A method and system for network routing security protection of unmanned aerial vehicles (UAVs)

Abstract

This application discloses a method and system for protecting unmanned aerial vehicle (UAV) network routing security, involving UAVs and deep learning technology. The method includes: collecting UAV network information data and performing data preprocessing; estimating UAV network routing traffic based on time-series data statistical adaptive regression using the traffic data obtained from the data preprocessing; calculating a network security factor based on a sequence nonlinear fitting learning method using the traffic estimation results; and optimizing and generating routes in the UAV swarm network using a multi-level dynamic data reconstruction-gradient strategy generation learning model based on the network traffic estimation results and the calculated network security factor, thereby achieving security protection for the UAV swarm network. The method of this application can solve the problems of insufficient data samples and untimely security policy response in traditional methods.

Description

Technical Field

[0001] This application relates to the fields of unmanned aerial vehicles (UAVs) and deep learning technology, and in particular to a method and system for protecting network routing security for UAVs. Background Technology

[0002] With the rapid development of drone technology, the security of drone networks has attracted increasing attention. Drone network security refers to a comprehensive security system that protects drone systems, including aircraft, ground control stations, communication links, and data, from cyberattacks, unauthorized access, data leaks, or malicious manipulation. Its core objective is to ensure the confidentiality, integrity, availability, and controllability of drones during operation. Drone network security is crucial for ensuring the safe operation of drone systems in complex digital environments, and its applications involve multiple aspects such as public safety, military security, and personal privacy. Drone network security can safeguard social order, reducing the threat to public safety posed by terrorist organizations using drones. Regarding personal privacy, the communication security of drone network devices is directly related to privacy leaks, and intrusions into weakly secure devices within the drone network are a significant source of information theft. Identity authentication and security mechanisms designed to address privacy leaks can effectively reduce the probability of data breaches and protect personal privacy and public safety. Therefore, drone network security has become a new research hotspot. Secure protection of drone network routing can resist attacks, reduce the probability of drone communication disruptions, and improve the robustness, integrity, and real-time performance of drone networks.

[0003] Currently, the relevant technologies for drone cybersecurity methods include:

[0004] "A Routing Security Protection Method and System for Unmanned Aerial Vehicle (UAV) Ad Hoc Networks" (Application No.: CN202110851751.4, Applicant: Xi'an University of Electronic Science and Technology, Inventors: Guo Jingjing; Gao Huamin; Liu Zhiquan; Huang Feiran; Wei Linfeng). The main idea of ​​this method is to divide network node messages into dynamic and static message fields, sign the messages before sending them, verify their integrity upon receipt, and update the topology based on the verification results. This method is a key-based routing security protection method, which updates routing messages relatively quickly. However, this method mainly verifies message integrity, and its effectiveness is weakened when malicious nodes use methods such as black hole intrusion or Sybil intrusion.

[0005] "A Relay Network Topology Generation Method for UAV Swarm Applications" (Application No.: CN202110851751.4, Applicant: University of Electronic Science and Technology of China, Inventors: Zhang Ke; Ma Yuhang; Li Jinke; Leng Supeng). This method mainly relies on adaptive adjustments based on network application layer control information to ensure effective coverage of mission machines and secure communication with the central control unit, as well as guaranteeing the performance and reliability of the UAV swarm network during mission execution. However, this method only analyzes network application layer control messages and does not generate network layer routing security policies. It does not employ a gradient-based network security analysis method and requires a significant amount of prior information, including location information and traffic volume. Furthermore, its security protection effect against network intrusion is unsatisfactory in complex environments; it cannot adaptively optimize and adjust in real time to derive network security protection strategies when the network topology changes dynamically; and it can only provide a certain degree of security protection against existing network intrusion methods, lacking the ability to defend against new network intrusion techniques. Summary of the Invention

[0006] This application provides a method and system for protecting network routing security for unmanned aerial vehicles (UAVs), which addresses the problems of insufficient data samples and untimely response of security policies in traditional methods.

[0007] This application provides a method for network routing security protection of unmanned aerial vehicles (UAVs), including:

[0008] Collect UAV network information data and perform data preprocessing;

[0009] Based on the traffic data obtained from data preprocessing, the routing traffic of the UAV network is estimated using a time-series data statistical adaptive regression method.

[0010] The network security factor is calculated based on the traffic estimation results using a sequence nonlinear fitting learning method.

