An Adaptive Cross-Domain Kill Chain Decision-Making Method Based on Deep Learning

Abstract

The present invention belongs to the field of computer technology, and specifically relates to an adaptive cross-domain kill network decision-making method based on deep learning, aiming to quickly provide a combat resource selection plan for battlefield commanders, including the following steps: 1) Based on the adaptive cross-domain kill network, configure resources according to the expected effect of the decision-maker's task, and establish a cross-domain combat resource selection model; 2) Take minimizing the number of combat resources as the optimization goal, and formalize the problem of minimizing combat resources into an integer programming problem model; 3) Solve the integer programming problem model to obtain the training set of deep learning; 5) Use the obtained training set to train the deep neural network; 4) Optimize the model parameters of the deep neural network so that the output result of the deep neural network model reaches the required accuracy. The invention solves the problem of difficult to make a quick decision considering a large number of combat resources in the adaptive cross-domain kill network, and reduces the task completion delay.

Description

Technical Field

[0001] The present invention belongs to the field of computer technology, and particularly relates to an adaptive cross-domain kill network decision-making method based on deep learning. Background Art

[0002] Most of the current combat mission planning involves cross-domain coordination. Decision-makers need to spend a lot of energy coordinating the resources of various services across domains. There is almost no development ability for combat mission selection and no comparison of combat missions. As a result, combat resources are statically allocated to specific resources. However, when the situation changes, it is difficult for decision-makers to determine the future trend. Therefore, some statically allocated resources may not be fully utilized, while some resources are overburdened and difficult to complete tasks. Therefore, it is necessary to build an adaptive cross-domain kill network. However, considering the task completion conditions and selecting suitable combat resources in multiple combat domains requires a large amount of computing time and is difficult to meet the rapidly changing decision-making needs. Summary of the Invention

[0003] In view of this, the present invention proposes an adaptive cross-domain kill network decision-making method based on deep learning. The kill network is based on a "bid invitation - bid submission - bid competition" architecture, and the kill network is classified according to roles to provide a service-oriented decision-making method. The present invention develops a real-time decision-making method to allocate combat resources to kill chains to meet specific task requirements, and can dynamically reallocate combat resources according to changes in the situation to assist commanders in quickly building a network.

[0004] In order to achieve the above technical objectives, the specific technical solutions adopted by the present invention are as follows:

[0005] An adaptive cross-domain kill network decision-making method based on deep learning, comprising the following steps:

[0006] 1) Based on the adaptive cross-domain kill network, configure resources according to the task expected effect of the decision-maker, establish a cross-domain combat resource selection model, and formalize the resource configuration in mathematical form;

[0007] 2) Take minimizing the number of combat resources as the optimization goal, and formalize the problem of minimizing combat resources into an integer programming problem model;

[0008] 3) Solve the integer programming problem model to obtain the training set of deep learning;

[0009] 4) Use the obtained training set to train a deep neural network;

[0010] 5) Optimize the model parameters of the deep neural network so that the output result of the deep neural network model reaches the required accuracy;

[0011] Wherein: The resources in the kill network are classified according to their roles in reality.

[0012] Furthermore, the specific method for constructing an adaptive cross-domain kill network includes the following steps:

[0013] S101: The adaptive cross-domain kill network consists of a decision maker, liaison software deployed on edge servers, and equipment resources in each domain; the adaptive cross-domain kill network is based on a bid-issuing - bid-submitting - bid-winning architecture. In this architecture, the role of the decision maker is equivalent to that of a consumer, the liaison software is equivalent to a battlefield liaison officer, and the liaison software is a representative of the equipment resources within a domain; the equipment resources are equivalent to suppliers, and the suppliers in each domain form a capabilities market. When the decision maker determines a target and the expected effects that need to be achieved for the task, the liaison software in each domain submits bids based on the availability of weapons and equipment within its domain, and the liaison software formulates the equipment selection for completing the task according to the bidding situation;

[0014] S102: Based on the roles of suppliers, consumers, liaison officers, and the capabilities market, and with the implementation effect and time delay as constraints, select appropriate suppliers from the capabilities market to construct a kill chain.

