Method and system for determining 6G air-space-ground network security policy based on AI
By deploying AI-driven cross-layer general model and hierarchical sub-models in the space-space integrated network, the problem of lack of endogenous design of the existing technology's air-space integrated network security means is solved, and the rapid identification and response of security events is achieved. , improves network security protection capabilities and decision-making efficiency.
Patent Information
- Application Number
- CN202410878682.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-02
- Publication Date
- 2024-11-15
AI Technical Summary
The existing integrated network architecture of the sky and the earth uses 'plug-in' and 'patch' security means, lacking a security endogenous design, making it difficult to eradicate network intrusion, network eavesdropping, message modification/injection, man-in-the-middle attack, and distributed denial of service attack. Such stubborn security issues make it difficult to achieve three-dimensional defense of heterogeneous aerospace networks.
Using AI-based 6G aerospace and earth network security strategy method, the real-time monitoring, analysis and decision-making of security events are realized by deploying cross-layer general large models, hierarchical sub-models and professional small models in the cloud and edge servers of the ground core network, and real-time monitoring, analysis and decision-making of security events are achieved, forming Security sub-strategy, unified scheduling and orchestration of network security capabilities, and realizes security endogenous design and three-dimensional defense of the integrated air-space and earth network.
The security endogenous design and three-dimensional defense of the integrated aerospace and earth network have been realized, the network security protection capabilities have been improved, and the ability to quickly and accurately judge and respond to security incidents have been improved, and the efficiency and intelligence of network security decisions have been improved.
Smart Images

Figure CN118972853A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of network security, and in particular to a method and system for determining a 6G air-space-ground network security strategy based on AI. Background Art
[0002] The sixth generation mobile networks (6G) are expected to be commercially available in 2030, supporting a variety of heterogeneous networks and massive terminal devices including satellite networks, industry networks, and body area networks, and realizing the true intelligent connection of all things. ITU-R proposed six typical scenarios for 6G, including immersive communication, ultra-large-scale connection, extremely reliable and low-latency communication, artificial intelligence (AI) and communication integration, interawareness and ubiquitous connection; it also clarified the four overall goals of the network, including sustainability, ubiquitous connection, security, privacy, resilience, and ubiquitous intelligence; and required the network to inherently support new capabilities such as AI, perception, and ubiquitous connection, to achieve a major shift from mobile communication to mobile information services.
[0003] With the increase in communication business volume and the diversification of business types, large-scale 5th Generation mobile communication technology (5G) network deployment requires high costs, and the existing ground mobile communication system is difficult to meet the communication needs of users in remote rural areas, deserts, etc. The 6G-oriented space-air integrated network deeply integrates space-based (high-orbit / medium-orbit / low-orbit satellites), air-based (air-based / high-altitude / low-altitude aircraft) and ground-based (cellular / WiFi / wired) networks to establish a three-dimensional space network for the entire region, meeting the full-area, all-weather ubiquitous coverage needs of the surface and three-dimensional space, and enabling users to access on demand anytime, anywhere.
[0004] Compared with traditional communication networks, the integrated air-ground-space network has the characteristics of massive terminal access, dynamic channel opening, heterogeneous network architecture, and cross-domain network element interconnection. It requires in-depth research on security issues such as authentication and authorization security, data transmission security, and cross-domain attack tracing. At present, the integrated air-ground-space network architecture uses "plug-in" and "patch" security methods, lacks inherent security design, and is difficult to cure network intrusion, network eavesdropping, message modification / injection, man-in-the-middle attacks, distributed denial of service attacks and other stubborn security issues, and it is difficult to achieve three-dimensional defense of heterogeneous air-ground-space networks. Summary of the invention
[0005] The present invention provides an AI-based method and system for determining a 6G air-space-ground network security strategy, which is used to solve the problem that the current air-space-ground integrated network architecture uses "plug-in" and "patch-type" security measures, lacks an inherent security design, is difficult to fundamentally solve stubborn security problems such as network intrusion, network eavesdropping, message modification / injection, man-in-the-middle attacks, and distributed denial of service attacks, and is difficult to achieve three-dimensional defense of air-space-ground heterogeneous networks. This invention realizes the inherent security design of the air-space-ground integrated network architecture and realizes three-dimensional defense of air-space-ground heterogeneous networks.
[0006] A method for determining 6G air-space-ground network security strategy based on AI is applied to the cloud corresponding to the ground core network, comprising: in response to a task instruction, sending the task instruction to the corresponding terminal device or node through at least one edge server in the ground-based network, the air-based network or the space-based network; receiving a security event set and a security sub-policy set corresponding to the task instruction; wherein the security event set is determined based on at least one professional small model, and the security sub-policy set is determined based on at least one hierarchical sub-model; the at least one professional small model is respectively deployed on the corresponding terminal device or node; the at least one hierarchical sub-model is distributedly deployed on the edge servers corresponding to each layer of the ground-based network, the air-based network and the space-based network; the security event set and the security sub-policy set are input into a pre-trained network security cross-layer general large model to generate a cross-layer security decision; the cross-layer security decision is used to uniformly schedule and orchestrate the network security capabilities of the ground-based network, the air-based network and the space-based network.
[0007] In one of the embodiments, in response to the task instruction, the task instruction is sent to the corresponding terminal device or node through at least one edge server in the ground-based network, the air-based network, and the space-based network, including: in response to the task instruction issued by the user or the state-aware task instruction of the network device, the task instruction is sent to the corresponding terminal device or node through at least one edge server in the ground-based network, the air-based network, or the space-based network.
[0008] In one embodiment, the pre-trained network security cross-layer general large model is a large language model. Before the security event set and the security sub-policy set are input into the pre-trained network security cross-layer general large model to generate a cross-layer security decision, the method further includes: based on a first preset prompt word set, fine-tuning the large language model to obtain the pre-trained network security cross-layer general large model; wherein the first preset prompt word set includes a historical security event set and a security sub-policy set determined based on the historical security event set.
