A dynamic network architecture maintenance system and method under adversarial conditions
By introducing attack prediction, decision update, and pre-deployment decision modules into a dynamic network architecture, and combining the A-CADer algorithm and multi-head attention mechanism, the problem of node prediction and maintenance of the network architecture under dynamic adversarial conditions is solved, and dynamic maintenance of network connectivity and system integrity is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- XIDIAN UNIV
- Filing Date
- 2024-12-03
- Publication Date
- 2026-05-26
AI Technical Summary
Existing technologies struggle to effectively predict and maintain critical nodes in a network under dynamic adversarial conditions, and are unable to predict adversary attacks and cover networks in rapidly changing network topologies, resulting in compromised network connectivity and system integrity.
A dynamic network architecture maintenance system is designed, including an attack prediction module, a decision update module, and a pre-deployment decision module. Through the A-CADer algorithm and multi-head attention mechanism, it predicts the opponent's possible attack targets in the next moment, and performs decision updates and redundancy node deployment according to changes in topology and attack strategy.
It enables effective maintenance of the network system in dynamic adversarial environments, maintaining network connectivity and system integrity, quickly responding to changes in the adversary's attack strategy, performing dynamic attack prediction and coverage, and supporting node importance calculation under both static and dynamic network topologies.
Smart Images
Figure CN119629071B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of dynamic network architecture maintenance technology, specifically relating to a dynamic network architecture maintenance system and method under adversarial conditions. Background Technology
[0002] In real-world scenarios where network system maintainers confront adversaries, effective coordination of various system units, such as command and control units, communication units, and strike units, is crucial. These units typically work in combination to execute tasks. However, the locations of these units in practical applications are often randomly distributed. To effectively coordinate these system units, communication units provide communication services to system units within their coverage area, connecting all units within the entire System of Systems (SoS) into a mesh topology. To ensure that communication between nodes in the SoS remains consistently effective and uninterrupted, redundant silent communication nodes always exist. These nodes are often deployed near vulnerable communication nodes to ensure that they can quickly take over and participate in the formation of the SoS if these communication nodes are attacked and forced to leave the SoS.
[0003] Assuming that an adversary's attack strategy can be observed at any given moment in reality, it is crucial to predict the sequence of key communication nodes in a SoS (Socially Targeted Systems) that may be attacked before the attack is carried out. Existing algorithms for calculating the importance of nodes in a mesh topology are difficult to apply to adversarial SoS that require rapid changes and constant connectivity.
[0004] The main reason why attack prediction lacks flexibility and timeliness in current research is that it only considers the static topology of the network. Commonly used methods for calculating the importance of nodes in a network include:
[0005] The disadvantages of both nearest neighbor-based and path-based sorting methods are as follows:
[0006] Network coverage methods primarily focus on rapid coverage and cannot predict adversary attacks on the network maintenance team's network system. They also fail to consider how to maintain the network system of the network maintenance team.
[0007] The CADer method primarily considers network coverage in dynamic networks, but it doesn't account for predicting nodes that could be targeted by an adversary attacking the network maintenance system. Therefore, it cannot perform system maintenance under dynamic adversarial conditions.
[0008] While this method takes into account the coverage of dynamic networks and can perform system maintenance under non-dynamic adversarial conditions, it does not consider the damage that adversary attacks can cause to the network maintaining party's network, and therefore cannot perform system maintenance under dynamic adversarial conditions.
[0009] The HCNT method models a real network by constructing a heterogeneous network for important node identification, but it does not consider coverage issues and therefore cannot support system integrity maintenance. It also cannot perform dynamic attack prediction.
[0010] This method can predict static attacks by identifying key nodes in a static network. However, it cannot predict dynamic attacks because it does not consider the damage caused by adversary attacks to the network maintainer's network system, nor the dynamic changes in the network topology. Furthermore, this method does not consider connectivity between nodes or network coverage issues, thus it cannot be used for system maintenance. Summary of the Invention
[0011] To overcome the shortcomings of the existing technology, the present invention aims to provide a dynamic network system maintenance system and method under adversarial conditions. The system and method design a decision-making architecture based on dynamic networks and changes in attack strategies in an adversarial environment. It can predict the nodes that the SoS may be attacked based on the changes in the topology of the system at time i-1 and time i, and make a decision in advance. The architecture will output the importance sequence of all communication nodes participating in the SoS in the next time step according to the opponent's attack strategy. The communication nodes in this sequence will likely become the targets of the attack in the next time step. Based on this sequence, the SoS can make pre-deployment decisions, so that the communication nodes in the sequence can make advance maneuvers or deploy redundant silent communication nodes around them.