[0015] Furthermore, the method for formalizing the problem of minimizing combat resources into an integer programming problem model in step 2) specifically includes the following steps

[0016] S201: Represent the set of liaison officers as: ;

[0017] S202: Divide out capabilities markets according to the number of liaison officers. The set of capabilities markets is represented as ; Each capabilities market consists of the connections between suppliers, represented as a tuple , where represents the set of suppliers in domain , represents the set of connections between suppliers in domain ;

[0018] S203: Set to indicate whether suppliers are connected. Its set is , that is , represents the set of , where

[0019]

[0020] S204: According to the requirements of the consumer, set the total time cost as , and set the sum of the time costs of each equipment resource to be less than the task time requirement;

[0021] S205: Set the requirement for the attack range as R, and each equipment resource needs to meet the attack range ;

[0022] Among them: The mathematical model of the kill network is:

[0023]

[0024]

[0025]

[0026]

[0027] Among them, The nodes in the kill network need to meet the combat range requirements.

[0028] Furthermore, the method for optimizing the deep learning network model is specifically as follows:

[0029] Adopt the greedy algorithm for each layer of the learning model, and endow the initial weights to the training model through unsupervised training;

[0030] Solve the problem through traditional optimization algorithms, obtain the training set, and perform supervised training and parameter tuning on the deep learning model.

[0031] Furthermore, in the above 5), the preset value of the required accuracy is 98%.

[0032] Furthermore, the traditional optimization algorithm is the branch and bound algorithm.

[0033] Compared with the prior art, the present invention has the following advantages:

[0034] The present invention plans and deploys the adaptive cross-domain kill network based on the "issuing bids - submitting bids - competing for bids" architecture, classifies the kill network according to roles, and provides a decision-making method for services. The battlefield commander does not need to understand the available combat resources and services in each domain. The method provided by the invention can assist the commander in selecting suitable kill network elements from the shared resource pool to form a kill chain, and can reallocate combat resources in real time according to the changes in the battlefield to quickly establish a kill chain.

[0035] Traditional battlefield decision-making methods mainly utilize traditional mathematical optimization methods. However, the mathematical problem after formalizing battlefield decision-making is an integer programming problem, which is considered an NP-hard problem and difficult to solve within polynomial time, making it difficult to meet the rapidly changing combat requirements on the battlefield. The present invention uses deep learning methods to transfer the computing time to the model training stage, greatly reducing the computing time of the model, and can solve the above problems to meet the requirements of time-delay sensitive combat tasks. It enables the constructed kill network to quickly coordinate and schedule to achieve reasonable allocation of resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0037] Figure 1 Schematic diagram of the adaptive cross-domain kill network architecture adopted by an adaptive cross-domain kill network decision-making method based on deep learning in a specific embodiment of the present invention;

[0038] Figure 2 Schematic diagram of the process of an adaptive cross-domain kill network decision-making method based on deep learning in a specific embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0039] The embodiments of the present disclosure will be described in detail below with reference to the drawings.

[0040] The following specific examples illustrate the embodiments of the present disclosure. Those skilled in the art can easily understand other advantages and effects of the present disclosure from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, not all of them. The present disclosure can also be implemented or applied through other different specific embodiments, and various modifications or changes can be made to the details in this specification based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that, without conflict, the following embodiments and the features in the embodiments can be combined with each other. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments in the present disclosure belong to the scope of protection of the present disclosure.

[0041] It should be noted that the following description relates to various aspects of embodiments within the scope of the appended claims. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on this disclosure, those skilled in the art should understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of the aspects set forth herein can be used to implement an apparatus and / or practice a method. Additionally, this apparatus can be implemented and this method can be practiced using other structures and / or functionality in addition to one or more of the aspects set forth herein.

[0042] It should also be noted that the diagrams provided in the following embodiments only schematically illustrate the basic concept of this disclosure. The diagrams only show the components related to this disclosure and are not drawn according to the number, shape, and size of the components in actual implementation. The type, quantity, and ratio of each component in its actual implementation can be arbitrarily changed, and the component layout type may also be more complex.

[0043] In addition, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the aspects described can be practiced without these specific details.

[0044] In an embodiment of the present invention, an adaptive cross-domain kill web decision-making method based on deep learning is proposed, including the following steps:

[0045] 1) Based on the adaptive cross-domain kill web, perform resource allocation according to the task expected effect of the decision maker, establish a cross-domain combat resource selection model, and formalize the resource allocation in a mathematical form.

[0046] 2) With minimizing the number of combat resources as the optimization goal, formalize the problem of minimizing combat resources into an integer programming problem model.

[0047] 3) Solve the integer programming problem model to obtain the training set for deep learning.

[0048] 4) Use the obtained training set to train a deep neural network.

[0049] 5) Optimize the model parameters of the deep neural network so that the output result of the deep neural network model reaches the required accuracy.

[0050] Wherein: The resources in the kill web are classified according to their roles in reality.