[0009] In one of the embodiments, the at least one hierarchical sub-model includes at least one ground-based network sub-model, at least one air-based network sub-model and at least one space-based network sub-model, and the edge server includes an edge service satellite, an edge service drone and a ground edge server; the at least one hierarchical sub-model is distributedly deployed on the edge servers corresponding to each layer of the ground-based network, the air-based network and the space-based network, including: the at least one ground-based network sub-model is distributedly deployed on the ground edge server; the at least one air-based network sub-model is distributedly deployed on the edge service drone; the at least one space-based network sub-model is distributedly deployed on the edge service satellite.
[0010] The present invention also provides an AI-based method for determining a 6G air-space-ground network security strategy, which is applied to edge servers corresponding to ground-based networks, air-based networks or space-based networks, respectively, and includes: receiving a task instruction from a cloud corresponding to a ground core network, and sending the task instruction to a corresponding terminal device or node; receiving a security event set corresponding to the task instruction; the security event set is determined based on at least one professional small model, and the at least one professional small model is deployed on a corresponding terminal device or node; based on the security event set, at least one pre-trained hierarchical sub-model, and a preset security policy database, a security sub-policy set is determined; the security sub-policy set is used to uniformly schedule and orchestrate the security capabilities of the distributed subnet where the edge server is located; the security event set and the security sub-policy set are sent to the cloud corresponding to the ground core network; the cloud corresponding to the ground core network is used to execute the AI-based method for determining a 6G air-space-ground network security strategy corresponding to any of the above embodiments.
[0011] In one embodiment, the pre-trained hierarchical sub-model includes a retrieval enhancement generation module and an autonomous organization generation module, and the security sub-policy set is determined based on the security event set, at least one pre-trained hierarchical sub-model, and a preset security policy database, including: retrieving the security policies corresponding to each security event in the security event set from the preset security policy database through the retrieval enhancement generation module; for each security event, when it is determined that the preset security policy database includes the security policy corresponding to the security event, determining the security policy corresponding to the security event as the security sub-policy of the security event; or, for each security event, when it is determined that the preset security policy database does not include the security policy corresponding to the security event, inputting the security event into the corresponding autonomous organization generation module, and the autonomous organization generation module generates a security sub-policy for the security event.
[0012] In one of the embodiments, before determining the security sub-policy set based on the security event set, the pre-trained hierarchical sub-model, and the preset security policy database, the method also includes: based on a second preset prompt word set, fine-tuning the lightweight large model to obtain the pre-trained hierarchical sub-model; wherein the lightweight large model is a large model after pruning or a large model after knowledge distillation; wherein the second preset prompt word set includes historical security event information, and the security event information includes the network environment involved in the security event, the type of security event, the priority of the security event, the location where the security event occurred, the way the security event occurred, and the impact caused by the security event.
[0013] The present invention also provides an AI-based method for determining a 6G air-space-ground-network security policy, which is applied to a terminal device or node, and includes: receiving a task instruction from a cloud corresponding to a ground core network; based on the task instruction and at least one professional small model, obtaining a security event set corresponding to the task instruction; sending the security event set to at least one edge server in a ground-based network, an air-based network, and a space-based network, the at least one edge server being used to execute the AI-based method for determining a 6G air-space-ground-network security policy corresponding to any of the above-mentioned embodiments; receiving and executing a security sub-policy in a security sub-policy set sent by the at least one edge server, or receiving and executing a cross-layer security decision sent by the cloud corresponding to the ground core network; the cloud corresponding to the ground core network is used to execute the AI-based method for determining a 6G air-space-ground-network security policy corresponding to any of the above-mentioned embodiments.
[0014] In one of the embodiments, the at least one professional mini-model includes a threat discovery and alarm mini-model, a vulnerability discovery and patching mini-model, a fault monitoring and repair mini-model, and an intrusion detection and interception mini-model.
[0015] The present invention also provides an AI-based 6G air-space-ground network security strategy determination system, the system comprising: a network security cross-layer general large model, at least one hierarchical sub-model and at least one professional small model; wherein the at least one professional small model is deployed on a terminal device and a node to realize real-time monitoring of network traffic, user behavior and system status, and discover security events; the at least one professional small model is also used to respond to security sub-strategies and cross-layer security decisions; the terminal device and the node are used to execute the AI-based 6G air-space-ground network security strategy determination method corresponding to any of the above embodiments; the at least one hierarchical sub-model is distributed in a distributed manner; The at least one hierarchical sub-model is used to generate a security sub-strategy; the at least one edge server is used to execute the AI-based 6G air-space-ground-network security strategy determination method corresponding to any of the above-mentioned embodiments; the network security cross-layer general large model is deployed on the cloud corresponding to the ground core network, and is used to generate cross-layer security decisions for the ground-based network, air-based network and space-based network based on the security incident and the security sub-strategy; the cloud corresponding to the ground core network is used to execute the AI-based 6G air-space-ground-network security strategy determination method corresponding to any of the above-mentioned embodiments.
[0016] The above-mentioned AI-based 6G air-space-ground network security strategy determination method and system, by deploying corresponding models in a layered manner in terminal equipment, at least one edge server in the ground-based network, air-based network or space-based network, and the cloud corresponding to the ground core network (air-space-ground integrated network security cross-layer general large model, hierarchical sub-model, professional small model), realizes a multi-layer intelligent collaboration mechanism among the air-space-ground integrated network security cross-layer general large model, hierarchical sub-model, and professional small model, integrates intelligent perception, intelligent analysis, intelligent decision-making and intelligent orchestration into every link of the security system, and realizes the architectural-level intelligent endogenous generation of network security strategies. BRIEF DESCRIPTION OF THE DRAWINGS
[0017] Figure 1 A schematic diagram of the air-ground integrated network security architecture provided by the present invention;
[0018] Figure 2 One of the flow charts of the method for determining the 6G air-space-ground network security strategy based on AI provided by the present invention;
[0019] Figure 3 The second flowchart of the method for determining the 6G air-space-ground network security strategy based on AI provided by the present invention;
[0020] Figure 4 The third flowchart of the method for determining the 6G air-space-ground network security strategy based on AI provided by the present invention;
[0021] Figure 5 A fourth flowchart of a method for determining a 6G air-space-ground network security strategy based on AI provided by the present invention;
[0022] Figure 6 A schematic diagram of the framework of the AI-based 6G air-space-ground network security strategy determination system provided by the present invention. DETAILED DESCRIPTION
[0023] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0024] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the embodiments of the present disclosure should be understood by people with ordinary skills in the field to which the present disclosure belongs. The "first", "second" and similar words used in the embodiments of the present disclosure do not indicate any order, quantity or importance, but are only used to distinguish different components. "Including" or "comprising" and similar words mean that the elements or objects appearing before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the described object changes, the relative positional relationship may also change accordingly.
