A combat force rapid reorganization method based on an architecture model

Abstract

The present application relates to the technical field of operational network analysis, and particularly relates to a combat force rapid reorganization method based on an architecture model, and proposes the following scheme: entities in a combat system are abstracted as nodes, a force deployment model containing reconnaissance nodes, command nodes and force nodes is constructed; under a random enemy situation, a task demand sequence is generated based on initial attributes of target nodes; a force allocation optimization model is established, taking task priority satisfaction, resource utilization and load balancing as optimization objectives, and a force allocation scheme is generated in combination with reinforcement learning technology; and under the premise of meeting resource, time and path feasibility constraint conditions, an optimal path of force nodes to target nodes is planned; the present application realizes task scheduling and force allocation optimization under a dynamic battlefield environment, improves combat response speed and resource utilization efficiency, and is suitable for a multi-task cooperative combat scene.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of operational network analysis, and particularly relates to a method for rapid reorganization of operational forces based on an architecture model. BACKGROUND

[0002] Under the current complex battlefield environment, the dynamic nature and diversity of operational tasks pose higher requirements for force deployment and scheduling. In the multi-task, multi-node cooperative combat scene, the optimization of force allocation not only needs to consider resource constraints, but also needs to meet the multi-dimensional constraints such as task priority, time window and path feasibility. However, the existing force scheduling methods mostly use static allocation models, which cannot adapt to the dynamic changes of the battlefield environment, such as the random occurrence of enemy situation, the fluctuation of path passability and the finiteness of resources. In addition, the traditional method lacks effective mechanism in task conflict detection, dynamic path planning and load balancing, which is easy to cause resource waste or task failure.

[0003] For example, the patent application with publication number CN115641235A provides a method for analyzing high-order structure of heterogeneous operational network based on operational motif clustering, including: regarding a subgraph mode composed of multiple equipment nodes in the operational network as an operational motif; converting the operational network into an operational motif adjacency matrix; dividing the corresponding operational network into a connected piece set according to the operational motif adjacency matrix; for each connected piece in the connected piece set, performing spectral clustering on its operational motif adjacency matrix, and gradually dividing it into clusters with node size not greater than a set threshold; aggregating all clusters of operational motifs in the connected piece to obtain the functional clusters of operational motifs, and then aggregating the functional clusters of all interested operational motifs to obtain the functional cluster set of different operational motifs in the complete operational network. The high-order functional clusters in the operational network can be found according to different operational motifs of interest.

[0004] The above patent has the problems proposed in the background: it is difficult to balance task priority, resource utilization and load balancing in a dynamic battlefield environment, and lacks effective dynamic feedback mechanism and real-time path planning capability, resulting in insufficient flexibility and efficiency of force allocation. In order to solve the above problems, the present application designs a method for rapid reorganization of operational forces based on an architecture model. SUMMARY

[0005] The technical problem solved by the present application is to provide a combat force rapid reorganization method based on an architecture model, which addresses the deficiencies of the prior art.

[0006] To achieve the above object, the present application provides the following technical scheme.

[0007] A combat force rapid reorganization method based on an architecture model, the method comprising:

[0008] Abstracting entities in a combat system as nodes and constructing a force deployment model based on the interaction relationship between the entities;

[0009] According to the random enemy situation detected by the nodes, updating the target nodes in the force deployment model, and generating a task demand sequence according to the initial attributes of the target nodes;

[0010] According to the task demand sequence and the force deployment model, establishing a force allocation optimization model, and generating a force allocation scheme through the force allocation optimization model according to the resource constraints and the task priority;

[0011] According to the force allocation scheme, planning the optimal path of the force from the force node to the target node.

[0012] The construction of the force deployment model comprises:

[0013] Dividing the entities in the combat system into reconnaissance nodes, command nodes and force nodes according to their functions, and adding attribute values to each node according to the functions of the entities, wherein the attribute values include position coordinates, task status, force status, corresponding time delay and task priority;

[0014] Establishing an interaction relationship according to the node type, wherein the interaction relationship includes information flow, instruction flow and resource flow;

[0015] Adding dynamic attributes to the nodes and the interaction relationship, and constructing the force deployment model in the form of a directed graph.

[0016] The dynamic attributes include:

[0017] adding a blank task sequence, wherein the blank task sequence comprises a task time window and a task timing dependency, the task time window is used to attach time constraints for tasks and nodes, including the earliest start time, the latest completion time and the task duration, and the task timing dependency is used to establish a priority order diagram between tasks according to the dependency relationship of multiple tasks;

[0018] A task conflict detection mechanism is established, and when a reconnaissance node or a force node is called by multiple tasks at the same time, the resource use order is adjusted through task priority and task time window;

[0019] The terrain change information is collected by the reconnaissance node, including obstacle generation, path blocking and passable width change, and a dynamic path feasibility diagram between nodes is established according to the battlefield terrain.