[0012] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0013] A dynamic network architecture maintenance system under adversarial conditions includes an attack prediction module, a decision update module, and a pre-deployment decision module;
[0014] The attack prediction module is used to predict the sequence of important nodes that the opponent will attack in the next moment to maintain the network system.
[0015] The decision update module is used to update the opponent's attack strategy and feed the results back to the attack prediction module;
[0016] The pre-deployment decision module is used to make deployment decisions based on the prediction results of the network system maintainer.
[0017] The strike prediction module predicts the target nodes based on changes in the network architecture maintainer's topology and the opponent's strike strategy. The decision update module observes whether the opponent's strike strategy has changed. If it has changed, the updated strike strategy is fed back to the strike prediction module for re-prediction. If it has not changed, the pre-deployment decision module makes the decision, allowing the network architecture maintainer to take action in advance or deploy redundant silent communication nodes next to the important nodes to be attacked.
[0018] A method for using a dynamic network architecture maintenance system under adversarial conditions includes the following steps;
[0019] Step (1): Construct an attack prediction module, establish an optimization model based on the opponent's current attack strategy and the node attributes of the network system maintainer, take the node attributes in the network system maintainer's SoS as input, and use the A-CADer algorithm to solve the optimization model, and output the sequence of nodes with the highest probability of being attacked by the opponent in the SoS connected at the current time.
[0020] Step (2): The decision update module observes whether the opponent's attack strategy has changed. If it has changed, the updated attack strategy of the opponent is fed back to the attack prediction module for re-prediction. If it has not changed, the pre-deployment decision module completes the decision.
[0021] Step (3): Deploy the decision module in advance. Based on the prediction results of the network system maintainer, make maneuvers in advance, or deploy redundant silent communication nodes next to important nodes that will be attacked.
[0022] The strike prediction module completes the opponent strike node sequence P{P1,P2,…,P n The system predicts the enemy's current attack strategy, then updates the decision update module, and finally the pre-deployment decision module makes the final decision for the network system maintainer.
[0023] The specific steps (1) are as follows:
[0024] To predict the targets, we need to assume, in the initial state, that the opponent's attack strategy is to prioritize attacking communication nodes with high importance. Let the importance of each communication node in the SoS be k. Then, the scoring system indicators constituting k are: the degree d of the communication node, the service range r that the communication node can provide, the maximum number of system nodes n that the communication node can serve, and the attack frequency f of that node.
[0025] k = a1d + a2r + a3n + a4f,
[0026] Where d is the degree of the communication node, r is the range of services that the communication node can provide, n is the maximum number of system nodes that the communication node can serve, f is the frequency at which the communication node is attacked; a1 to a4 are weighted average coefficients.
[0027] The specific steps are as follows:
[0028] Step 1: Set the initial attack strategy as follows: prioritize attacking communication nodes with high importance;
[0029] Step 2: Define the importance of communication nodes in the SoS as follows:
[0030] k = a1d + a2r + a3n + a4f,
[0031] Where d is the degree of the communication node, r is the range of services that the communication node can provide, n is the maximum number of system nodes that the communication node can serve, and f is the frequency at which the node is attacked.
[0032] Step 3: Observe the topological changes of SoS at time i-1 and time i, and update the node attributes in SoS;
[0033] Step 4: Establish an attack prediction optimization model based on the attributes of communication nodes and system nodes within the SoS;
[0034] Step 5: Use the A-CADer algorithm to solve the optimization model, and finally output the sequence of nodes with the highest probability of being attacked by the opponent in the connected SoS at the current time.
[0035] Step 3 determines the characteristic information contained in the node attributes based on the node status at the current moment: the four indicators constituting the importance k of the communication node are the degree d of the communication node, the range r of the service that the communication node can provide, the maximum number of system nodes n that the communication node can serve, the frequency of the node being attacked f, and the location information of each node.
[0036] The specific steps are as follows:
[0037] Step 3.1: Update the degree index d of the nodes in the current network based on the network architecture maintainer;
[0038] Step 3.2: Update the index r based on the range of services that communication nodes in the current network can provide, according to the network architecture maintainer.