[0025] In order to facilitate understanding, the technical terms involved in the present invention are first explained.
[0026] In traditional deep learning methods, downstream tasks usually need to be fine-tuned independently on specific domains or tasks, which may require a large amount of labeled data, and the performance of the model is constrained by the differences between tasks and insufficient data. Therefore, researchers began to think about how to better utilize the versatility of pre-trained models and large-scale text and image data to improve the performance of multimodal models on various downstream tasks.
[0027] The essence of prompt learning is to restructure different downstream tasks into the form of pre-trained tasks. The core idea of this method is to use a specific template to convert the data of downstream tasks into natural language form, so as to better utilize the language understanding and visual understanding capabilities of the pre-trained model. The design of this template is the key, which needs to fully tap the potential of the pre-trained model itself so that it can better adapt to various downstream tasks with as little labeled data as possible.
[0028] The process of Prompt Learning can be summarized in the following three steps: First, pre-training is performed, in which the model is trained on large-scale text and image data to acquire rich language and visual knowledge. Then, by designing specific templates, the data of the downstream task is converted into a natural language form similar to the pre-training task. Finally, the model is fine-tuned under these natural language prompts to adapt to specific downstream tasks, thereby achieving better performance.
[0029] In summary, Prompt Learning is an innovative approach that fully exploits the versatility and knowledge of pre-trained models on multimodal data by reframing downstream tasks, thereby improving the performance of models on various tasks.
[0030] Next, the inventive concept of the present invention is described.
[0031] The open heterogeneity and cross-domain interconnection of the air-ground integrated network increase the network security exposure. The traditional "plug-in" and "patch-type" network security architecture is difficult to meet the security requirements of the 6G air-ground integrated network. In addition, the various layers of the network in the air-ground integrated network have different technical standards, frequency bands and transmission methods. The traditional centralized security decision-making method with low efficiency and slow response is no longer applicable.
[0032] The big model has powerful natural language processing capabilities and has achieved remarkable results in many fields. As a pre-trained basic model, the scope, quantity and type of intelligent business undertaken by the big model are unmatched by previous AI models. The big model has excellent generalization capabilities and can adapt to specific tasks with very little downstream data, saving a lot of time for data collection, cleaning and annotation for various downstream tasks, greatly saving development costs. The big model in the field of communications will be able to make full use of the rich semantic knowledge it contains, adapt to complex and heterogeneous air-space-ground networks and multimodal data, and further improve network security management and control capabilities.
[0033] Therefore, first of all, the present invention proposes a method for determining an AI-based 6G air-space-ground network security strategy based on a large model. This method distributes the large language model to multiple nodes in an air-space-ground integrated network, allowing information exchange, knowledge sharing, and collaborative reasoning between models to complete rapid and accurate judgment of security incidents, promote the efficiency and intelligence of security decision-making in ubiquitous connection scenarios, and realize intelligent security collaboration of cloud-edge-end in an air-space-ground integrated distributed network.
[0034] Secondly, the AI-endogenous overall security architecture of the air-space-ground integrated network proposed in the present invention distributes the layered language model of cross-layer general large models, hierarchical sub-models and professional small models in the air-space-ground integrated network architecture, forming a security mechanism with layered protection, dynamic adjustment and overall control, and realizing the endogenous security, ubiquitous intelligence and three-dimensional collaborative defense of the 6G air-space-ground integrated network.
[0035] In addition, the current security challenges of the integrated air-space-ground-integrated network are multidimensional. There are differences between space-based, air-based and ground-based networks, which require specific security capabilities. The large-model security framework proposed in the present invention is deployed through the entire network, and professional small models are deployed on terminal nodes to perceive information. The hierarchical sub-models process the information and generate strategies. The cross-layer general large model arranges security capabilities to ensure the full-scenario security of the integrated air-space-ground-integrated network, enhance the overall security protection capabilities of the integrated air-space-ground-integrated network, and form a complete security chain of "perception-decision-response" based on the layered large model.
[0036] The following describes the AI-based 6G air-space-ground network security strategy determination method and system provided by the present invention in conjunction with the accompanying drawings.
[0037] Figure 1 A schematic diagram of the air-ground-integrated network security architecture provided by the present invention.
[0038] like Figure 1 As shown, the network security architecture includes space-based network, air-based network, and ground-based network.
[0039] The space-based network includes edge service satellites and other satellites. The edge service satellite is equipped with a hierarchical sub-model of integrated space-ground network security, and the other satellites are equipped with a small professional sub-model of integrated space-ground network security. There may be security threats such as signal interference, network attacks, signal eavesdropping and data leakage in the space-based network.
[0040] The air-based network includes edge service drones and other drones. The edge service drone is deployed with an air-ground integrated network security hierarchical sub-model, and the other drones are deployed with an air-ground integrated network security professional small model. There may be security threats such as signal interference and deception, network attacks, physical security threats, data theft and leakage, identity disguise and man-in-the-middle attacks in the air-based network.