[0051] In this embodiment, the specific method for constructing an adaptive cross-domain kill web includes the following steps:

[0052] S101: The adaptive cross - kill network consists of a decision - maker, liaison software deployed on edge servers, and equipment resources in each domain; the adaptive cross - kill network is based on a tender - bid - competition architecture. In this architecture, the role of the decision - maker is equivalent to that of a consumer, the liaison software is equivalent to a battlefield liaison, and the liaison software is the representative of the equipment resources within a domain; the equipment resources are equivalent to suppliers, and the suppliers in each domain form a capabilities market. When the decision - maker determines the target and the expected effects that the task needs to achieve, the liaison software in each domain bids according to the availability of weapons and equipment within its domain, and the liaison software formulates the equipment selection for completing the task based on the bidding situation.

[0053] S102: Based on the roles of suppliers, consumers, liaisons, and the capabilities market, and with the implementation effect and time delay as constraints, select appropriate suppliers from the capabilities market to construct a kill chain.

[0054] In this embodiment, the method of formalizing the problem of minimizing combat resources into an integer programming problem model in step 2) specifically includes the following steps

[0055] S201: Represent the set of liaisons as: ;

[0056] S202: Divide out capabilities markets according to the number of liaisons. The set of capabilities markets is represented as ; Each capabilities market consists of the connections between suppliers, and is represented as a tuple , where represents the set of suppliers in domain , represents the set of connections between suppliers in domain ;

[0057] S203: Set to represent whether a supplier is connected, and its set is , that is , represents 's set, where

[0058]

[0059] S204: According to the needs of the consumer, set the total time cost as , and set the sum of the time costs of each equipment resource to be less than the task time requirement;

[0060] S205: Set the requirement for the attack range as R, and each equipment resource needs to meet the attack range ;

[0061] Among them: The mathematical model of the kill network is:

[0062]

[0063]

[0064]

[0065]

[0066] Among them, The nodes in the kill network need to meet the combat range requirements.

[0067] In this embodiment, the method for optimizing the deep learning network model is specifically as follows:

[0068] The greedy algorithm is adopted for each layer of the learning model, and the initial weight is assigned to the training model through unsupervised training;

[0069] The traditional optimization algorithm is used to solve the problem, obtain the training set, and perform supervised training and parameter tuning on the deep learning model.

[0070] In this embodiment, in the above 5), the preset value of the required accuracy is 98%.

[0071] In this embodiment, the traditional optimization algorithm is the branch and bound algorithm.

[0072] The flow of the adaptive cross-domain kill network decision-making method based on deep learning in this embodiment is as Figure 2 shown, and the adopted adaptive cross-kill network architecture is as Figure 1 shown.

[0073] The following combines Figure 1 to further introduce the above-mentioned adaptive cross-domain kill network architecture. The adaptive cross-kill network consists of a decision maker, liaison software deployed on the edge server, and equipment resources in each domain. The kill network is built based on the "issuing bids - submitting bids - competing for bids" architecture. In this architecture, the role of the decision maker is equivalent to that of a "consumer", the role of the liaison software acts as a "battlefield liaison officer", and it is a representative of the equipment resources within a domain. The equipment resources act as the role of "suppliers", and the "suppliers" in each domain form a "capability market". When the decision maker determines the target and the expected effect that the task needs to achieve, the liaison software in each domain conducts "bidding" according to the available situation of the weapons and equipment within the domain, and the liaison software formulates the equipment selection for task completion based on the "bidding" situation.

[0074] Based on the roles of "suppliers", "capability market" and "consumers", considering the implementation effect, time delay and other constraint conditions, select appropriate "suppliers" from the "capability market" to construct a kill chain;

[0075] Solve integer programming model problems to obtain deep learning datasets;

[0076] Use the generated dataset to train the deep learning network model and tune the model parameter values to ensure the accuracy meets the requirements.

[0077] The following combination Figure 2 The following further describes the process of online software upgrades for edge terminals:

[0078] Step 1: After the combat commander determines the combat target and assigns the combat mission, all the information required for a kill chain is formed based on the combat mission and the results of the mission implementation.

[0079] Step 2: The liaison officers in each domain determine whether available combat resources and services are available in their respective domains based on the expected mission results.