[0039] Step 3.3: Update the index n based on the maximum number of system nodes that a communication node in the current network can serve, according to the network architecture maintainer.
[0040] Step 3.4: Update the index f based on the attack frequency of nodes in the current network by the network system maintainer.
[0041] Step 4 specifically involves:
[0042] Step 4.1: Configure communication node C i The coverage area r, only within this range, system node S i Communication services are available;
[0043] Step 4.2: Configure communication node C i For system node S i Various communication transmission indicators;
[0044] Step 4.3: Configure communication node C i The maximum number of services and the principle of proximity service, i.e., a certain system node S i Exceeding a certain communication node C i If the maximum number of services is exceeded, or if the distance to the node is greater than that, then communication node C cannot be accepted. i Services;
[0045] Step 4.4: Set node x i For redundant communication node C i Communication nodes that participate in forming a SoS;
[0046] Step 4.5: Configure each communication node C i Service cost v i ;
[0047] Step 4.6: Using steps 4.1 to 4.5 and step 3 as constraints, establish the objective as follows: The optimization model.
[0048] Step 5 is described in detail below:
[0049] Step 5.1: Generate the embedding vector for each node during the encoding stage. This vector contains static structure (node attributes maintained by the network architecture) and dynamically processed feature information. The encoder consists of two encodings: static and dynamic.
[0050] Step 5.2: Employ a multi-head attention mechanism to extract feature information from the network architecture maintainer nodes and generate static embeddings. The feature information of the communication nodes includes location, coverage radius r, and cost information v. i And the importance k of the communication node, and the system node S i Includes location information;
[0051] Step 5.3: Define the vector For communication node C i Within the coverage area, the number of system nodes not covered by the communication node selected in step t;
[0052] Step 5.4: In the set of uncovered system nodes, there are key system nodes for the entire solution. The dynamically encoded information is projected onto the embedding vector through linear projection. superior;
[0053] Step 5.5: Combine static and dynamic coding information to construct the final embedding vector of the communication node;
[0054] Step 5.6: The decoder takes the embedding vector output in step 5.5 as input. The decoding stage consists of k steps. In each step, the agent selects an unselected communication node and adds the selected communication node to the output node sequence. Then, the dynamic embedding is updated according to the environmental state.
[0055] Step 5.7: Repeat step 5.6 iteratively until a sequence of output nodes P{P1,P2,…,P} that satisfies all constraints is constructed. n} or reach the maximum number of iterations.
[0056] Step (2) specifically involves:
[0057] As the strike strategy is updated, SoS can determine the weighting coefficients of each indicator in the importance k of communication nodes by analyzing the distribution of the actual strike nodes targeted by the adversary, thus changing the value of k. Before the pre-deployment decision module makes its final decision, the strike prediction module is executed again to obtain a new sequence P based on the updated adversary strike strategy. ′ {P1,P2,…,P n The details are as follows:
[0058] Step 1: If the attack strategy is updated to prioritize attacking the communication node with the highest number of service system nodes in the SoS at a certain moment, the importance is adjusted by increasing the coefficient a1.
[0059] Step 2: If the attack strategy is updated to prioritize attacking the communication node with the largest service range in the SoS of the network system maintainer at a certain time, the importance is adjusted by increasing the coefficient a2.
[0060] Step 3: If the attack strategy is updated to prioritize attacking the communication node with the strongest service capability in the SoS at a certain moment, the importance is adjusted by increasing the coefficient a3.
[0061] Step 4: If the attack strategy is updated to prioritize attacking the communication node that has been attacked the most times in the SoS, the importance is adjusted by increasing the coefficient a4.
[0062] Through the above steps, the decision update module can automatically adjust the model established by the strike prediction module and update the prediction sequence according to the changes in the strike strategy.
[0063] Step (3) specifically involves:
[0064] The prediction attack module outputs the sequence of nodes P{P1,P2,…,P} in the currently connected SoS with the highest probability of being attacked by the opponent. n After that, make decisions and deployments based on the confrontation situation at this moment;
[0065] Specifically as follows:
[0066] Step 1: Observe whether the opponent's attack strategy has been updated;
[0067] Step 2: If the attack strategy has been updated, execute the decision update module and output the new sequence P′{P1,P2,…,P n Otherwise, follow the existing sequence P{P1,P2,…,P}; n Make decisions;
[0068] Step 3: Based on this sequence and the actual adversarial situation, SoS can make pre-deployment decisions, enabling communication nodes in the sequence to maneuver in advance or deploy redundant silent communication nodes around them.