[0041] The ground-based network includes immersive communications, ultra-large-scale connections, cellular access networks, the Internet of Things, the Internet, smart factories and other ground-based network scenarios. Each network scenario includes corresponding edge servers and terminal devices, and each network can be connected to the core network. Among them, a large cross-layer general model of integrated space-ground network security is deployed on the core network, a hierarchical sub-model of integrated space-ground network security is deployed on the edge server, and a small professional model of integrated space-ground network security is deployed on the terminal device. There may be security threats such as fake base station attacks, radio interference and blocking, network slicing hijacking, signaling storm attacks, distributed denial of service attacks, cross-network attacks, eavesdropping attacks, signaling / data tampering, and unauthorized access in the ground-based network.
[0042] Figure 2 A flow chart of a method for determining a 6G air-space-ground network security strategy based on AI provided by the present invention. It can be understood that the method for determining a 6G air-space-ground-network security strategy based on AI is applied to the cloud corresponding to the ground core network. In the present invention, virtualization technology is used to abstract physical network resources into multiple virtual networks, each of which can be independently configured and managed. Security functions are abstracted and no longer rely on specific hardware. Instead, as part of the virtualization service, they can be flexibly deployed anywhere in the network.
[0043] like Figure 2 As shown, in one embodiment, a method for determining a 6G air-space-ground network security strategy based on AI is proposed, which may specifically include the following steps:
[0044] Step 210, in response to the task instruction, the task instruction is sent to the corresponding terminal device or node through at least one edge server in the ground-based network, air-based network or space-based network.
[0045] Among them, the mission instructions may include instructions for monitoring network security threats. Among them, network security threats include ground-based network security threats, air-based network security threats, and space-based network security threats. For specific network security threats, please refer to Figure 1 For the sake of brevity, the relevant description in is not repeated here.
[0046] Among them, the at least one edge server includes an edge service satellite, an edge service drone and a ground edge server; the terminal device or node can be a satellite in a space-based network, a drone in an air-based network, and a ground edge server in a ground-based network.
[0047] Step 220, receiving a security event set and a security sub-policy set corresponding to the task instruction; wherein, the security event set is determined based on at least one professional small model, and the security sub-policy set is determined based on at least one hierarchical sub-model; the at least one professional small model is deployed on a corresponding terminal device or node respectively; the at least one hierarchical sub-model is distributedly deployed on edge servers corresponding to each layer of the ground-based network, air-based network and space-based network.
[0048] Among them, the security event set includes security events in terms of network traffic, user behavior and system status, which can specifically correspond to the above-mentioned security threats.
[0049] It can be understood that the main function of the professional small model is to collect and process multimodal information in the network environment in real time to achieve comprehensive monitoring of the network status and anomaly detection. The professional small model can be an existing small model with good performance. Different professional small models are obtained based on pre-training of data for different scenarios. The present invention does not limit the professional small model.
[0050] It should be noted that by adopting professional small models, various types of information are deeply mined and analyzed to improve network security protection capabilities and ensure the security of the entire architecture. For different security scenarios, corresponding state-aware professional small models can be deployed to collect multimodal information, such as network traffic information, identity authentication information, user abnormal behavior information, etc., to achieve refined processing of complex security issues. That is, the present invention utilizes the advantages of professional small models in big data analysis, behavior recognition and pattern recognition, and deploys professional small models on terminal devices and nodes, so that terminal devices and nodes have the ability of scene perception and information processing, and realizes real-time monitoring of network traffic, user behavior and system status, and timely discovers and responds to potential security threats in specific scenarios and fields.
[0051] It can also be understood that once the professional small model discovers a potential risk, the professional small model immediately reports the security incident to the integrated air-ground-space network security hierarchical sub-model.
[0052] Among them, the hierarchical sub-models include ground-based network sub-models, air-based network sub-models, and space-based network sub-models. These hierarchical sub-models are responsible for disassembling, analyzing, and arranging solutions for security incidents, and ultimately generating security sub-strategies. In this process, the interaction process between the hierarchical sub-model of integrated air-ground-air network security and the professional small models is very close. The security sub-strategies are sent from the hierarchical sub-models to the professional small models, and the professional small models make specific security responses based on the sent security sub-strategies. This hierarchical interaction method not only improves the efficiency of security decision-making, but also realizes the effective coordination of security decision-making support between different scenarios, comprehensively improving the security of the entire network.
[0053] It can be understood that in the traditional centralized architecture, data and computing resources are concentrated in a central node, which is easy to cause resource bottlenecks and single point failures. In addition, since the data form that needs to be processed by the centralized architecture is relatively single, the large model deployment method does not need to be very complicated. The integrated air-space-ground network covers a variety of different access domains, and each access domain will generate its own specific data stream. Facing the 6G ultra-large-scale connection scenario, different application scenarios and business needs are involved in the network, and a large number of sensors and devices will also be involved in the Internet of Things, and the format and content of the data will also be diversified. In the present invention, by distributing and deploying professional small models and hierarchical sub-models, computing tasks can be assigned to each node in the network, making full use of distributed computing power and data resources, thereby achieving more efficient data processing and decision reasoning.
[0054] Step 230, input the security event set and the security sub-strategy set into a pre-trained network security cross-layer general model to generate cross-layer security decisions; the cross-layer security decisions are used to uniformly schedule and orchestrate the network security capabilities of ground-based networks, air-based networks, and space-based networks.
[0055] Combined with the previous text, it can be seen that the hierarchical sub-model is a model deployed in different environments (ground-based networks, air-based networks, and space-based networks) to solve specific field problems, but the hierarchical sub-model is deployed on the edge service area of a single subnet, which limits its ability to solve cross-layer problems, that is, the hierarchical sub-model is still difficult to solve the network security problems under the intersection of multiple environments (ground-based networks, air-based networks, and space-based networks). Therefore, for cross-layer events, it is necessary to transfer data to the cloud-based air-ground integrated network security cross-layer general large model for processing. This multi-layer large model interactive architecture solution can not only effectively solve professional specific field problems, but also can handle general cross-domain problems, thereby comprehensively improving the overall performance and continuously solving network security problems. Therefore, this application generates cross-layer security decisions based on a cross-layer general large model of network security to solve cross-layer security problems.