[0020] The generated task demand sequence comprises:

[0021] The initial attributes of the task are obtained from the target node by the reconnaissance node, wherein the initial attributes include target position, threat level and task type;

[0022] The task priority is calculated according to the initial attributes, the task demand sequence is sorted according to the task priority, and the task in the task demand sequence is updated according to descending order, and the calculation formula of the task priority is:

[0023]

[0024] Wherein, P represents the task priority, W represents the threat level of the target node, ω1 represents the weight of the threat level, T s represents the earliest start time of the task, T e represents the latest completion time of the task, T c represents the current time, ω2 represents the weight of the time constraint, D represents the average distance between the force node and the target node, F represents the path accessibility average value between the force node and the target node, ω3 represents the weight of the path constraint, R d represents the difference between the resource demand of the target node and the current remaining resource, and ω4 represents the weight of the resource matching degree.

[0025] The optimization target of the force allocation optimization model comprises task priority satisfaction, resource utilization and force node load balancing.

[0026] The constraint conditions of the force allocation optimization model comprise:

[0027] The resource demand of each task is less than or equal to the total resource capacity of the force node;

[0028] The total amount of resource allocation of each force node is less than or equal to its available capacity.

[0029] The resource allocation of each task is completed within its task time window;

[0030] The resource transfer path between the force node and the target node meets the path feasibility requirement.

[0031] The force allocation scheme is generated by the force allocation optimization model, comprising:

[0032] Input the task demand sequence and the force deployment model, generate a task-node state diagram, the vertices of the task-node state diagram represent the task nodes and the force nodes, the edges represent the resource allocation possibility between the task nodes and the force nodes, and the edge weights are determined by the task priority, path distance, passing probability and resource matching degree;

[0033] Create a reinforcement learning agent, initialize the state space, action space and reward function of the reinforcement learning agent, take the task-node state diagram as the input parameter of the reinforcement learning agent, and output the force allocation scheme.

[0034] The calculation formula of the reward function is:

[0035]

[0036] Wherein, Z represents the reward value, i represents a single task in the task demand sequence, N represents the total number of tasks in the task demand sequence, M represents the total number of force nodes allocated by the task i, j represents a single force node allocated by the task i, represents the weight of priority satisfaction degree, represents the weight of resource utilization rate, represents the weight of load balancing, P i represents the task priority of task i, η represents the gain coefficient of resource allocation to task priority, R i,j represents the resource demand satisfaction degree of force node j to task i, u j represents the allocated resource amount of force node j, U j represents the total resource capacity of force node j, λ j represents the load standard deviation of force node j, U represents the average value of all node allocation resources in the force allocation scheme.

[0037] The reinforcement learning agent comprises:

[0038] According to the state space, calculate the allocation probability between the task nodes and the force nodes, recombine the allocation probability through a random strategy, and generate an initial scheme;

[0039] Conduct constraint checking on the initial scheme in the action space, judge whether the constraint condition is satisfied, if not satisfied, re-generate the initial scheme in the state space, and set the reward value of this scheme to 0, if satisfied, calculate the reward value of the initial scheme according to the reward function;

[0040] Judge whether the reward value satisfies the condition that the three consecutive rounds are less than the set threshold, if satisfied, output the initial scheme corresponding to the maximum reward value in the three rounds as the force allocation scheme, if not satisfied, reserve the initial scheme in the state space and optimize the loss function of the random strategy according to the reward value, and re-generate the initial scheme.

[0041] The optimal path comprises:

[0042] Initialize the path graph, set the force node in the force allocation scheme as the starting point and the target node as the ending point, and mark them in the path graph;

[0043] Set the feasible path between the force node and the target node as the edge of the path graph, and the weight of each edge represents the comprehensive cost of the path, wherein the comprehensive cost comprises path distance and path resource demand;

[0044] Select the edge with the minimum cumulative weight as the optimal path of the force node to the target node, traverse each force node in the force allocation scheme, and output the optimal path scheme.