[0069] Through the above process, the decision-making architecture can predict the sequence of nodes that the adversary may attack at a certain time based on the adversary's attack strategy and the changes in the topology of the SoS network maintained by the network system, so as to make decisions in advance.
[0070] The beneficial effects of this invention are:
[0071] This invention presents a strike prediction and decision-making architecture based on dynamic networks and changes in adversary strike strategies. It can predict the sequence of key nodes targeted by the adversary in the next moment, while maintaining overall network connectivity, based on changes in the topology of the network system maintainer's SoS and adjustments to the adversary's strike strategies. This enables the network system maintainer to make more advantageous decisions. The method proposed in this invention has the following advantages:
[0072] This invention can achieve network coverage while completing strike prediction, thus keeping nodes in the network system connected.
[0073] This method can support static calculation of important node sequences based on network topology, thus enabling static attack prediction.
[0074] This method can support dynamic attack prediction based on the dynamic changes in the network topology of the network system maintainer and the opponent's attack strategy.
[0075] In a non-confrontational environment, this method can be used for network coverage and system maintenance in dynamic environments.
[0076] This method takes into account the impact of adversary attacks on the network system maintainer's network system, can predict attacks in response to adjustments in adversary attack strategies, and can perform system maintenance in dynamic adversarial environments. Attached Figure Description
[0077] Figure 1 This is the overall flowchart of the present invention.
[0078] Figure 2 This is a flowchart of the attack prediction module. Detailed Implementation
[0079] The present invention will now be described in further detail with reference to the accompanying drawings.
[0080] like Figure 1 As shown, this invention discloses a dynamic network architecture maintenance system under adversarial conditions, including an attack prediction module, a decision update module, and a pre-deployment decision module;
[0081] The attack prediction module is used to predict the sequence of important nodes that the opponent will attack in the next moment to maintain the network system.
[0082] The decision update module is used to update the opponent's attack strategy and feed the results back to the attack prediction module;
[0083] The pre-deployment decision module is used to make deployment decisions based on the prediction results of the network system maintainer.
[0084] The strike prediction module predicts the target nodes based on changes in the network architecture maintainer's topology and the opponent's strike strategy. The decision update module observes whether the opponent's strike strategy has changed. If it has changed, the updated strike strategy is fed back to the strike prediction module for re-prediction. If it has not changed, the pre-deployment decision module makes the decision, allowing the network architecture maintainer to take action in advance or deploy redundant silent communication nodes next to the important nodes to be attacked.
[0085] This invention discloses a dynamic network architecture maintenance method under adversarial conditions. It designs a decision-making architecture capable of predicting nodes that the network architecture maintainer's SoS (Social System of Entities) might be attacked in advance under adversarial conditions, referring to... Figure 1 The process and specific implementation of this invention are as follows:
[0086] Step 1: Set the opponent's initial attack strategy as follows: prioritize attacking communication nodes that are of high importance to the network system maintainers.
[0087] Step 2: Define the importance of communication nodes in the network architecture maintainer SoS as k = a1d + a2r + a3n + a4f, where d is the degree of the communication node, r is the range of services that the communication node can provide, n is the maximum number of system nodes that the communication node can serve, and f is the attack frequency of the node.
[0088] Step 3: Observe the topological changes of the SoS at time i-1 and time i, and determine the node attributes in the SoS based on the node state at the current time. The included feature information includes: the four indicators d, r, n, and f that constitute the importance k of the communication node, and the location information of each node. (e.g.) Figure 2(As shown)
[0089] Step 4: Establish an attack prediction optimization model based on the attributes of communication nodes and system nodes within the SoS; (e.g.) Figure 2 (As shown)
[0090] Step 4.1: Configure communication node C i The coverage area r, only within this range, system node S i Communication services are available;
[0091] Step 4.2: Configure communication node C i For system node S i Various communication transmission indicators; Communication node C i The signal transmission power and communication quality between nodes;
[0092] Step 4.3: Configure communication node C i The maximum number of services and the principle of proximity service, i.e., a certain system node S i Exceeding a certain communication node C i If the maximum number of services is exceeded, or if the distance to the node is greater than that, then communication node C cannot be accepted. i Services;
[0093] Step 4.4: Set node x i For redundant communication node C i Communication nodes that participate in forming a SoS;
[0094] Step 4.5: Configure each communication node C i Service cost v i ;
[0095] Step 4.6: Using steps 4.1 to 4.5 and step 3 as constraints, establish the objective as follows: The optimization model.