[0056] The AI-based method for determining the 6G air-space-ground network security strategy provided by the present invention realizes a multi-layer intelligent collaboration mechanism among the cross-layer general large model, hierarchical sub-model, and professional small model of the air-space-ground integrated network security by deploying corresponding models in a layered manner in the terminal device, at least one edge server in the ground-based network, the air-based network, or the space-based network, and the cloud corresponding to the ground core network, thereby integrating intelligent perception, intelligent analysis, intelligent decision-making, and intelligent orchestration into every link of the security system and realizing the endogenous intelligence of the network security strategy at the architectural level.
[0057] In one embodiment, the above step 210 specifically includes: in response to a task instruction issued by a user or a state-aware task instruction of a network device, the task instruction is sent to a corresponding terminal device or node through at least one edge server in a ground-based network, an air-based network, or a space-based network.
[0058] The network device may be any network device among ground-based network, air-based network and space-based network.
[0059] In one embodiment, the pre-trained network security cross-layer general large model is a large language model. Before the security event set and the security sub-policy set are input into the pre-trained network security cross-layer general large model to generate a cross-layer security decision, the method further includes: based on a first preset prompt word set, fine-tuning the large language model to obtain the pre-trained network security cross-layer general large model.
[0060] Among them, the large language model can be an existing artificial intelligence-based language model, such as a pre-trained language model based on the Transformer architecture (Generative Pre-trained Transformer, GPT).
[0061] The first preset prompt word set includes a historical security event set and a security sub-policy set determined based on the historical security event set.
[0062] Exemplarily, one of the preset prompt words in the above-mentioned first preset prompt word set can be: "You are an expert in the field of integrated air-ground-space network security, and are good at solving large-scale network attacks, complex network attacks, cross-domain network attacks, etc. Now attack 1 is found in network 1, attack 1 is found in network 2, and attack 3 is found in network 3. The security strategies given by each hierarchical sub-model are {strategy 1, strategy 2, strategy 3} respectively. You now need to comprehensively consider the strategies of all hierarchical sub-models and give a comprehensive strategy. It can be understood that the above-mentioned network 1, network 2, and network 3 can correspond to any one of the ground-based network, air-based network, and space-based network. Similarly, attack 1, attack 2, and attack 3 correspond to security incidents in actual scenarios, and strategy 1, strategy 2, and strategy 3 also correspond to security sub-strategies in actual applications."
[0063] It can be understood that the large language model is fine-tuned by the first preset prompt word set so that it can have the ability to generate cross-layer and cross-layer security decisions.
[0064] In one of the embodiments, the at least one hierarchical sub-model includes at least one ground-based network sub-model, at least one air-based network sub-model and at least one space-based network sub-model, and the edge server includes an edge service satellite, an edge service drone and a ground edge server; the at least one hierarchical sub-model is distributedly deployed on the edge servers corresponding to each layer of the ground-based network, the air-based network and the space-based network, including: the at least one ground-based network sub-model is distributedly deployed on the ground edge server; the at least one air-based network sub-model is distributedly deployed on the edge service drone; the at least one space-based network sub-model is distributedly deployed on the edge service satellite.
[0065] It can be understood that since the hierarchical sub-model of the present invention is deployed on the edge server, and the performance of the edge server is poor, by distributing the hierarchical sub-model, the computing tasks can be distributed to each node in the network, making full use of the distributed computing power and data resources, thereby achieving more efficient data processing and decision reasoning.
[0066] In one embodiment, if Figure 3 As shown, the present invention also provides a method for determining a 6G air-space-ground network security strategy based on AI, which is applied to edge servers corresponding to a ground-based network, an air-based network, or a space-based network, respectively, and includes the following steps:
[0067] Step 310, receiving a task instruction from the cloud corresponding to the ground core network, and sending the task instruction to the corresponding terminal device or node.
[0068] Step 320, receiving a security event set corresponding to the task instruction; the security event set is determined based on at least one professional small model, and the at least one professional small model is deployed on a corresponding terminal device or node respectively.
[0069] Step 330, based on the security event set, at least one pre-trained hierarchical sub-model, and a preset security policy database, determine a security sub-policy set; the security sub-policy set is used to uniformly schedule and orchestrate the security capabilities of the distributed subnet where the edge server is located.
[0070] Step 340, sending the security event set and the security sub-policy set to the cloud corresponding to the ground core network; the cloud corresponding to the ground core network is used to execute the AI-based 6G air-space-ground network security policy determination method corresponding to any of the above corresponding embodiments.
[0071] It can be understood that since this method is similar to the above Figure 2 The methods in can correspond, so for the sake of brevity, similar content is not repeated here.
[0072] In one embodiment, if Figure 4 As shown, the pre-trained hierarchical sub-model includes a retrieval enhancement generation module and an autonomous organization generation module, and the above step 330 may include the following steps:
[0073] Step 410: The security policy corresponding to each security event in the security event set is retrieved from the preset security policy database through the retrieval enhancement generation module.
[0074] The preset security policy database includes core resource libraries such as network-wide security data, threat intelligence, and basic resource libraries corresponding to ground networks, satellite networks, and user terminals. In the process of storing and retrieving data from the preset security policy database, vector database technology can be used to accelerate the model's access to and processing of data.
[0075] It can be understood that through the retrieval augmented generation module, which mainly adopts the retrieval augmented generation technology (RAG), combined with the retrieved external knowledge, the model can access the latest research and case analysis, thereby providing more accurate solution strategies and defense suggestions, thereby enhancing the model's security strategy generation capabilities.
[0076] Step 420: For each security event, when it is determined that the preset security policy database includes the security policy corresponding to the security event, the security policy corresponding to the security event is determined as the security sub-policy of the security event.
[0077] Step 430, for each security event, when it is determined that the preset security policy database does not include the security policy corresponding to the security event, the security event is input into the corresponding autonomous organization generation module, and the autonomous organization generation module generates a security sub-policy for the security event.