[0011] Based on network traffic estimation results and calculated network security factors, a multi-level dynamic data reconstruction-gradient strategy is adopted to generate a learning model to optimize and calculate routes in the UAV swarm network, thereby achieving security protection for the UAV swarm network.

[0012] This application also proposes a UAV network routing security protection system, including a processor and a memory. The memory stores a computer program, which, when executed by the processor, implements the steps of the aforementioned UAV network routing security protection method.

[0013] The method in this application employs a multi-level dynamic data reconstruction model, which exhibits good iterative processing performance for data in complex environments and can better approximate complex network environment data. By extracting reliable network training samples through adaptive parameter caching, the accuracy of loss function calculation and convergence speed of the routing gradient strategy generation evaluation learning model network are improved, thus solving the problems of insufficient data samples and untimely response of security policies in traditional methods.

[0014] The above description is only an overview of the technical solution of this application. In order to better understand the technical means of this application and to implement it in accordance with the contents of the specification, and to make the above and other objects, features and advantages of this application more obvious and understandable, the following are specific embodiments of this application. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0016] Figure 1 This is a schematic diagram of the overall process of the UAV network routing security protection method in the embodiments of this application;

[0017] Figure 2 This is a schematic diagram of the multi-level dynamic data reconstruction model and gradient strategy generation learning model architecture in an embodiment of this application;

[0018] Figure 3 This is a simulation example of UAV network routing in the embodiments of this application;

[0019] Figure 4 This is a schematic diagram of a secure routing transmission path updated based on a deep learning joint model, as shown in the simulation example of the embodiments of this application.

[0020] Figure 5 The training error curve of the multi-level dynamic data reconstruction and gradient strategy generation learning joint model in the simulation example of the embodiment of this application;

[0021] Figure 6a , 6b The transmission latency performance of the method of this application in generating network security routes is demonstrated in the simulation example of a network intrusion scenario in the embodiments of this application.

[0022] Figure 7a , 7b The packet loss rate performance of the method for generating network security routes in this application is demonstrated in the simulation example of a network intrusion scenario in the embodiments of this application.

[0023] Figure 8a , 8b The throughput before and after generating a secure route using the method described in this application is shown in the embodiments of this application. Detailed Implementation

[0024] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0025] This application provides a method for network routing security protection of unmanned aerial vehicles (UAVs), such as... Figure 1 As shown, it includes the following steps:

[0026] In step S101, UAV network information data is collected and preprocessed. Specifically, UAV network information is collected by parsing UAV network protocol messages to obtain data related to network routing information, including the routing table of the data packets, node location information, routing protocol control information, node status information, and other dynamic routing-related extended information. Through active route probing and tracing technology, customized network routing information messages are sent, data packets with different lifespans are generated, triggering a routing timeout response mechanism and receiving response data packets, thereby completing the collection of routing information for subsequent network data preprocessing.

[0027] After completing the network information data collection, the data undergoes preprocessing, including data cleaning, stationarity testing, and data augmentation. First, the data is cleaned to handle missing values, outliers, and duplicate data. Missing values ​​are handled through deletion, imputation, or interpolation, while outliers are identified and handled using statistical methods or prior knowledge of UAV networks. Duplicate data is either deleted or integrated. This cleaning and preprocessing yields a network routing information dataset, providing an information source and data samples for subsequent network routing traffic estimation.

[0028] Because UAV network nodes operate in a dynamically changing environment, the routing tables, node location information, and control message fields in the acquired network information data also change over time. Therefore, this network information data is treated as time series data, and the moving mean and moving variance parameters of the series are calculated to perform a time-domain stationarity test. If the test results show that the information series data is non-stationary, time-domain filtering methods are used to preprocess the data to obtain stationary information series data, reducing data fluctuations and increasing the accuracy of the estimation.

[0029] Converting indicator data to a single dimension is specifically reflected in indicator normalization, which transforms all indicator values ​​into dimensionless values ​​between 0 and 1, thereby eliminating the influence of values ​​and dimensions on the calculation.

[0030] Then, information enhancement is achieved by constructing a noise suppression model, and statistical features are used to locally correct the time-series data. A sliding window mechanism is adopted to calculate statistical quantities such as the sample mean within the window, and the parameters are dynamically adjusted based on the local characteristics of the data and the obtained parameter values ​​to achieve spatial adaptive analysis of noise intensity constraints, effectively eliminating short-term interference. The normalized and filtered preprocessed UAV network routing information data is input into the next model for network traffic estimation.