[0096] Step 5: Solve the model using the A-CADer algorithm, and finally output the sequence P{P1,P2,…,P} of nodes in the currently connected SoS with the highest probability of being attacked by the opponent. n};
[0097] Step 5.1: Generate the embedding vector for each node during the encoding stage. This vector contains static structure (node attributes maintained by the network architecture) and dynamically processed feature information. The encoder consists of two encodings: static and dynamic.
[0098] Step 5.2: Employ a multi-head attention mechanism to extract feature information from the network architecture maintainer nodes and generate static embeddings. The feature information of the communication nodes includes location, coverage radius r, and cost information v. i And the importance k of the communication node, and the system node S i Includes location information;
[0099] Step 5.3: Define the vector For communication node C i Within the coverage area, the number of system nodes not covered by the communication node selected in step t;
[0100] Step 5.4: In the set of uncovered system nodes, there are key system nodes for the entire solution. The dynamically encoded information is projected onto the embedding vector through linear projection. superior;
[0101] Step 5.5: Combine static and dynamic coding information to construct the final embedding vector of the communication node;
[0102] Step 5.6: The decoder takes the embedding vector output in step 5.5 as input. The decoding stage consists of k steps. In each step, the agent selects an unselected communication node and adds the selected communication node to the output node sequence. Then, the dynamic embedding is updated according to the environmental state.
[0103] Step 5.7: Repeat step 5.6 iteratively until a sequence of output nodes P{P1,P2,…,P} that satisfies all constraints is constructed. n} or reach the maximum number of iterations.
[0104] Step 6: Execute the pre-deployment decision module to observe the opponent's targets to determine if they have updated their attack strategy. If the opponent updates their attack strategy, proceed to Step 7; otherwise, use the existing sequence P{P1,P2,…,P}. n} Make deployment decisions.
[0105] Step 7: Execute the decision update module, adjust the importance k of the communication nodes, and output the new sequence P′{P1,P2,…,P n}
[0106] Based on this sequence and the actual adversarial situation, the network system maintainer (SoS) can make pre-deployment decisions, enabling communication nodes in the sequence to maneuver in advance or deploy redundant silent communication nodes around them.
[0107] The multi-head attention mechanism described in the above steps is an extension of the attention mechanism widely used in the Transformer model. It obtains the attention distribution of different subspaces of the input sequence by running multiple independent attention mechanisms in parallel, thereby capturing more comprehensive potential semantic associations in the sequence. This invention uses this method to extract the feature information of the network architecture maintenance nodes.
[0108] The main purpose of the static encoding described in the above steps is to extract the static structural information of the input node, which remains unchanged as the decoding steps proceed.
[0109] The dynamic encoding described above involves the agent selecting a communication node as the output at each decoding step. The coverage area of the selected communication node dynamically changes as the decoding process progresses. This dynamic state influences the selection of the next decoding step.
[0110] The embedding vector mentioned in the above steps Defined as communication node C i Within the coverage area, the number of system nodes not covered by the communication node selected in step t.
[0111] The update of the opponent strike prediction in the decision update module mentioned above is mainly based on observing the frequency changes of nodes within the opponent strike network system maintainer. If the frequency of the opponent's strikes against a certain node in the system increases around a certain time, it indicates that the opponent's strike strategy tends to target nodes with this characteristic during that period. This characteristic corresponds to d, r, and n in the node importance scoring index.
[0112] Table 1 compares the differences between the method of this invention and existing methods. Network coverage methods mainly consider how to quickly achieve coverage; some methods, such as CADer, can be used for system maintenance under non-dynamic adversarial conditions. Node importance assessment methods do not consider the coverage problem and therefore cannot support system integrity maintenance. The solution of this invention considers how to quickly maintain the system under dynamic adversarial conditions and can simultaneously support static attack prediction based on network topology and attack prediction based on the adversary's attack strategy. Therefore, the solution of this invention is superior to existing solutions in addressing the maintenance of command network systems under dynamic system adversarial conditions.