[0045] Compared with the prior art, the beneficial effects of the present application are:

[0046] The present application abstracts entities in the combat system as nodes, constructs a dynamic force deployment model, optimizes the task allocation scheme combined with reinforcement learning technology, and improves the resource allocation efficiency and the reliability of task completion by using real-time path planning; at the same time, the task priority, resource utilization and load balancing are comprehensively considered, and the battlefield environment changes are dynamically adapted, so that efficient, flexible and intelligent force rapid reorganization is realized, and the combat response speed and task execution effect are improved. BRIEF DESCRIPTION OF DRAWINGS

[0047] Other features, objects and advantages of the present application will become more apparent from the following detailed description of non-limiting embodiments made with reference to the accompanying drawings:

[0048] Figure 1 Fig. 1 is a flowchart of a combat force rapid reorganization method based on an architecture model according to Embodiment 1 of the present application;

[0049] Figure 2 Fig. 2 is a network mode schematic diagram of a force deployment model according to Embodiment 1 of the present application;

[0050] Figure 3The force allocation flowchart of embodiment 1 of the present application. DETAILED DESCRIPTION

[0051] The technical solutions in the embodiments of the present application will be clearly and completely described in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application.

[0052] Embodiment 1

[0053] Please refer to Figure 1 An embodiment provided by the present application is a combat force rapid reorganization method based on an architecture model, and the specific steps of the method are as follows:

[0054] S1: Abstract entities in a combat architecture as nodes, and build a force deployment model based on the interaction relationship between the entities;

[0055] In this step, key entities in the combat architecture are abstracted as nodes, such as reconnaissance nodes, command nodes and force nodes, and the nodes are classified according to functions. The interaction relationship between the nodes is represented by information flow, instruction flow and resource flow, a dynamic directed graph model is constructed, the vertex represents the node, the edge represents the interaction relationship, and each node is attached with attribute values (such as position coordinates, task state and resource state) to reflect the structure and state of the combat architecture;

[0056] S2: Update target nodes in the force deployment model to generate a task demand sequence;

[0057] In this step, the target nodes in the force deployment model are updated according to the real-time enemy situation data provided by the reconnaissance nodes. The initial attributes of each target node include target position, threat level and task type. The task demand sequence is generated according to the attributes of the target nodes, the task demand contains priority, resource demand and time window, etc. information, and is sorted by priority to provide basis for subsequent allocation;

[0058] S3: Build a force allocation optimization model, and generate a force allocation scheme through the force allocation optimization model according to the resource limitation and the task priority;

[0059] In this step, the force allocation optimization model is built based on the task demand sequence and the force deployment model, the objective function and the constraint condition are set, and the optimization objectives include the task priority satisfaction degree, the resource utilization rate and the load balancing. The force allocation scheme is generated by using the reinforcement learning technology, and the matching relationship between the tasks and the force nodes, the resource allocation amount and the task execution priority are outputted;

[0060] S4: According to the force allocation scheme, plan the optimal path of the force from the force node to the target node;

[0061] In this step, the geographic information and path accessibility of task nodes and force nodes are extracted according to the force allocation scheme, and a path graph model is constructed. Through path weight calculation, factors such as distance, accessibility and resource consumption are considered, and a graph search algorithm is used to plan the optimal path from the force node to the target node, and to adjust it in real time in the dynamic battlefield environment;

[0062] The entities in the combat system refer to the key elements that play specific functions in the combat process, including reconnaissance equipment, command center, force unit, etc. According to the function of the entity in the combat, the nodes are divided into the following categories:

[0063] Reconnaissance nodes are responsible for collecting battlefield intelligence, such as unmanned aerial vehicles, radar equipment;

[0064] Command nodes are responsible for analyzing intelligence and making combat decisions, such as command centers;

[0065] Force nodes are responsible for executing combat tasks, such as combat units and firepower units;

[0066] Each node is assigned dynamic attributes to accurately describe its function and state in the battlefield situation, including:

[0067] Position coordinates are used to represent the geographic location of the node in two-dimensional or three-dimensional coordinate form, which is used for path planning and battlefield distribution calculation;

[0068] Task status is used to indicate whether the node is executing a task, the task completion progress and the current load status;

[0069] Force status includes the total amount of resources, the remaining amount of resources, the amount of resource allocation, and whether it is overloaded;

[0070] Response delay is used to represent the response time of the node to the instruction, which depends on the communication delay, task complexity and resource availability;

[0071] Task priority is dynamically calculated by threat level, resource demand and time urgency;

[0072] Please refer to Figure 2, the embodiment of the present application is a force deployment model network mode schematic diagram, the investigation part is responsible for collecting battlefield information and reporting enemy situation, such as enemy position, activity range and target characteristics. It is the starting point of information flow, providing basic information for the whole combat process, the command part is responsible for receiving the information transmitted by the investigation node, analyzing information and generating decision, sending instructions to the combat node for commanding combat, it is the core node of information analysis and decision. The combat part receives the combat instructions of the command node, implements the corresponding combat task, including firepower attack, support task, etc., the target is the target of combat action, such as enemy facilities, important strongholds, etc. The combat node directly acts on the target node through resource allocation, the information flow transmits intelligence data from the investigation node to the command node, such as target position and enemy activity information, which is used to support decision, the instruction flow is sent from the command node to the combat node, such as specific attack target or combat plan, the resource flow is put into the target node by the combat node, such as firepower support, material supply, etc., to complete the target task.