[0080] Step 3: Model the nodes in each domain. The “Liaison (L)” set is represented as: ; According to the number of "liaisons", they can be divided into The set of “capability markets” is represented as: ,Each capability market consists of a connection between “supplier” and “supplier”, representing the tuple ,in Representation Domain The "suppliers" collection in Representation Domain The set of connections between "suppliers" in the; Represents the "Suppliers" collection The elements of , express A collection of

[0081]

[0082] According to the needs of "consumers", the total time spent is , the time spent on each equipment resource should be less than the total time requirement; the attack range requirement is R, and each equipment resource needs to meet the attack range ;

[0083] Step 4: Build a model based on Step 3, mathematically formalizing it as a minimum equipment resource problem, that is, the minimum combat resources required to complete the mission, so that the kill chain can save combat resources and have strong resilience;

[0084] Step 5. The mathematical optimization problem established in step 4 is an integer programming problem, which is solved by the branch and bound algorithm;

[0085] Step 6: The data set obtained by solving in Step 5 can be used as the training set for the deep learning model.

[0086] Step 7: Train the deep neural network model with the data set obtained in Step 6, and transfer the solving time to the model training stage to reduce the model calculation time.

[0087] Step 8: Test according to the deep learning model trained in Step 7. If the training accuracy meets the required requirements, the model meets the requirements. If the model accuracy is lower than the required accuracy, continue to optimize the model parameters.

[0088] As described above, it is only the specific implementation manner of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed by the present disclosure should be covered by the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be subject to the protection scope of the claims.

Claims

1. An adaptive cross-domain kill web decision-making method based on deep learning, characterized in that, It includes the following steps: 1) Based on the adaptive cross-domain kill network, perform resource allocation according to the expected task effects of the decision maker, establish a cross-domain combat resource selection model, and formalize the resource allocation in a mathematical form; 2) With minimizing the number of combat resources as the optimization goal, formalize the problem of minimizing combat resources into an integer programming problem model; 3) Solve the integer programming problem model to obtain the training set for deep learning; 4) Use the obtained training set to train a deep neural network; 5) Optimize the model parameters of the deep neural network so that the output results of the deep neural network model reach the required accuracy. The resources in the kill network are classified according to their roles in reality. Among them, the specific method for constructing an adaptive cross-domain kill network includes the following steps: S101: The adaptive cross-domain kill network consists of a decision maker, liaison software deployed on an edge server, and equipment resources in each domain; the adaptive cross-domain kill network is based on a bid invitation-bid submission-bidding architecture. In this architecture, the role of the decision maker is equivalent to that of a consumer, the liaison software is equivalent to a battlefield liaison officer, and the liaison software is a representative of the equipment resources within a domain; the equipment resources are equivalent to suppliers, and the suppliers in each domain form a capabilities market; when the decision maker determines the target and the expected effects that the task needs to achieve, the liaison software in each domain submits bids according to the availability of weapons and equipment within its domain, and the liaison software formulates the equipment selection for completing the task based on the bidding situation; S102: Based on the roles of suppliers, consumers, liaison officers, and capabilities market, select appropriate suppliers from the capabilities market to construct a kill chain with the implementation effect and time delay as constraints; The method for formalizing the problem of minimizing combat resources into an integer programming problem model specifically includes the following steps: S201: Represent the liaison officer set as: L = {1, 2, …, l}; S202: Divide l into capability markets according to the number of liaison officers. The set of capability markets is represented as CM = {cm1, cm2, …, cm l}; Each capability market is composed of connections between suppliers and is represented as a tuple cm n = (s n , c n ), where s n represents the set of suppliers in the domain cm n , and c n represents the set of connections between suppliers in the domain cm n ; S203: Set To indicate whether the supplier is connected, and its set is S n , that is, S n={ , …, } , s n Indicates The set of, where S204: Set the total time spent as T according to the consumer's needs, set the sum of the time spent on each equipment resource to be less than the task time requirement, and the time spent on a single equipment resource to be t n ; S205: Set the requirement for the attack range as R, and each equipment resource needs to meet the attack range R， Among them, the mathematical model of the adaptive cross-domain kill network is as follows: Among them, the nodes in the kill net need to meet the requirements of the combat range.

2. The adaptive cross-domain kill web decision-making method based on deep learning according to claim 1, wherein The method for optimizing the deep neural network model is specifically as follows: Adopt a greedy algorithm for each layer of the learning model, and endow the training model with initial weights through unsupervised training; Solve the problem through a traditional optimization algorithm, obtain the training set, conduct supervised training on the deep learning model, and perform parameter optimization.

3. The adaptive cross-domain kill web decision-making method based on deep learning according to claim 2, characterized in that, In 5), the preset value of the required accuracy is 98%.

4. The adaptive cross-domain kill chain decision-making method based on deep learning according to claim 3, wherein The traditional optimization algorithm is the branch and bound algorithm.

Images