[0031] In step S102, based on the traffic data obtained from data preprocessing, the UAV network routing traffic is estimated using a time-series data statistical adaptive regression method. Specifically, this includes:

[0032] The processing order is adaptively determined based on the highest priority probability information criterion. The information criterion function is defined according to the sample data variance, sample size and order, and the order parameter is determined.

[0033] First, the order parameter is determined, which is the time scale of the data used in the estimation method. The processing order is adaptively determined based on the highest priority probability information criterion. An information criterion function is defined according to the sample data variance, sample size, and order, and this function is optimized to determine the order parameter.

[0034] After setting the processing order, a time-series data statistical adaptive regression method based on autoregressive coefficients is constructed. The traffic estimation result is obtained by linearly combining network routing information data with traffic data from past times, and the parameter bias is corrected by using the optimized estimated regression coefficients.

[0035] Then, based on the joint parameter estimation probability maximization method, the processing parameters such as the regression coefficients are estimated. The processing parameters are adaptively optimized by maximizing the estimated probability function under the existing data samples, as shown in the following expression:

[0036]

[0037] in, To estimate the probability function for joint parameters, These are the autoregressive coefficients. The regression coefficients are the information data.

[0038] After optimizing the statistical processing parameters, the obtained method is used to analyze and calculate the network routing information data to obtain real-time estimates of network routing traffic. As shown in the following formula:

[0039]

[0040] in, Indicates the current time The output estimation results, This represents the estimation result from the previous moment. For flow rate estimates of order difference, For the input drone network information data, This indicates the processing order set. The obtained network traffic estimates are used as input layer elements for subsequent sequence nonlinear fitting learning methods to further analyze the network security factors of the drone swarm.

[0041] In step S103, the network security factor is calculated based on the traffic estimation results using a sequence nonlinear fitting learning method.

[0042] In step S104, based on the network traffic estimation results and the calculated network security factors, a multi-level dynamic data reconstruction-gradient strategy is used to generate a learning model to optimize and generate routes in the UAV swarm network, so as to achieve security protection for the UAV swarm network.

[0043] The method in this application employs a multi-level dynamic data reconstruction model, which exhibits good iterative processing performance for data in complex environments and can better approximate complex network environment data. By extracting reliable network training samples through adaptive parameter caching, the accuracy of loss function calculation and convergence speed of the routing gradient strategy generation evaluation learning model network are improved, thus solving the problems of insufficient data samples and untimely response of security policies in traditional methods.

[0044] In some embodiments, calculating the network security factor based on a sequence nonlinear fitting learning method according to the traffic estimation results includes:

[0045] To calculate the reliability of data transmission from network nodes, we first compare the routing data sent by the nodes at different times based on the routing data packets they send, and then calculate the measured value of the transmitted data from the sending node.

[0046]

[0047] in Indicates the length of the routed data packet transmitted by the node. This represents the maximum threshold for processing data packets per unit of time. At any given time, the sending node Transmission measurement values ​​are used express. For nodes The neighboring nodes, The neighbor node number. This represents the radius threshold for selecting neighboring nodes.

[0048] Measurement sending node The transmission measurement values ​​are used to calculate the difference between transmission measurement values ​​at different times to determine whether a node is sending duplicate transmissions, and to calculate the node's transmission measurement values. The network security factor is obtained by taking the average measurement difference of all neighboring nodes. :

[0049]

[0050] in, For nodes The number of neighboring nodes, For nodes At a radius of All neighboring nodes within the range, This represents the difference between the node's measurement and the transmitted data.

[0051] After the security factor of the node is calculated, the security factor of the node is used as a network security factor sample. The input sequence nonlinear fitting learning method is used to optimize the linear operation unit with nonlinear function and process the input network traffic data layer by layer.

[0052] After obtaining the output, the obtained output results are compared with the safety factor data obtained from the previous analysis, the loss function is calculated and the processing parameters are updated.

[0053]

[0054] in, To update the preprocessing parameters, This indicates the updated processing parameter value. This represents a non-linear processing function. By processing parameters at different levels and through joint analysis, features can be effectively refined to further obtain signal features. The learnability of the processed parameters also makes it a primary target for training and optimization. The number of levels in nonlinear fitting learning methods. For loss function, For gradient calculation, Network traffic data input as a nonlinear fitting learning method For the nonlinear operation functions of each layer, For compound operations, This is a sample of network security factors.