[0078] In one of the embodiments, before determining the security sub-policy set based on the security event set, the pre-trained hierarchical sub-model, and the preset security policy database, the method also includes: fine-tuning the lightweight big model based on a second preset prompt word set to obtain the pre-trained hierarchical sub-model; wherein the lightweight big model is a pruned big model or a big model after knowledge distillation.
[0079] Among them, the second preset prompt word set includes historical security event information, and the security event information includes the network environment involved in the security event, the type of security event, the priority of the security event, the location where the security event occurred, the way the security event occurred, and the impact caused by the security event.
[0080] For example, taking drone GPS spoofing attack as an example, one of the preset prompt words in the second preset prompt word set may be:
[0081] “You are an expert in integrated air-space-ground cyber with expertise in this field.
[0082] The following is a description of the drone network environment. Drones currently communicate with ground control stations and other drones via wireless signals, and the network topology is highly complex, including dynamic connections between multiple drones and one or more ground control stations. Drones are transmitting real-time flight data, video streams, and sensor data to the ground control station, which is encrypted to ensure its security. The ground control station or edge computing node on the drone performs data analysis to support mission execution and decision-making.
[0083] The following is a description of the characteristics of a GPS spoofing attack on a drone.
[0084] Drone ID: UAV-5874
[0085] Category: GPS spoofing attack drones
[0086] Priority: High
[0087] Original flight path: 34.0522°N, 118.2437°W
[0088] Interference location: 34.0000°N, 118.0000°E
[0089] Jamming method: Faking GPS signals
[0090] Affected Systems: Navigation
[0091] Description: A drone GPS spoofing event was detected, where the drone's flight path was altered using a fake GPS signal. The original coordinates were 34.0522°N, 118.2437°E, and were changed to 34.0000°N, 1180000°E. This interference caused a deviation from the designated route, with the potential for a collision.
[0092] Please formulate a defense strategy against this attack based on the above information."
[0093] To facilitate understanding, the training process of the hierarchical sub-model can be explained by taking the response to drone GPS spoofing attacks as an example. First, preset prompt words are given. The preset prompt words need to accurately define the specific problem scenarios of drone GPS spoofing, including the attack methods, common attack patterns, and specific needs of drone networks. Collect relevant data, such as historical attack cases, drone flight data, GPS signal logs, etc. Then select a suitable pre-trained large language model as the basis, such as Llama, and use the collected professional data to fine-tune the model to better understand the specific scenarios and technical details related to GPS spoofing. The key to this step is to transform the model's general language understanding capabilities into specific field applications. Implement RAG technology to enhance the model's response capabilities. By combining retrieved external knowledge, the model can access the latest research and case studies to provide more accurate solution strategies and defense recommendations. In this process, the model not only generates answers, but also retrieves information from a large number of relevant documents to support the content it generates. Deploy the fine-tuned model as a decision support system for the drone network. The model can analyze incoming flight data and GPS signals in real time, identify potential GPS spoofing activities, and provide response strategies. This includes emergency measures such as dynamically adjusting flight paths and switching to backup navigation systems. Continuously collect the latest attack data and drone operation feedback, and continuously optimize and update the model. Ensure that the model can adapt to new attack techniques and strategies to maintain the timeliness and effectiveness of defense measures.
[0094] It can be understood that deploying hierarchical sub-models in various scenarios can enable the big model to obtain richer and more diverse data information, better utilize the data in specific scenarios, and through in-context learning, write the data and information of various communication scenarios as prompts and input them into the big model, continuously deepen the big model's understanding of the scenario, strengthen the capture of its features and regularities, and improve its ability to handle problems in specific scenarios.
[0095] It can also be understood that the hierarchical sub-models of the present invention are deployed on edge servers, and the performance of edge servers is poor, so it is necessary to perform lightweight operations on these large models, and lightweight operations are aimed at optimizing the model structure, compressing the model volume, and improving the computing efficiency without significantly sacrificing performance. There are various methods to achieve lightweighting, including pruning technology based on removing redundant neurons or connections to streamline the model structure; knowledge distillation technology that transfers the knowledge of large models to small models; and quantization technology that reduces the size of the model by reducing the precision of numerical representation.
[0096] By adopting the above method, the hierarchical sub-model can quickly and effectively learn the specific requirements of a specific scenario while maintaining its original knowledge.
[0097] In one embodiment, if Figure 5 As shown, the present invention also provides a method for determining a 6G air-space-ground network security strategy based on AI, which is applied to a terminal device or a node, including:
[0098] Step 510, receiving a task instruction from the cloud corresponding to the ground core network.
[0099] Step 520: Based on the task instruction and at least one professional small model, obtain a set of security events corresponding to the task instruction.
[0100] Specifically, after receiving the task instruction, the terminal device can monitor the relevant network traffic, user behavior and system status based on the task instruction, and input the monitored data into the corresponding professional small model respectively, so as to obtain a security event.
[0101] Step 530: Send the security event set to at least one edge server in the ground-based network, the air-based network, and the space-based network. The at least one edge server is used to execute the AI-based 6G air-space-ground network security policy determination method corresponding to any of the above embodiments.
[0102] Step 540, receive and execute the security sub-policy in the security sub-policy set sent by the at least one edge server, or receive and execute the cross-layer security decision sent by the cloud corresponding to the ground core network; the cloud corresponding to the ground core network is used to execute the AI-based 6G air-space-ground network security policy determination method corresponding to any of the above embodiments.
[0103] It can be understood that since this method is similar to the above Figures 2 to 4 The methods in can correspond, so for the sake of brevity, similar content is not repeated here.
[0104] Among them, the at least one professional small model may include a threat discovery and alarm small model, a vulnerability discovery and repair small model, a fault monitoring and repair small model, and an intrusion detection and interception small model.
