, This represents the action space of the Red Team's heterogeneous unmanned swarm. This represents the action space of the blue team's heterogeneous unmanned swarm. Represents the state transition probability. Let represent the utility function of the red team's heterogeneous unmanned cluster. Represents the utility function of the blue team's heterogeneous unmanned cluster; By solving the Nash equilibrium or approximate Nash equilibrium of the stochastic dynamic game problem, the optimal cooperative confrontation strategy of the red and blue sides under different situations can be obtained. In step 3, the evolution trend of the opponent's strategy between the red and blue sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results. Specifically, this includes: Introducing a Long Short-Term Memory (LSTM) network to capture snapshots of the battlefield situation at multiple consecutive historical moments. Given the input, obtain the opponent's strategy for the next time step. The predicted feature vector, where, express A snapshot of the battlefield situation at any given moment. Indicates the number of historical moments. express The predicted feature vector of the opponent's strategy at any given moment; the battlefield situation snapshot is obtained by instantiating and filling the battlefield situation space at the corresponding moment. Will Snapshot of the battlefield situation at any moment The predicted opponent's strategy is input into the policy network model, which solves the stochastic dynamic game problem to obtain the optimal cooperative adversarial strategy under the current situation. and ,in, express The optimal cooperative adversarial strategy for the red team's heterogeneous unmanned swarm at all times. express The optimal cooperative adversarial strategy for the heterogeneous unmanned swarm of Blue Team at any given moment; Step 3 involves solving for the approximate Nash equilibrium of the stochastic dynamic game problem, specifically including: Using mean-field game theory, the interaction between each unmanned platform in the cross-domain heterogeneous unmanned cluster and the overall state of the cluster is approximated as the interaction between each unmanned platform and the average state of the cluster. For an unmanned platform in our cluster, its influence on all other unmanned platforms in the cross-domain heterogeneous unmanned cluster is aggregated into a mean-field term. Individual value function and the mean field Their evolution is mutually coupled; among them, Indicates the first The individual states of each unmanned platform, i.e., the joint state space. The value function corresponds to a portion of the state components of the unmanned platform. The evolution is described by the Hamilton-Jacobi-Bellman equation, the mean field The evolution is described by the Fokker-Planck equations, and the coupling relationship between the two is achieved through joint solution; An approximate Nash equilibrium strategy is obtained by solving the coupled Hamilton-Jacobi-Bellman equation and Fokker-Planck equation.

2. The collaborative game wargaming simulation method for cross-domain heterogeneous unmanned clusters according to claim 1, characterized in that, Step 1 involves constructing a multi-domain heterogeneous collaborative combat model, specifically including: The geographical environment, meteorological conditions, electromagnetic spectrum and communication network in the dynamic battlefield environment are parametrically modeled to form a unified battlefield situation space. The battlefield situation space is a unified digital representation that includes static battlefield environment data and real-time status data storage structure of each unmanned platform. The real-time status data of each unmanned platform is continuously updated in time sequence according to the simulation process. Digital modeling is performed on each unmanned platform, which is distributed in multiple combat domains including air, ground, surface and underwater domains. The attributes of the unmanned platforms include kinematic model, dynamic constraints, sensor detection range, weapon payload and communication capability, resulting in digital models and attributes of each unmanned platform. Define cross-domain interaction rules between various unmanned platforms, including the collaborative relationships, information exchange protocols, and operational constraints between unmanned platforms in different operational domains; The battlefield situation space, the digital models and attributes of each unmanned platform, and the cross-domain interaction rules are bound and integrated to generate the multi-domain heterogeneous collaborative combat model.

3. The collaborative game wargaming simulation method for cross-domain heterogeneous unmanned clusters according to claim 2, characterized in that, Step 2 involves performing a multi-granularity real-time situation assessment, specifically including: The battlefield situation space and real-time status data of each unmanned platform are obtained from the multi-domain heterogeneous collaborative combat model. Based on a global perspective, the overall battlefield situation is aggregated and analyzed to generate a global strategic situation, which includes the troop deployment situation of both sides, the control situation of key areas, and the evolution trend of the overall battlefield situation. Based on a local viewpoint, fine-grained analysis is performed on each unmanned platform within a preset spatial range to generate a local tactical situation. The local tactical situation includes the distribution of each unmanned platform within the local area, the real-time threat level, and the local mission completion progress.