[0073] S1. The specific steps are as follows:

[0074] S1.1: the entities in the combat system are divided into investigation nodes, command nodes and force nodes according to functions, ensuring that each type of node has clear tasks and attributes in the model, and the attribute values are added to each node according to the functions of the entities, wherein the attribute values include position coordinates, task state, force state, corresponding time delay and task priority;

[0075] Through node classification and attribute assignment, the model can reflect the dynamic state of each node in the combat system in real time, providing a basis for subsequent task allocation and path planning.

[0076] S1.2: establish interaction relationship according to node type, wherein the interaction relationship includes information flow, instruction flow and resource flow;

[0077] The information flow is initiated by the investigation node, which transmits real-time intelligence to the command node, the instruction flow is initiated by the command node, which sends combat instructions to the force node, the resource flow is initiated by the force node, which provides firepower support, material supply and other resources to the target node, through the above interaction relationship modeling, the cooperation between nodes can be accurately described, and a multi-level and multi-functional combat network is constructed;

[0078] S1.3: add dynamic attributes to nodes and interaction relationship, and build a force deployment model in the form of directed graph;

[0079] The abstract nodes and interaction relations are combined to further build a force deployment model. The vertices represent reconnaissance nodes, command nodes and force nodes. The edges represent information flow, instruction flow and resource flow between nodes. The vertices and edges are attached with dynamic attributes to reflect the real-time changes of the battlefield situation. The dynamic attributes of the vertices include the task state, resource state and position change of the nodes. The dynamic attributes of the edges include the path accessibility, resource flow volume and communication time delay. According to the feedback of the reconnaissance nodes and the task execution situation, the directed graph structure is dynamically updated. For example, when the position of a target node or the threat level changes, the model will adjust the path weight and resource allocation scheme in real time. If a path is blocked or communication fails, the edge weight will be recalculated to find an alternative path.

[0080] The additional dynamic attributes include:

[0081] A blank task sequence is added. The blank task sequence includes a task time window and a task timing dependency. The task time window is used to attach time constraints to tasks and nodes, including the earliest start time, the latest completion time and the task duration. The task timing dependency is used to establish a priority order diagram between tasks according to the dependency relationship of multiple tasks.

[0082] The blank task sequence is a preset task container used to provide a framework of time constraints and dependency relationships for node and resource allocation before actual task generation. The task time window includes three key time parameters.

[0083] The earliest start time refers to the earliest time point at which a task can start, which is determined by the completion time of the upstream dependent task.

[0084] The latest completion time refers to the last time point at which a task must be completed, which is determined by the task urgency or strategic demand.

[0085] The task duration refers to the estimated time required to execute the task, which is dynamically adjusted based on the task nature and node capability.

[0086] A task priority order diagram is generated based on the dependency relationship of the tasks. Each task node in the diagram is connected to its dependent task node. A topological sorting algorithm is used to sort the task nodes to ensure that the pre-depending dependencies of the task allocation are met. The task priority order diagram can be dynamically updated according to the task priority.

[0087] A task conflict detection mechanism is established. When a reconnaissance node or a force node is called by multiple tasks simultaneously, the resource usage order is adjusted through the task priority and the task time window.

[0088] The reconnaissance node collects terrain change information, including obstacle generation, path blocking and passable width change. According to the battlefield terrain, a dynamic path feasibility diagram between nodes is established.

[0089] The specific steps of S2 are as follows:

[0090] S2.1: obtaining initial attributes of the task from the target node by the reconnaissance node, wherein the initial attributes include a target position representing geographical coordinates of the target node for path planning and force scheduling, a threat level determined based on indexes such as enemy combat power and firepower strength for task priority calculation, and a task type representing the nature of the task such as reconnaissance, attack, and defense to determine the types and quantities of resources required by the task;

[0091] S2.2: calculating a task priority according to the initial attributes, sorting in a task demand sequence according to the task priority, and updating the task in the task demand sequence according to descending order, and the calculation formula of the task priority is:

[0092]

[0093] wherein P represents the task priority, W represents the threat level of the target node, ω1 represents the weight of the threat level, T s represents the earliest start time of the task, T e represents the latest completion time of the task, T c represents the current time, ω2 represents the weight of the time constraint, D represents the average distance between the force node and the target node, F represents the path accessibility average between the force node and the target node, ω3 represents the weight of the path constraint, R d represents the difference between the resource demand of the target node and the current remaining resources, and ω4 represents the weight of the resource matching degree.