[0105] It can be understood that the professional small model can be an existing small model with good performance. Different professional small models are obtained based on data pre-training for different scenarios. The present invention does not limit the professional small model. Exemplarily, the powerful sequence data processing capability of the long short-term memory network (LSTM) model can be used to efficiently analyze complex time series patterns in network traffic and monitor complex network intrusion behaviors in real time.
[0106] The following describes the AI-based 6G air-space-ground network security strategy determination system provided by the present invention. Figure 6 As shown, in one embodiment, a system for determining a 6G air-space-ground network security strategy based on AI is provided. The system for determining a 6G air-space-ground network security strategy based on AI may include: a network security cross-layer general large model, at least one hierarchical sub-model and at least one professional small model.
[0107] Among them, the at least one professional small model is deployed on the terminal device and the node, and is used to realize real-time monitoring of network traffic, user behavior and system status, and discover security incidents; the at least one professional small model is also used to respond to security sub-strategies and cross-layer security decisions; the terminal device and the node are used to execute the AI-based 6G air-space-ground network security policy determination method corresponding to any of the above-mentioned embodiments; the at least one hierarchical sub-model is distributedly deployed on at least one edge server corresponding to each layer of the ground-based network, the air-based network and the space-based network; the at least one hierarchical sub-model is used to generate a security sub-strategy; the at least one edge server is used to execute the AI-based 6G air-space-ground network security policy determination method corresponding to any of the above-mentioned embodiments.
[0108] The network security cross-layer general large model is deployed in the cloud corresponding to the ground core network, and is used to generate cross-layer security decisions for ground-based networks, air-based networks and space-based networks based on the security events and the security sub-strategies; the cloud corresponding to the ground core network is used to execute the AI-based 6G air-space-ground network security strategy determination method corresponding to any of the above-mentioned embodiments.
[0109] The AI-based 6G air-space-ground network security strategy determination system provided by the present invention realizes a multi-layer intelligent collaboration mechanism among the cross-layer general large model, hierarchical sub-model, and professional small model of the air-space-ground integrated network security by deploying corresponding models in a layered manner in the terminal device, at least one edge server in the ground-based network, the air-based network, or the space-based network, and the cloud corresponding to the ground core network, thereby integrating intelligent perception, intelligent analysis, intelligent decision-making, and intelligent orchestration into every link of the security system and realizing the architectural-level intelligent endogenous generation of network security strategies.
[0110] It should be noted that the present invention integrates the virtualization technology of 5G network and the software defined network (SDN) technology to provide integrated integration and abstract processing for the security functions of the air-ground integrated network. Through virtualization technology, security functions are abstracted and centrally constructed into a unified resource pool to achieve on-demand customization, dynamic deployment and elastic scaling of security capabilities, and adapt to the security needs of cloud networks. The model in the present invention can score the current environment through real-time detection and reported environmental information, and generate corresponding security capability orchestration strategies in combination with specific attack information. Specifically, for example, when a DDoS attack is detected in the current environment, the model evaluates that the current environment score is low, then the system will call firewalls and other protective capabilities from the security capability pool, and customize and dynamically deploy specific firewall rules on demand. In addition, the system can elastically scale and dynamically adjust the firewall filtering scale to achieve efficient security protection. Based on the system provided by the present invention, the air-ground integrated network can dynamically call appropriate security capabilities according to the current security policy, effectively implement security protection measures, and ensure the security and reliability of the network.
[0111] The present invention uses a state perception model (professional sub-model) to collect relevant information, build a multi-level model architecture, and arrange security capabilities according to the division of labor at different levels. The architecture can generate corresponding security policies based on the collected information, and feed these policies back to the terminal for execution to implement effective security protection. This solution can ensure that the security of the network is fully protected, while improving the pertinence and effectiveness of security protection measures.
[0112] Among them, security capabilities include: 1. Intrusion detection system: Real-time detection and alert of potential malicious behavior or policy violations by monitoring and analyzing network activities. 2. Unified authentication mechanism for integrated air-ground-space network: Adopt multi-level security measures to ensure that only authorized users can access and operate network resources. 3. Abnormal traffic monitoring module: Identify and record abnormal traffic patterns in the network to help prevent and mitigate DDoS attacks and other security threats. 4. Air interface security mechanism based on physical security isolation and channel keys: Combine physical isolation and encryption technology to ensure the security and integrity of data transmission. 5. Adopt blockchain, differential privacy, federated learning, and establish trust models to further ensure the protection of user privacy and data in the domain of the integrated air-ground-space network architecture. First, blockchain technology is used to record and verify data transmission to ensure the integrity and tamper-proofness of data. Secondly, when processing user sensitive data, differential privacy technology protects individual privacy and prevents sensitive information leakage by adding noise during data analysis and sharing. In addition, federated learning technology allows multiple distributed nodes to jointly train models without sharing original data, thereby reducing the risk of data leakage. Finally, the present invention establishes a trust model in the architecture to evaluate and verify the reliability and security of the equipment and environment to ensure the security and stability of communications.
[0113] It can be understood that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A method for determining a 6G air-space-ground network security strategy based on AI, characterized in that: Applied to the cloud corresponding to the ground core network, including: In response to the task instruction, the task instruction is sent to a corresponding terminal device or node through at least one edge server in a ground-based network, an air-based network, or a space-based network; Receive a security event set and a security sub-strategy set corresponding to the task instruction; wherein the security event set is determined based on at least one professional small model, and the security sub-strategy set is determined based on at least one hierarchical sub-model; the at least one professional small model is respectively deployed on a corresponding terminal device or node; the at least one hierarchical sub-model is distributedly deployed on edge servers corresponding to each layer of the ground-based network, the air-based network, and the space-based network; The security event set and the security sub-strategy set are input into a pre-trained network security cross-layer general large model to generate cross-layer security decisions; the cross-layer security decisions are used to uniformly schedule and orchestrate the network security capabilities of ground-based networks, air-based networks, and space-based networks.