[0055] The loss function is calculated based on the data samples, and the processing parameters are adjusted according to the loss function to gradually achieve the effect of minimizing the loss, i.e., parameter convergence. The optimized sequence nonlinear fitting learning method is then obtained, which is used to extract features from the network routing traffic of UAV swarms and output network security factor values, providing analytical data for subsequent network security route generation calculations.

[0056] In some embodiments, based on network traffic estimation results and calculated network security factors, a multi-level dynamic data reconstruction-gradient strategy is used to generate a learning model to optimize and calculate routes in the UAV swarm network, thereby achieving security protection for the UAV swarm network. Specifically, this includes:

[0057] Construct a multi-level dynamic data reconstruction model and a gradient-based strategy-based generative learning model, where, for example... Figure 2 As shown, the learning model includes: a policy generation module, a policy inference module, and an adaptive parameter caching module.

[0058] The training process of the learning model includes:

[0059] First, data samples such as network routing traffic and security factors are sampled from the adaptive parameter caching module.

[0060] Data reconstruction is achieved by performing multi-level optimization processing on the network routing traffic data and network node security factors of the input network nodes through a multi-level dynamic data reconstruction model.

[0061] After the reconstruction is completed, the input data is calculated based on the generation function of the generation module to obtain the network routing strategy.

[0062] The obtained strategy is then output to the strategy inference module to calculate the inference result of the current strategy, and the model parameters are updated according to the result for iterative training.

[0063] After training is complete, the parameters are copied to the policy generation submodule and inference generation submodule with the same structure and the parameters are frozen to obtain a gradient-based policy generation learning model for route generation calculation, thus achieving separation of model training and use.

[0064] Gradient-based deep learning models are dynamic decision-making models that gradually approach the optimal strategy through interaction and feedback with the environment. Unlike some traditional deep learning models, which rely on static datasets for offline training and cannot adapt to real-time environmental changes, gradient-based deep learning, based on Markov decision processes, models the temporal correlation of multi-step decisions. The model obtains feedback through real-time interaction with the environment and dynamically adjusts its strategy accordingly. Furthermore, gradient-based deep learning proactively discovers potentially high-return decision behaviors by designing exploration strategies, avoiding getting trapped in local optima and stabilizing returns.

[0065] Compared to traditional deep learning models such as LSTM networks, the gradient-based policy generation and learning model adjusts routing security protection strategies in real time according to changes in the network environment under complex conditions, ensuring the effectiveness of the strategies. Furthermore, the gradient-based policy generation and inference model can directly generate UAV network security routing protection strategies, eliminating the need for a network security status assessment model to evaluate the network security situation and adjust the UAV network routing strategy accordingly. The gradient-based policy generation and learning model combines policy generation and policy inference frameworks, separating the computation of training and usage processes through two sets of deep learning models, reducing mutual interference between modules and exhibiting better policy robustness. Therefore, the gradient-based policy generation and learning model proposed in this application can generate safer and more reliable UAV network security routing protection strategies, providing support for network security.

[0066] In some embodiments, the model training process further includes:

[0067] Based on the input network routing traffic data sample of the drone network nodes Security factors of network nodes The sample data is iteratively processed through a multi-level dynamic data reconstruction model to filter out noise and interference and reconstruct network traffic.

[0068]

[0069] in, , Sample data for network traffic and security factors. The model parameters are reconstructed for multi-level dynamic data and are affected by noise in the samples, so they can be updated through training; The weights in the iterative process can be estimated using variational posterior equations.

[0070] Through iterative analysis and optimization of the calculation equations of the dynamically reconstructed model, the flow and safety factor data were analyzed. , Disturbance elimination processing is performed to achieve dynamic descrambling of input network routing data;

[0071] Will pass The security factor data and traffic data reconstructed in the next iteration , The input gradient strategy generates the policy generation module for the learning model.

[0072] In some embodiments, it also includes:

[0073] The policy generation module and the policy reasoning module are two important components of the model. The reconstructed network traffic and security factor data are input into the policy generation module to calculate and generate corresponding routing policies. :

[0074]

[0075] in, This represents the normalized strategy output vector, i.e., selecting each node as a node. The probability of the next hop node. Represents a node The set of neighboring nodes, The policy generation module represents the input traffic and security factor. , The processing function, This indicates that its parameters can be updated during training. The value is the sequence number in which the drone sends routing information to its neighboring nodes, i.e. At that time, drone nodes to neighboring nodes Send routing information.