[0094] By comprehensively considering the threat level, time constraint, path cost, and resource matching, the task priority is dynamically adjusted to ensure the scientificity and flexibility of battlefield resource allocation, to preferentially complete tasks with high threat, high urgency, and low cost, and to optimize the task execution order;

[0095] The optimization objectives of the force allocation optimization model include task priority satisfaction, resource utilization, and force node load balancing;

[0096] The task priority satisfaction reflects the completion degree of high-priority tasks and is a key factor in the optimization objectives. High-threat and high-urgency tasks are preferentially completed, while the path cost and resource matching degree are also considered. In practical applications, the task priority satisfaction is achieved through dynamic resource allocation. Tasks with high priority obtain more allocation proportion under limited resources. The real-time updating mechanism of the task priority ensures that high-priority tasks can quickly obtain resource support when the battlefield situation changes;

[0097] Resource utilization measures the efficiency of resource utilization of force nodes, the goal is to reduce resource idling and waste as much as possible through reasonable allocation, in the model, the resource allocation ratio of force nodes is dynamically adjusted according to the demand of the task, and the resource allocation of each force node is as close to its available capacity as possible, the optimization model dynamically monitors the resource usage state of the force node, and allocates more resource tasks to the low-load node;

[0098] Load balancing aims to avoid the low efficiency caused by resource overload or uneven resource allocation of force nodes. Through dynamic regulation, the task load is evenly distributed to multiple nodes to ensure the cooperative combat capability among nodes. In the process of load balancing, the task demand, resource matching degree and path cost are considered, and the node with lower load is preferentially selected for distribution to avoid the influence of single node overload on the overall combat efficiency and improve the cooperative ability among nodes to ensure the smooth execution of combat tasks;

[0099] The constraint conditions of the force allocation optimization model include:

[0100] The resource demand of each task is less than or equal to the total resource capacity of the force node, and in the resource allocation process, it is ensured that the resource demand of each task will not exceed the total resource capacity of the force node, and the remaining resource amount of the node is dynamically adjusted, and the node with sufficient resources is preferentially selected to avoid task allocation failure due to insufficient resources and improve the task completion rate;

[0101] The total amount of resource allocation of each force node is less than or equal to its available capacity, the remaining available resources of the node are updated in real time, and each allocation in the allocation scheme needs to meet the resource capacity constraint. In the case of task allocation conflict, the resources are redistributed through priority adjustment and time window optimization to avoid resource overflow, improve the rationality and feasibility of the allocation scheme, and ensure the smooth execution of the task;

[0102] The resource allocation of each task is completed within its task time window, and the time window constraint ensures that the resource allocation and execution of each task are completed within the available time range of the task. In actual implementation, the remaining time window of the task is calculated to dynamically adjust the allocation priority, and the task with time urgency is ensured to be completed first. The optimization model will sort the tasks according to the remaining time window and dynamically optimize the allocation order in combination with the resource allocation situation to ensure the timely completion of time-sensitive tasks and avoid task failure caused by delay;

[0103] The resource transfer path between the force node and the target node meets the path feasibility requirement. In the path planning, the path weight is dynamically updated in combination with the force deployment model, and the path with high trafficability and low cost is preferentially selected for distribution to improve the efficiency and safety of resource transfer and reduce the risk in the transfer process;

[0104] Please refer to Figure 3The embodiment of the present application is a force allocation flowchart, which generates a force allocation scheme through a force allocation optimization model, including:

[0105] A task requirement sequence and a force deployment model are inputted to generate a task-node state diagram, wherein the vertices of the task-node state diagram represent task nodes and force nodes, the edges represent the resource allocation possibility between the task nodes and the force nodes, and the edge weights are determined by the task priority, path distance, passing probability and resource matching degree;

[0106] The construction of the state diagram is based on the dynamic interaction relationship between the task nodes and the force nodes, and aims to represent the task allocation possibility and the comprehensive evaluation factors. The task nodes represent specific task requirements, such as resource requirements, time windows and task priorities, while the force nodes represent available resources for executing tasks. The weights of the edges comprehensively consider the matching degree between the task nodes and the force nodes;

[0107] A reinforcement learning agent is created, and the state space, action space and reward function of the reinforcement learning agent are initialized. The task-node state diagram is taken as the input parameter of the reinforcement learning agent, and the force allocation scheme is outputted;

[0108] The state space takes the task-node state diagram as the input, and the state space includes the dynamic attributes of the task nodes (such as task priority, remaining time window), the resource states of the force nodes (such as remaining resource amount, load) and the path passing ability between the task nodes. The state space is automatically updated after each task allocation, ensuring that the agent perceives the latest battlefield situation;