2. The method for determining the 6G air-space-ground network security strategy based on AI as claimed in claim 1, characterized in that: The step of sending the task instruction to a corresponding terminal device or node through at least one edge server in a ground-based network, an air-based network, or a space-based network in response to the task instruction includes: In response to a task instruction issued by a user or a state-aware task instruction of a network device, the task instruction is issued to a corresponding terminal device or node through at least one edge server in a ground-based network, an air-based network or a space-based network.
3. The method for determining the 6G air-space-ground network security strategy based on AI as claimed in claim 1, characterized in that: The pre-trained network security cross-layer universal large model is a large language model. Before inputting the security event set and the security sub-strategy set into the pre-trained network security cross-layer universal large model to generate a cross-layer security decision, the method further includes: Based on the first preset prompt word set, fine-tune the large language model to obtain the pre-trained network security cross-layer universal large model; The first preset prompt word set includes a historical security event set and a security sub-policy set determined based on the historical security event set.
4. The method for determining the 6G air-space-ground network security strategy based on AI as claimed in claim 1, characterized in that: The at least one hierarchical sub-model includes at least one ground-based network sub-model, at least one air-based network sub-model, and at least one space-based network sub-model, and the edge server includes an edge service satellite, an edge service drone, and a ground edge server; The at least one hierarchical sub-model is distributedly deployed on edge servers corresponding to each layer of the ground-based network, the air-based network, and the space-based network, including: The at least one ground-based network sub-model is distributedly deployed on a ground edge server; The at least one air-based network sub-model is distributedly deployed on the edge service drone; The at least one space-based network sub-model is distributedly deployed on the edge service satellite.
5. A method for determining 6G air-space-ground network security strategy based on AI, characterized in that: The edge servers corresponding to the ground-based network, air-based network or space-based network include: Receive task instructions from the cloud corresponding to the ground core network, and send the task instructions to the corresponding terminal device or node; Receiving a security event set corresponding to the task instruction; the security event set is determined based on at least one professional small model, and the at least one professional small model is respectively deployed on a corresponding terminal device or node; Determine a security sub-policy set based on the security event set, at least one pre-trained hierarchical sub-model, and a preset security policy database; the security sub-policy set is used to uniformly schedule and orchestrate the security capabilities of the distributed subnet where the edge server is located; The security event set and the security sub-policy set are sent to the cloud corresponding to the ground core network; the cloud corresponding to the ground core network is used to execute the method described in any one of claims 1 to 4 above.
6. The method for determining the 6G air-space-ground network security strategy based on AI as claimed in claim 5, characterized in that: The pre-trained hierarchical sub-model includes a retrieval enhancement generation module and an autonomous organization generation module, and the security sub-policy set is determined based on the security event set, at least one pre-trained hierarchical sub-model, and a preset security policy database, including: Retrieve the security policies corresponding to the respective security events in the security event set from the preset security policy database through the retrieval enhancement generation module; For each security event, when it is determined that the preset security policy database includes a security policy corresponding to the security event, the security policy corresponding to the security event is determined as a security sub-policy of the security event; or, For each security event, when it is determined that the preset security policy database does not include the security policy corresponding to the security event, the security event is input into the corresponding autonomous organization generation module, and the autonomous organization generation module generates a security sub-policy for the security event.
7. The method for determining the 6G air-space-ground network security strategy based on AI as claimed in claim 5, characterized in that: Before determining the security sub-policy set based on the security event set, the pre-trained hierarchical sub-model, and the preset security policy database, the method further includes: Based on the second preset prompt word set, fine-tune the lightweight large model to obtain the pre-trained hierarchical sub-model; wherein the lightweight large model is a pruned large model or a knowledge distilled large model; Among them, the second preset prompt word set includes historical security event information, and the security event information includes the network environment involved in the security event, the type of security event, the priority of the security event, the location where the security event occurred, the way the security event occurred, and the impact caused by the security event.
8. A method for determining 6G air-space-ground network security strategy based on AI, characterized in that: Applied to terminal devices or nodes, including: Receive mission instructions from the cloud corresponding to the ground core network; Based on the task instruction and at least one professional small model, obtaining a security event set corresponding to the task instruction; Sending the security event set to at least one edge server in a ground-based network, an air-based network, or a space-based network, wherein the at least one edge server is used to execute the method described in any one of claims 5 to 7 above; Receive and execute the security sub-policies in the security sub-policy set sent by the at least one edge server, or receive and execute the cross-layer security decisions sent by the cloud corresponding to the ground core network; the cloud corresponding to the ground core network is used to execute the method described in any one of claims 1 to 4 above.
9. The method for determining the 6G air-space-ground network security strategy based on AI as claimed in claim 8, characterized in that: The at least one professional small model includes a threat discovery and alarm small model, a vulnerability discovery and repair small model, a fault monitoring and repair small model, and an intrusion detection and interception small model.
10. A system for determining 6G air-space-ground network security strategy based on AI, characterized in that: The system comprises: A network security cross-layer general large model, at least one hierarchical sub-model and at least one specialized small model; Wherein, the at least one professional small model is deployed on the terminal device and the node, and is used to realize real-time monitoring of network traffic, user behavior and system status, and discover security events; the at least one professional small model is also used to respond to security sub-strategies and cross-layer security decisions; the terminal device and the node are used to execute the method of any one of claims 8-9 above; The at least one hierarchical sub-model is distributedly deployed on at least one edge server corresponding to each layer of the ground-based network, the air-based network and the space-based network; the at least one hierarchical sub-model is used to generate a security sub-strategy; the at least one edge server is used to execute the method described in any one of claims 5 to 7 above; The cross-layer general large model of network security is deployed in the cloud corresponding to the ground core network, and is used to generate cross-layer security decisions for ground-based networks, air-based networks and space-based networks based on the security events and the security sub-strategies; the cloud corresponding to the ground core network is used to execute the method described in any of claims 1 to 4 above.
Citation Information
Cited By
Distributed intelligent air and space defense combat adaptive system architecture
CN119620626A
Security management against vulnerabilities in machine learning models in a communication network environment
WO2026201426A1