[0076] The policy generation module calculates the probability of selecting each neighbor node based on the input security state and chooses the neighbor node with the highest probability as the next-hop route, thus generating the network security routing policy result. Subsequently, the routing policy is input into the policy reasoning module, which calculates the reward function. Obtain the inference value of the strategy The system measures the effectiveness of drone nodes in executing corresponding routing strategies, calculates the loss function, and updates the model parameters.

[0077] The routing strategy is input into the policy inference module, and the reward function is calculated. Obtain the inference value of the strategy To determine the effectiveness of the drone nodes in implementing the corresponding routing strategy, the loss function is calculated and the model parameters are updated, as shown in the following formula:

[0078]

[0079] in Indicates the number of samples. For the corresponding strategy inference value, The output routing policy of the policy generation module, The parameter is The corresponding operational functions. For the strategy reasoning module, the first Parameters of each strategy reasoning submodule Parameters for the strategy generation module. To calculate the gradient, The weights are updated for the parameters. Furthermore, to enhance the sample dataset and improve model training speed, actions, states, and values ​​are stored in the parameter cache module and periodically sampled for calculating the network loss and updating the model parameters. The model parameters are iteratively updated until the model converges by calculating the loss function.

[0080] After updating the model parameters until the model converges, a learning model is generated based on multi-level dynamic data reconstruction and routing strategies, and the results of the safe routing strategy are output.

[0081] In some embodiments, after completing the construction and training analysis of the improved gradient-based policy generation learning model, based on the cooperative distribution optimization discrimination criterion, the multi-level dynamic data reconstruction model and the gradient-based policy generation learning model are jointly performed to achieve refined mining of data characteristics and improve the accuracy of weight allocation. The method further includes:

[0082] First, the network routing traffic data is input into the gradient strategy generation learning model trained in the previous step to obtain the corresponding routing output results. And,

[0083] Traffic data is input into a multi-level dynamic data reconstruction model, and the output vector is calculated and normalized. After obtaining the outputs of the two models, the weights of the two models are adjusted based on the collaborative distribution optimization criterion to reduce weight calculation errors and obtain a more accurate joint model output, as shown in the following equation:

[0084]

[0085] in, These represent the outputs of the gradient-based strategy-generated learning model and the multi-level dynamic data reconstruction model, respectively. This represents the output network security routing policy vector, which is the probability of selecting each node as the next-hop route. This represents the information distribution characteristics calculated through routing policy vectors. Solving for the average uncertainty of each value; As a balancing factor, it dynamically changes with the model loss during the training of both models. This formula represents the difference in information distribution between the probability vectors output by the two models. Based on this formula, the joint model is used to optimize the network security routing weights, which are then used to calculate the network security routing policy output and to analyze the performance of network security routing protection strategies.

[0086] In some embodiments, it also includes:

[0087] Based on the routing vector weights calculated using the cooperative distribution optimization criterion, a network security routing policy is output to protect against malicious node intrusions into the drone network. The routing policy output calculation satisfies the following:

[0088]

[0089] in, To output the routing policy probability vector, The weight values ​​for the multi-level dynamic data reconstruction model and the gradient strategy are respectively used to generate the learning model. These are the routing result vectors for the two models, respectively.

[0090] After calculating the joint routing vector, the vector is normalized, the probability values ​​are calculated, and the nodes with the highest probability are selected. This is the next-hop routing output of the routing policy.

[0091] First, information data from the UAV network is acquired and preprocessed, including data cleaning and normalization. This data is then input into a time-series data statistical adaptive regression model to estimate UAV network traffic. Based on the estimated values, a security factor is calculated using a sequence nonlinear fitting learning model. The network routing traffic estimates and security factor obtained in the above steps are used as input to a multi-level dynamic data reconstruction and gradient-based policy generation learning model. The model is optimized and reconstructed to obtain de-scratched network traffic data, which is then input into the policy generation module of the gradient-based policy generation learning model. The routing output is obtained based on the policy function. The resulting routing policy is input into the policy inference module. The reward value is calculated based on the sample data of network traffic and security factor, and the model parameters are updated for training. The routing policy output is then calculated based on the trained joint model.