4. The collaborative game wargaming simulation method for cross-domain heterogeneous unmanned clusters according to claim 1, characterized in that, Step 3 introduces an adaptive task allocation mechanism, which specifically includes: The local tactical situation and real-time status data of each unmanned platform are acquired. The real-time status data includes the current position, remaining energy, payload type, and mission execution status. Define task set and unmanned platform integration ,in, Indicates the first One task, Indicates the total number of tasks. Indicates the first An unmanned platform Indicates the total number of unmanned platforms; each task Each unmanned platform carries the mission type, required capability vectors, location, and time window attributes. Carry information including current location, remaining energy, payload type, and current mission status; Constructing a dynamic utility matrix The dynamic utility matrix Indicates unmanned platform Execute the task The expected return is calculated using the following formula: ； in, Indicates task Real-time priority weights Indicates unmanned platform Abilities and Tasks Degree of matching requirements Indicates unmanned platform Move to task Resource consumption at the location , , These represent the weighting coefficients for priority weight, capability matching degree, and resource consumption, respectively. Based on the dynamic utility matrix, online task reallocation is performed, generating dynamic grouping and task weight adjustment instructions.

5. The collaborative game wargaming simulation method for cross-domain heterogeneous unmanned clusters according to claim 4, characterized in that, Online task reallocation based on the dynamic utility matrix specifically includes: When the first preset condition is met, a distributed auction algorithm is used for task allocation: each unmanned platform acts as a bidder, each task acts as an auction item, and each unmanned platform uses the dynamic utility matrix mentioned above. The platform uses the utility value of each task as the basis for bidding, and after multiple rounds of iteration, it obtains an allocation scheme that satisfies individual rationality and budget balance. When the second preset condition is met, a rapid reallocation mechanism is triggered: the affected tasks and surviving platforms are identified, and the dynamic utility matrix is ​​used to determine the dynamic utility matrix. The utility value submatrix corresponding to the affected tasks and surviving platforms is used as input. The centralized Hungarian algorithm is executed to redistribute the affected subset within a preset time threshold, generate a new grouping structure and task weight instructions and issue them to each unmanned platform. The first preset condition is that the rapid redistribution mechanism is not triggered, and the second preset condition is that at least one of the following is detected: the platform damage ratio in the cross-domain heterogeneous unmanned cluster reaches a preset damage threshold, the critical communication link is interrupted, a sudden threat occurs, or the mission timeliness level reaches a preset emergency level.

6. The collaborative game wargaming simulation method for cross-domain heterogeneous unmanned clusters according to claim 1, characterized in that, Step 4 specifically includes: The optimal cooperative countermeasure strategy and the task allocation instruction set are parsed into a sequence of underlying control instructions for each unmanned platform. The sequence of underlying control instructions includes waypoint instructions, sensor on / off instructions, and weapon launch instructions. Parallel computing is used to divide the complete simulation process into multiple sub-region simulation threads. The calculation tasks of different unmanned platforms in each sub-region are assigned to multiple computing nodes. Each computing node independently calculates the state update of each unmanned platform under its responsibility within a unit simulation step. Data synchronization between computing nodes ensures that the simulation time step is consistent with the actual combat time step. During the simulation, the state change data of each unmanned platform is collected in real time, and the state change data is fed back to the situation assessment stage in step 2 to update the input data for the next round of situation assessment. Record all decision points and state change data during the simulation process to form a complete simulation log and debriefing data package, and finally output the simulation results and debriefing analysis report.

7. A collaborative game wargaming simulation system for cross-domain heterogeneous unmanned clusters, characterized in that, The system is used to implement the collaborative game wargaming simulation method for cross-domain heterogeneous unmanned clusters as described in any one of claims 1 to 6, the system comprising: The model building unit is used to build a multi-domain heterogeneous collaborative combat model, which is used to uniformly represent the attributes, cross-domain interaction rules and dynamic battlefield environment of each unmanned platform in a cross-domain heterogeneous unmanned cluster. The collaborative game decision-making unit is used to perform multi-granularity real-time situation assessment based on the multi-domain heterogeneous collaborative combat model, and generate global strategic situation and local tactical situation respectively. Based on the aforementioned global strategic situation, the confrontation process between the red and blue sides is modeled as a dynamic game problem. Dynamic game theory is used to solve for the optimal cooperative confrontation strategy. Before solving, the evolution trend of the opponent's strategy on both sides is predicted based on the historical situation sequence, and the output of the strategy network model is corrected according to the prediction results. This strategy network model is used to solve the dynamic game problem and output the optimal cooperative confrontation strategy. Simultaneously, based on the aforementioned local tactical situation, an adaptive task allocation mechanism is introduced to dynamically adjust the task weights and groupings of each unmanned platform, generating a set of task allocation instructions. The simulation control unit is used to drive the wargame simulation process based on the optimal cooperative confrontation strategy and task allocation instruction set, and to feed back the status change data of each unmanned platform to the situation assessment stage of the cooperative game decision unit in real time during the simulation, and finally output the simulation results and debriefing analysis report.

Images