[0113] Table 1 Comparison of Functions of Different Methods
[0114]
[0115] This method can achieve network coverage in a dynamic adversarial environment, ensuring that all nodes in the SoS (Social System Architecture) of the network maintainer remain connected.
[0116] This method can predict the opponent's attack nodes in a dynamic adversarial environment based on the static attributes of the network system maintainer's nodes and the opponent's dynamic attack strategy.
[0117] This method can quickly update the strike prediction model based on changes in the opponent's strike strategy, ensuring the accuracy of the strike prediction sequence.
[0118] This method can predict attacks in dynamic environments based on changes in the network topology of the network maintainer and the opponent's attack strategies.
[0119] This method takes into account the impact of adversary attacks on the network system maintainer's network system, can predict attacks in response to adjustments in adversary attack strategies, and can perform system maintenance in dynamic adversarial environments.
[0120] The above embodiments are only used to illustrate the design concept and features of the present invention, and their purpose is to enable those skilled in the art to understand the content of the present invention and implement it accordingly. The protection scope of the present invention is not limited to the above embodiments. Therefore, all equivalent changes or modifications made based on the principles and design ideas disclosed in the present invention are within the protection scope of the present invention.
Claims
1. A method for using a dynamic network architecture maintenance system under adversarial conditions, characterized in that, Includes the following steps; Step (1): Constructing the strike prediction module: An optimization model is established based on the opponent's current attack strategy and the node attributes of the network system maintainer. The attributes of each node in the network system maintainer's SoS are used as input, and the A-CADer algorithm is used to solve the optimization model. The output is the sequence of nodes with the highest probability of being attacked by the opponent in the SoS connected at the current time. Step (2): Decision Update Module: Observe whether the opponent's attack strategy changes. If it changes, feed back the updated attack strategy to the attack prediction module for re-prediction. If it does not change, the pre-deployment decision module will make the decision. Step (3): Pre-deployment decision module: Based on the predictions of the network system maintainers, take proactive measures in advance, or deploy redundant silent communication nodes next to key nodes that will be attacked. The specific steps in step (1) are as follows: Step 1: Set the initial attack strategy as follows: prioritize attacking communication nodes with high importance; Step 2: Define the importance of communication nodes in the SoS as follows: ， in d The degree of the communication node. r The range of services that a communication node can provide. n The maximum number of system nodes that a communication node can serve. ，f This is the frequency at which the communication node was attacked; The weighted average coefficient; Step 3: Observe SoS at the 1st... i -1 time and the i Changes in the topology at any given time update the node attributes in the SoS; Step 4: Establish an optimization model based on the attributes of communication nodes and system nodes within the SoS; Step 5: Use the A-CADer algorithm to solve the optimization model, and finally output the sequence of nodes with the highest probability of being attacked by the opponent in the connected SoS at the current time. In step 3, the feature information contained in the node attributes is determined based on the node state at the current moment: the importance of the communication node. The four indicators of communication node degree The range of services that communication nodes can provide r The maximum number of system nodes that a communication node can serve. n The frequency at which this node is attacked f And the location information of each node; The specific steps are as follows: Step 3.1: Update the degree index of nodes in the current network based on the network architecture maintainer's current data. d ; Step 3.2: Update the indicators based on the range of services that communication nodes in the current network can provide, according to the network architecture maintainer. r ; Step 3.3: Update the index based on the maximum number of system nodes that communication nodes in the current network can serve, according to the network architecture maintainer. n ; Step 3.4: Update the indicators based on the attack frequency of nodes in the current network maintained by the network architecture maintainer. f ; Step 5 is described in detail below: Step 5.1: Generate the embedding vector for each node during the encoding stage. This vector contains static structure and dynamic processing feature information. The encoder consists of two encodings: static and dynamic. Step 5.2: Extract feature information of the network architecture maintainer nodes using a multi-head attention mechanism and generate static embeddings. The feature information of the communication nodes includes location and coverage radius. Cost information and the importance of communication nodes , and system nodes Includes location information; Step 5.3: Define the vector For communication nodes Within the coverage area, in the steps t The number of system nodes not covered by the selected communication node; Step 5.4: In the set of uncovered system nodes, there are key system nodes for the entire solution. The dynamically encoded information is projected onto the embedding vector through linear projection. superior; Step 5.5: Combine static and dynamic coding information to construct the final embedding vector of the communication node; Step 5.6: The decoder uses the embedding vector output in step 5.5 as input, and the decoding stage is performed by... h The process consists of several steps. In each step, the agent selects an unselected communication node and adds the selected communication node to the output node sequence. Then, it updates the dynamic embedding based on the environmental state. Step 5.7: Repeat step 5.6 iteratively until a sequence of output nodes that satisfies all constraints is constructed. Or it may reach the maximum number of iterations.