[0092] This application further proposes simulation experiments to verify the method of this application. After generating the UAV network security routing strategy, in order to analyze the performance of the network security routing strategy, a certain number of malicious nodes are added to the UAV network to pose a security threat to the network. A multi-level dynamic data reconstruction and gradient strategy generation learning joint model is used to calculate the network security routing strategy. The network routing is protected according to the obtained security strategy, and the status and quantity of data packet sending and receiving in the network, as well as the network data packet sending and arrival time, are monitored and statistically analyzed in real time. The network packet loss rate, network transmission latency, network throughput, and other performance parameters are calculated. Compared with the case without network routing security strategy and the traditional method, the security performance effect of the routing strategy output by the new method in this paper can be analyzed. After the routing strategy is output based on the multi-level dynamic data reconstruction and gradient strategy generation learning joint model and the security performance effect is analyzed, the security routing performance of the model is verified through simulation experiments. The simulation scenario uses a UAV self-organizing network composed of 20 UAV nodes, the security threat method is black hole intrusion, the loss function is MSE loss function, and the maximum number of training iterations is set. Step size for parameter updates .

[0093] A simulation experiment was conducted based on network layer security threats to unmanned aerial vehicles (UAVs). The network routing diagram is shown below when no network security threats are present. Figure 3 As shown, the drone network nodes are blue, indicating that all nodes are in normal status and have not been compromised. Node 4 can send routing information to nodes 3 and 9. However, by using a black hole intrusion method to compromise node 4, the routing information cannot be transmitted normally after the network node 4 is compromised.

[0094] Secure routing transmission paths updated based on a joint model of multi-level dynamic data reconstruction and gradient-based learning, such as... Figure 4 As shown, network nodes 3 and 9 change their next-hop routes to avoid sending routing messages to the compromised network node 4, and instead set their respective next-hop routes to nodes 10 and 11.

[0095] The model was trained 100 times based on the set parameters, and the training error curve of the joint model of multi-level dynamic data reconstruction and gradient strategy generative learning was obtained as follows: Figure 5 As shown, it is compared with the traditional method, which analyzes and evaluates the link status and effective link lifetime, and selects the corresponding link for the routing strategy in real time based on the evaluation results.

[0096] Figure 5 The horizontal axis represents the number of training iterations for the joint model of multi-level dynamic data reconstruction and gradient-based strategy generative learning, while the vertical axis represents the training error. From Figure 5As can be seen from the above, the secure routing policy generation method proposed in this application has a faster convergence speed and better error performance compared to the Q-learning method.

[0097] Based on the secure routing policy generation method of this application, the changes in network transmission latency, packet loss rate, and throughput under the influence of black hole intrusion are statistically analyzed, and the effects are compared with those of not using the secure routing policy generation method and using the traditional network topology-aware routing policy generation method. The resulting curves of network transmission latency, packet loss rate, and throughput changes are shown in Figures 6, 7, and 8, respectively.

[0098] As shown in Figures 6, 7, and 8 above, the horizontal axis represents statistical time in seconds, and the vertical axis represents network transmission latency, network packet loss rate, and network throughput performance parameters. The transmission latency before generating a secure route using the method of this application is in seconds; the transmission latency after generating a secure route using the new method is in milliseconds (ms), and the network throughput is in Mbps.

[0099] Figure 6 illustrates the network transmission latency performance curves of network security routes generated using the method of this application after a network intrusion. As shown in Figure 6(a), the network transmission latency is approximately 9.8 seconds after the network intrusion. As shown in Figure 6(b), the network transmission latency of the network security routes generated by the method of this application, after protection, is approximately 90 milliseconds. In contrast, the traditional security route generation method results in a network transmission latency of approximately 290 milliseconds after protection. Therefore, the UAV swarm network security route generation calculation method based on a joint model of multi-level dynamic data reconstruction and gradient-based strategy generation learning can effectively protect UAV swarm networks.

[0100] Figure 7 shows the network packet loss rate performance curves after a network intrusion, using the method of this application to generate secure routes. As shown in Figure 7(a), after a network intrusion, the packet loss rate is approximately 88%. As shown in Figure 7(b), the network packet loss rate after protection using the method of this application to generate secure routes is approximately 6%. In contrast, the traditional secure route generation method results in a network packet loss rate of approximately 12% after protection. Therefore, the UAV swarm secure route generation calculation method based on a joint model of multi-level dynamic data reconstruction and gradient strategy generation learning can effectively protect UAV swarm networks.