[0109] The action is defined as the resource allocation behavior between the task nodes and the force nodes, and the agent evaluates the feasibility and benefits of each action according to the state space. The action space is dynamically adjusted to exclude unfeasible resource allocation schemes, such as resource shortage or path impassability allocation actions;

[0110] The reward function guides the agent to select the optimal allocation strategy, and comprehensively considers the task priority satisfaction degree, resource utilization rate and load balancing. The high and low of the reward value depends on the completion of the task, the resource allocation efficiency and the balance degree of the global load;

[0111] Specifically, in the process of generating the force allocation scheme by the force allocation optimization model, first, the task demand sequence and the force deployment model are input into the system to construct a task-node state graph, in which the task nodes and the force nodes are taken as vertices, and the edges represent the possibility of resource allocation between the task nodes and the force nodes, and the weights of the edges are calculated by the dynamic attributes such as task priority, path distance, passage probability and resource matching degree. The setting of the task priority is dynamically adjusted based on the threat level and time urgency of the task, and the path distance and passage probability reflect the geographical relationship between the force nodes and the target nodes and the dynamic changes of the battlefield environment, and the resource matching degree is used to evaluate whether the current force node has the resource conditions to execute the task. On the basis of constructing the task-node state graph, the reinforcement learning agent is initialized and receives the parameters of the state space, the action space and the reward function. The state space includes the priority, time window and resource demand of the task node, and the current resource state, availability and path passage of the force node, the action space defines the resource allocation decision options between the task and the node, and the reward function measures the pros and cons of the allocation scheme through multi-objective optimization. Based on the input task-node state graph, the agent generates a resource allocation scheme through reinforcement learning strategy iteration, and dynamically updates the task priority and the node state to meet the constraints of the task time window and the resource demand. Through the reinforcement learning technology, the force allocation model can realize efficient allocation in the multi-task and dynamic battlefield environment, and optimize and adjust the allocation scheme according to the real-time feedback, so as to ensure the rapid response of high-priority tasks, improve the resource utilization and node load balancing, and finally generate a force allocation scheme that meets the global optimization goal.

[0112] The calculation formula of the reward function is:

[0113]

[0114] wherein Z represents the reward value, i represents a single task in the task demand sequence, N represents the total number of tasks in the task demand sequence, M represents the total number of force nodes allocated to the task i, j represents a single force node allocated to the task i, represents the weight of the priority satisfaction degree, represents the weight of the resource utilization rate, represents the weight of the load balancing, P i represents the task priority of the task i, η represents the gain coefficient of the resource allocation to the task priority, R i,j represents the resource demand satisfaction degree of the force node j to the task i, u j represents the allocated resource amount of the force node j, U j represents the total resource capacity of the force node j, λ j represents the load standard deviation of the force node j, U represents the average value of the allocated resources of all nodes in the force allocation scheme;

[0115] The reinforcement learning agent comprises:

[0116] According to the state space, the allocation probability between the task nodes and the force nodes is calculated, and the initial scheme is generated by recombining the allocation probability through a random strategy;

[0117] In this step, the reinforcement learning agent first dynamically constructs the mapping relationship between the task nodes and the force nodes based on the current state space, comprehensively considers the demand attributes of the task nodes (such as task priority, time window, resource demand, etc.) and the state attributes of the force nodes (such as resource capacity, geographical location, load condition, etc.), and constructs the mapping relationship between the task nodes and the force nodes. This mapping relationship is realized by calculating the allocation probability matrix, where each probability value represents the matching possibility between a certain task node and a certain force node. In order to ensure the exploration and diversity of the allocation scheme, the agent uses a random strategy to rearrange and sample the allocation probability matrix, and balances the exploration by introducing a greedy strategy, thereby generating an initial force allocation scheme.

[0118] Specifically, the calculation of the allocation probability needs to comprehensively evaluate the path distance, passing probability and resource matching degree between the task and the node, and dynamically adjust the allocation probability to adapt to the changes of the battlefield environment. For example, when the load of a certain force node is too high, its allocation probability will automatically decrease, thereby avoiding the concentration of tasks to a few nodes and causing the system performance to decline. The introduction of the random strategy further enhances the exploration ability of the agent to new allocation possibilities, preventing the occurrence of local optimal solutions.

[0119] The reasonable mapping between the tasks and the force nodes is realized, while the flexibility and adaptability of the allocation scheme are preserved, which can dynamically adjust the allocation strategy in complex battlefield environments and meet the scheduling needs of multiple tasks and multiple nodes. The principle is that the dynamic relationship between the tasks and the force nodes is captured through state space modeling, and the feasibility evaluation and random adjustment are performed through the allocation probability matrix, so that the agent can consider the global optimization and local adjustment needs when generating the allocation scheme.