2. The method of using a dynamic network architecture maintenance system under adversarial conditions according to claim 1, characterized in that, Step 4 specifically involves: Step 4.1: Configure the communication node Coverage Within this scope only, system nodes Obtain communication services; Step 4.2: Configure the communication node For system nodes Various communication transmission indicators; Step 4.3: Configure the communication node The maximum number of services and the principle of proximity service, i.e., a certain system node Exceeding a certain communication node If the maximum number of services is exceeded, or if the distance to the node is too far, the communication node cannot be accepted. Services; Step 4.4: Configure nodes For redundant communication nodes Communication nodes that participate in forming a SoS; Step 4.5: Configure each communication node service costs ; Step 4.6: Using steps 4.1 to 4.5 and step 3 as constraints, establish the objective as follows: The optimized model.
3. The method of using a dynamic network architecture maintenance system under adversarial conditions according to claim 1, characterized in that, In step (2), as the attack strategy is updated, SoS determines the importance of communication nodes based on the distribution of the actual attack nodes by the adversary. Changes in the weighting coefficients of each indicator The value of is used to execute the strike prediction module again before the pre-deployment decision module makes the final decision, resulting in a new sequence based on the updated adversary strike strategy. .
4. The method of using a dynamic network architecture maintenance system under adversarial conditions according to claim 3, characterized in that, Step (2) is as follows: Step 1: If the attack strategy is updated to prioritize attacking the communication node with the highest number of service system nodes in the SoS at a certain moment, increase the coefficient. Adjust the importance level; Step 2: If the attack strategy is updated to prioritize attacking the communication node with the largest service range in the SoS of the network system maintainer at a certain time, increase the coefficient. Adjust the importance level; Step 3: If the attack strategy is updated to prioritize attacking the communication node with the strongest service capacity in the SoS at a certain moment, increase the coefficient. Adjust the importance level; Step 4: If the attack strategy is updated to prioritize attacking the communication node in the SoS that has been attacked the most times, increase the coefficient. Adjust the importance level; Through the above steps, the decision update module automatically adjusts the model established by the strike prediction module according to the changes in the strike strategy, and updates the prediction sequence.
5. The method of using a dynamic network architecture maintenance system under adversarial conditions according to claim 4, characterized in that, Step (3) specifically involves: The predictive attack module outputs the sequence of nodes with the highest probability of being attacked by the opponent in the currently connected SoS. Then, make decisions and deployments based on the confrontation situation at this moment; Specifically as follows: Step 1: Observe whether the opponent's attack strategy has been updated; Step 2: If the attack strategy has been updated, execute the decision update module and output the new sequence. Otherwise, follow the existing sequence. Make decisions; Step 3: Based on this sequence and the actual adversarial situation, SoS makes a pre-deployment decision, so that the communication nodes in the sequence can make maneuvers in advance, or deploy redundant silent communication nodes around them.
6. The method of using a dynamic network architecture maintenance system under adversarial conditions according to claim 1, characterized in that, The method is implemented through a dynamic network architecture maintenance system under adversarial conditions, which includes an attack prediction module, a decision update module, and a pre-deployment decision module. The attack prediction module is used to predict the sequence of important nodes that the opponent will attack in the next moment to maintain the network system. The decision update module is used to update the opponent's attack strategy and feed the results back to the attack prediction module; The pre-deployment decision module is used to make deployment decisions based on the prediction results of the network system maintainer; The attack prediction module predicts attack nodes based on changes in the network architecture maintainer's topology and the opponent's attack strategy. The decision update module observes whether the opponent's attack strategy has changed. If it has changed, the updated attack strategy is fed back to the attack prediction module for re-prediction. If it has not changed, the pre-deployment decision module makes the decision, so that the network system maintainers can take action in advance or deploy redundant silent communication nodes next to the important nodes to be attacked.