[0101] Figure 8 shows the network throughput performance curves after a network intrusion, generated using the method of this application to create a secure route. As shown in Figure 8(a), after the network intrusion, the network throughput is approximately 4.9 Mbps. As shown in Figure 8(b), after protection, the network throughput of the secure route generated using the method of this application is approximately 184 Mbps. In contrast, the traditional secure route generation method results in a network throughput of approximately 145 Mbps after protection. Therefore, the UAV swarm secure route generation calculation method based on a joint model of multi-level dynamic data reconstruction and gradient strategy generation learning can effectively protect UAV swarm networks.

[0102] Compared with the prior art, the method of this application has the following advantages:

[0103] To address the shortcomings of traditional methods in UAV network security routing regarding the effectiveness and real-time performance of security strategies, this paper proposes a joint model for generating secure routes using multi-level dynamic data reconstruction and gradient-based strategy generation. Multi-level dynamic data reconstruction effectively reduces the impact of noise in complex environments on routing data acquisition and security strategy generation, resulting in superior noise resistance. The gradient-based strategy generation model outputs efficient and accurate routing strategies through in-depth processing and computation, based on precise analysis of acquired network traffic data, achieving better security protection. By efficiently acquiring dynamic changes in network data and iteratively updating the routing security strategy, the paper ensures good timeliness and avoids performance degradation caused by untimely updates.

[0104] The UAV network security routing protection method proposed in this application generates dynamic intelligent optimization routing security policies to ensure the robustness of message transmission at each network node, thereby improving the fine-grained access capability of network layer routing and the availability of UAV network communication delivery. As an important part of the overall security characteristics of the UAV network, it accurately generates network security policies, providing important support for the safe operation of UAV network systems in complex environments.

[0105] This application also proposes a UAV network routing security protection system, including a processor and a memory. The memory stores a computer program, which, when executed by the processor, implements the steps of the aforementioned UAV network routing security protection method.

[0106] It should be noted that, in the embodiments of this application, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0107] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0108] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0109] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims. All of these forms are within the protection scope of this application.