[0120] In the action space, the initial scheme is subjected to constraint condition inspection to ensure that the initial scheme meets the realistic constraints of task execution and resource allocation, so as to avoid task failure or force node overload due to unreasonable allocation. It is judged whether the constraint condition is met. If not, the initial scheme is regenerated in the state space, and the reward value of this scheme is set to 0. If it is met, the reward value of the initial scheme is calculated according to the reward function;

[0121] If the initial scheme fails to meet any constraint condition, the agent will set the reward value of the scheme to 0, and back off to the state space to generate a scheme while retaining the known local optimization information in the current allocation to guide the next round of scheme generation. Through this dynamic adjustment mechanism, the agent can gradually filter out infeasible allocation actions, optimizing the global feasibility and execution efficiency of the scheme;

[0122] When the initial scheme meets all the constraint conditions, the agent will calculate the reward value of the scheme according to the set reward function. The reward function combines task priority, resource utilization, load balancing, and path cost, etc. to evaluate the quality of the current allocation scheme. The completion of high-priority tasks, efficient use of resources, reduction of path cost, and improvement of load balancing will significantly improve the reward value;

[0123] Specifically, through constraint condition testing and reward function evaluation, the agent can efficiently select feasible schemes that meet the actual needs of the battlefield, and gradually optimize the force allocation strategy, ultimately achieving coordinated scheduling of multiple tasks and rational use of force nodes. The combination of action space and constraint conditions realizes the preliminary evaluation and screening of the allocation scheme, while the design of the reward function guides the agent to approach the global optimization goal. This mechanism ensures the effectiveness of the training direction and strategy optimization in the reinforcement learning process;

[0124] Determine whether the reward value meets the set threshold for three consecutive rounds. If it meets, output the initial scheme corresponding to the maximum reward value in the three rounds as the force allocation scheme. If it does not meet, retain the initial scheme in the state space and optimize the loss function of the random strategy according to the reward value to generate a new initial scheme;

[0125] Specifically, after each round of scheme generation and evaluation, the agent will record the reward value of the current initial scheme and compare it with the reward values of the previous two rounds to determine whether the reward values of the three consecutive rounds are less than the set threshold. The set threshold of the reward value is a dynamic adjustment parameter used to measure the overall quality of the allocation scheme.

[0126] The new initial scheme is more inclined to the schemes that perform better in the historical data in terms of allocation probability, while retaining the exploration ability for new allocation possibilities through randomness, so as to be more likely to generate high-quality allocation schemes in the next iteration. The three-round judgment of the reward value not only measures the overall trend of the allocation scheme, but also ensures the practicality of the output scheme through the local optimal scheme selection mechanism. Through the dynamic adjustment of the random strategy, the reinforcement learning agent gradually approaches the global optimal solution. The core of this mechanism lies in balancing exploration and utilization, which not only utilizes existing high-quality allocation strategies, but also retains flexibility and innovation space for scheme optimization through randomness, thereby continuously improving the adaptability and execution efficiency of force scheduling in dynamic battlefield environments.

[0127] The S4 specific steps are as follows:

[0128] S4.1: initialize the path graph, set the force node in the force allocation scheme as the starting point and the target node as the end point, and mark them in the path graph;

[0129] S4.2: set the feasible path between the force node and the target node as the edge of the path graph, and the weight of each edge represents the comprehensive cost of the path, wherein the comprehensive cost includes path distance and path resource demand;

[0130] S4.3: select the edge with the minimum cumulative weight as the optimal path of the force node to the target node, traverse each force node in the force allocation scheme, and output the optimal path scheme.

[0131] Although the embodiments of the present application have been shown and described above, it should be understood that the above embodiments are exemplary and should not be construed as limiting the present application, and those skilled in the art can make changes, modifications, replacements and variations to the above embodiments within the scope of the present application.

Claims

1. A method for rapid reorganization of combat forces based on an architecture model, characterized in that, The method comprises: Abstracting entities in a combat system as nodes, and constructing a force deployment model based on interaction relationships between the entities; According to the random enemy situation detected by the nodes, updating the target nodes in the force deployment model, and generating a task demand sequence according to the initial attributes of the target nodes; According to the task demand sequence and the force deployment model, establishing a force allocation optimization model, and generating a force allocation scheme through the force allocation optimization model according to resource limitations and task priorities; According to the force allocation scheme, planning an optimal path for the force from the force nodes to the target nodes; The force deployment model is constructed, comprising: Dividing the entities in the combat system into reconnaissance nodes, command nodes and force nodes according to functions, and attaching attribute values to each node according to the functions of the entities, wherein the attribute values include position coordinates, task states, force states, corresponding time delays and task priorities; Establishing interaction relationships according to node types, wherein the interaction relationships include information flow, instruction flow and resource flow; Attaching dynamic attributes to the nodes and the interaction relationships to construct the force deployment model in the form of a directed graph; The dynamic attributes include: Adding a blank task sequence, wherein the blank task sequence includes a task time window and a task timing dependency, the task time window is used to attach time constraints to tasks and nodes, including the earliest start time, the latest completion time and the task duration, and the task timing dependency is used to establish a priority order diagram between tasks according to the dependency relationship of multiple tasks; Establishing a task conflict detection mechanism, when a reconnaissance node or a force node is called by multiple tasks at the same time, adjusting the resource use order through task priority and task time window; Collecting terrain change information through the reconnaissance node, including obstacle generation, path blocking and passable width change, and establishing a dynamic path feasibility diagram between nodes according to the battlefield terrain; The task demand sequence is generated, comprising: Obtaining the initial attributes of the tasks from the target nodes through the reconnaissance nodes, wherein the initial attributes include target positions, threat levels and task types; Calculating the task priority according to the initial attributes, sorting the tasks in the task demand sequence according to the task priority, and updating the tasks in the task demand sequence according to the descending order, and the calculation formula of the task priority is: ， wherein, represents a task priority, represents a threat level of a target node, represents a weight of a threat level, represents an earliest start time of a task, represents a latest finish time of a task, represents a current time, represents a weight of a time constraint, D represents an average distance between a force node and a target node, represents a path accessibility mean between a force node and a target node, represents a weight of a path constraint, represents a difference between a target node resource requirement and a current remaining resource, represents a weight of a resource matching degree.

2. The method of claim 1, wherein the method further comprises: The optimization objectives of the force allocation optimization model include task priority satisfaction, resource utilization and force node load balancing.

3. The method of claim 1, wherein the method further comprises: The constraint conditions of the force allocation optimization model include: The resource demand of each task is less than or equal to the total resource capacity of the force node; The total amount of resource allocation of each force node is less than or equal to its available capacity; The resource allocation of each task is completed within its task time window; The resource transfer path between the force node and the target node meets the path feasibility requirement.

4. The method of claim 3, wherein the method further comprises: The force allocation scheme is generated through the force allocation optimization model, comprising: Input a task demand sequence and a force deployment model to generate a task-node state graph, wherein vertices of the task-node state graph represent task nodes and force nodes, edges represent resource allocation possibilities between the task nodes and the force nodes, and edge weights are determined by task priorities, path distances, passage probabilities, and resource matching degrees; create a reinforcement learning agent, initialize a state space, an action space, and a reward function of the reinforcement learning agent, input the task-node state graph as an input parameter of the reinforcement learning agent, and output a force allocation scheme.

5. The method of claim 4, wherein, The calculation formula of the reward function is: ， Wherein, Z represents the reward value, i represents a single task in the task demand sequence, N represents the total number of tasks in the task demand sequence, M represents the total number of force nodes allocated to the task i, j represents a single force node allocated to the task i, represents the weight of the priority satisfaction degree, represents the weight of the resource utilization rate, represents the weight of the load balancing, represents the task priority of the task i, represents the gain coefficient of the resource allocation to the task priority, represents the resource demand satisfaction degree of the force node j to the task i, represents the allocated resource amount of the force node j, represents the total resource capacity of the force node j, represents the load standard deviation of the force node j, and U represents the average value of the allocated resources of all nodes in the force allocation scheme.

6. The method of claim 5, wherein the method further comprises: The reinforcement learning agent comprises: According to the state space, calculate the allocation probability between the task nodes and the force nodes, recombine the allocation probability through a random strategy, and generate an initial scheme; perform constraint condition inspection on the initial scheme in the action space, determine whether the constraint condition is met, if the constraint condition is not met, re-generate the initial scheme in the state space, and set the reward value of the scheme to 0, if the constraint condition is met, calculate the reward value of the initial scheme according to the reward function; determine whether the reward value meets a set threshold value for three consecutive rounds, if the reward value meets the set threshold value, output the initial scheme corresponding to the maximum reward value in the three rounds as the force allocation scheme, if the reward value does not meet the set threshold value, retain the initial scheme in the state space and optimize a loss function of the random strategy according to the reward value, and re-generate the initial scheme.

7. The method of claim 1, wherein the method further comprises: The optimal path comprises: Initialize a path graph, mark the force nodes in the path graph as starting points and the target nodes as ending points in the force allocation scheme; set feasible paths between the force nodes and the target nodes as edges of the path graph, and the weight of each edge represents the comprehensive cost of the path, wherein the comprehensive cost comprises a path distance and a path resource demand; select an edge with the minimum cumulative weight as the optimal path of the force nodes to the target nodes, traverse each force node in the force allocation scheme, and output an optimal path scheme.

Images