Claims

1. A method for securing routing in a UAV network, the method comprising: The method comprises the following steps: Collecting unmanned aerial vehicle network information data and performing data preprocessing; Based on the traffic data obtained by data preprocessing, the traffic of unmanned aerial vehicle network routing is estimated based on time series data statistical adaptive regression method; According to the traffic estimation result, the network security factor is calculated based on the sequence nonlinear fitting learning method; According to the network traffic estimation result and the calculated network security factor, a multi-level dynamic data reconstruction-gradient strategy learning model is generated to optimize the calculation of the routing in the unmanned aerial vehicle cluster network, so as to realize the security protection of the unmanned aerial vehicle cluster network; According to the traffic estimation result, the network security factor is calculated based on the sequence nonlinear fitting learning method, which comprises the following steps: According to the routing data packet sent by the network node, the routing data sent by the node at different time is compared, and the transmission data measurement value of the sending node is calculated: wherein represents the length of a routing packet transmitted by a node, represents the maximum threshold of packets that can be processed per unit of time, represents the transmission measurement value of a sending node at time , is a neighbor node of the node , is the neighbor node number, represents the radius threshold for selecting a neighbor node; Measuring transmission node , calculating the difference of transmission measurement values at different time, judging whether the node exists repeated transmission, and calculating the average measurement difference of all neighbor nodes of the node , obtaining network security factor : wherein, the number of neighbor nodes of a node the number of neighbor nodes of a node all neighbor nodes of a node all neighbor nodes of a node within a radius of all neighbor nodes of a node within a radius of denotes the difference of a node to the transmission data measure The security factor of the node is taken as the network security factor sample, which is input into the sequence nonlinear fitting learning method, and the linear operation unit in the nonlinear function is optimized, and the input network traffic data is processed layer by layer; The obtained output result is compared with the calculated network security factor data, the loss function is calculated and the processing parameters are updated; According to the loss function, the processing parameters are adjusted to obtain the sequence nonlinear fitting learning method after optimization of the parameters, so as to extract the routing traffic characteristics of the unmanned aerial vehicle cluster network and output the network security factor value; According to the network traffic estimation result and the calculated network security factor, a multi-level dynamic data reconstruction-gradient strategy learning model is generated to optimize the calculation of the routing in the unmanned aerial vehicle cluster network, which comprises the following steps: A multi-level dynamic data reconstruction model and a gradient strategy learning model are constructed, wherein the learning model comprises a strategy generation module, a strategy reasoning module and an adaptive parameter cache module; The training process of the learning model comprises the following steps: Selecting network routing traffic and security factor data samples from the adaptive parameter cache module; Through the multi-level dynamic data reconstruction model, the network routing traffic data and network node security factor of the input network node are optimized and reconstructed; After reconstruction, the input data is calculated based on the generation function of the strategy generation module to obtain the network routing strategy; The obtained strategy is output to the strategy reasoning module to calculate the reasoning result of the current strategy, and the model parameters are updated for iterative training; After training, the parameters are copied to the strategy generation submodule and reasoning generation submodule with the same structure to obtain the gradient strategy learning model for routing generation calculation. 2.The method of claim 1, wherein, Based on the time series data statistical adaptive regression method, the traffic of unmanned aerial vehicle network routing is estimated, which comprises the following steps: Adaptively determine the order based on the optimal prior probability information criterion, define the information criterion function based on the sample data variance, sample size and order, and determine the order parameter; The processing parameters are estimated by using the joint parameter estimation probability maximization method, and the processing parameters are adaptively optimized by maximizing the estimation probability function under the existing data samples; After completing the optimization of the statistical processing parameters, the network routing information data is analyzed and calculated to obtain the estimation result of the network routing traffic. 3.The method of claim 1, wherein, In the model training process, the following steps are also included: Network routing traffic data samples of input unmanned aerial vehicle network nodes With network node security factors Sample data is iteratively processed by a multi-level dynamic data reconstruction model to filter out noise and interference in the data and reconstruct network traffic: wherein, , is the sample data of network traffic and security factors, is the multi-level dynamic data reconstruction model parameters, is the weight coefficient in the iteration process; The cumulative analysis optimization of the dynamic reconfiguration model calculation equation is carried out on the flow and safety factor data , The disturbance elimination processing is carried out, and the dynamic disturbance elimination of the input network routing data is realized; The security factor data and the traffic data after the sub-iterative reconstruction are combined , The strategy generation module inputs the gradientization strategy to generate a learning model. 4.The method of claim 3, wherein, Also included inputting the reconstructed network traffic and the security factor data into a policy generation module to calculate and generate a corresponding routing policy : wherein, denotes a normalized policy output vector, denotes a set of neighbor nodes of a node , denotes a processing function of the policy generation module on input traffic and security factors , , denotes a value of a parameter thereof, is a sequence number of the UAV sending routing information to the neighbor node. The strategy generation module calculates the probability of selecting each neighbor node according to the input security state, and selects the neighbor node with the maximum value as the next hop routing; The routing strategy is input to a strategy inference module, and an inference value of the strategy is obtained by calculating a reward function The inference value of the strategy is obtained by calculating a reward function to judge the effect of the unmanned aerial vehicle node executing the corresponding routing strategy, calculate a loss function, and update model parameters After updating the model parameters to the model convergence, a multi-level dynamic data reconstruction and routing strategy generation learning model is generated based on the routing strategy result. 5.The method of claim 4, wherein, After completing the construction and training analysis of the improved gradient strategy generation learning model, it also includes: Input the network routing traffic data into the gradient strategy generation learning model obtained by training to obtain the corresponding routing output result; and, Input the traffic data into the multi-level dynamic data reconstruction model, calculate the output vector and perform normalization processing; According to the output of the two models, adjust the weights of the two models based on the collaborative distribution optimization criterion to reduce the weight calculation error. 6.The method of claim 5, wherein, It also includes: According to the routing vector weight calculated based on the collaborative distribution optimization criterion, output the network security routing strategy to achieve protection against malicious node intrusion in the unmanned aerial vehicle network, wherein the routing strategy output calculation satisfies: wherein, is an output routing strategy probability vector, are weight values of the multi-level dynamic data reconstruction model and the gradient-based strategy generation learning model, respectively, are routing result vectors of the two models, respectively.

7. A UAV network routing security protection system, characterized in that, A processor and a memory are included, the memory has a computer program stored thereon, and the computer program implements the steps of the unmanned aerial vehicle network routing security protection method according to any one of claims 1-6 when executed by the processor.

Images