Unmanned aerial vehicle network security verification effect evaluation method, system, device and medium

Abstract

The invention relates to the technical field of unmanned aerial vehicle network security, and discloses an unmanned aerial vehicle network security verification effect evaluation method, system and device, and a medium, and the method comprises the steps: carrying out the preprocessing of the index data related to the network security of an unmanned aerial vehicle, and constructing an index matrix; performing dimension reduction processing on the index matrix based on a nonlinear excitation learning model, and extracting weight information based on a spatial information dynamic path iteration model to obtain an evaluation threshold vector of unmanned aerial vehicle network security and a weight vector thereof; calculating a normalized comprehensive threshold based on the evaluation threshold vector and the weight vector thereof; and calculating an evaluation vector of each node of the unmanned aerial vehicle network, and comparing the evaluation vector with the normalized comprehensive threshold to obtain an unmanned aerial vehicle network security evaluation result. The method is suitable for a complex network protocol operation mechanism, more accurate and objective analysis is realized by fully combining a model hierarchical structure, a more reliable and comprehensive evaluation effect is achieved, and the method is closer to an actual application scene of an unmanned aerial vehicle network.

Description

Technical Field

[0001] The present invention relates to the field of drone network security technology, and in particular to a drone network security verification effect evaluation method, system, equipment and medium. Background Art

[0002] With the rapid development of drone technology and communication network technology, the security issues of drone networks have also received increasing attention. In practical applications, the drone network architecture is relatively complex. Traditional evaluation methods do not consider the mutual influence between the various levels of drone network protocols and security verification effect indicators, and the accuracy of network security verification evaluation is relatively poor. Traditional AHP (AHP: Analytic Hierarchy Process, Hierarchical Analysis Method) and ANP (ANP: Analytic Network Process, Network Analysis Method) and other ideas rely on a lot of prior information and are greatly affected by subjective aspects, and cannot fully reflect the objective evaluation effect. Therefore, the evaluation of drone network security verification effect needs to analyze the overall structural characteristics of drone networks on the basis of combining complex network theory with network security evaluation ideas, and gradually analyze the key factors such as networking transmission delay and network packet loss rate that are closely related to network security performance, so as to better adapt to the drone network security verification and evaluation capabilities in complex scenarios.

[0003] Most of the network protocols used in drone networks adopt a standardized overall processing architecture, such as Figure 1 As shown in the figure, the UAV network protocol has the characteristics of optimized routing topology, dynamic time slot allocation, fast access networking and adaptive resource adjustment. The network layer and MAC layer of the UAV network are important core components of the network.

[0004] Among them, the network layer of the UAV network is mainly responsible for network routing planning and management, routing information interaction, network routing addressing, etc., which can realize the optimization control of the network topology.

[0005] The MAC layer of the drone network mainly completes functions such as network time slot scheduling and link frame transmission. It can realize flexible and dynamic allocation of time slots for newly connected nodes and ensure accurate synchronization of processing time of each node in the network.

[0006] For the evaluation of the network security verification effect of drones, it is necessary to comprehensively consider the complex factors at all levels related to the network protocol processing mechanism, conduct research on the multi-level links of the drone network protocol processing process, and calculate the weight of the network decision threshold based on the availability and integrity of the drone network; use deep learning, model hierarchical analysis and other methods for fusion optimization, and further generate a more robust and accurate joint decision threshold, so as to obtain a more reasonable drone network security verification and evaluation result that is closer to the actual application scenario.

[0007] The following are the similar methods with patent applications retrieved so far: 1. "A method for network security assessment of unmanned cluster systems based on graph neural networks" (Application No.: CN202211438933.X). This invention provides a method for network security assessment of unmanned cluster systems based on graph neural networks. For the communication topology set of an unmanned cluster system composed of multiple unmanned cluster systems, the corresponding adjacency matrix set is obtained, and then the original adjacency matrix set is preprocessed. The adjacency matrix set and the feature matrix set are input into the graph neural network module for feature extraction to obtain the global representation vector of the graph. The output vector of the fully connected layer is input into the Softmax function for normalization processing, and the performance of the model is evaluated according to relevant indicators. This invention is an evaluation method based on network topology, which considers the correlation between unmanned cluster nodes and the communication status of nodes, but is not sensitive to changes in the graph structure. When the network communication topology is not completely transparent, or network nodes are interfered and fail, the evaluation method based on graph neural networks will have a large error. In addition, when the number of network nodes is large and the topology is complex, the memory and computing resource requirements are high.

[0008] 2. "A method for network security risk assessment for unmanned aerial vehicle systems" (Application No.: CN202410334451.2). This invention proposes a network security risk assessment method based on an improved vulnerability tree. By constructing a vulnerability tree model, two risk indices are calculated, and the vulnerability level of the unmanned aerial vehicle system is quantified through the risk index values. At the same time, by comparing the system risk index values after using various network security measures, the evaluation and ranking of security priorities are carried out. However, this method requires collecting data such as control loss, production loss, labor cost loss, equipment repair loss, software repair loss, and the probability of attack occurrence for various attack methods, has high requirements for prior knowledge, and the data is not easy to collect. Moreover, the accuracy of data collection for unmanned aerial vehicle security index data will greatly affect the evaluation accuracy. Summary of the Invention

[0009] Aiming at the complex network architecture of the unmanned aerial vehicle network, there are many influencing factors, and the traditional evaluation method does not consider the mutual influence between each layer of the network protocol and the attack effect indicators, and the evaluation accuracy of the evaluation means for the network security verification effect is relatively poor. Through the research on the influence relationship between the evaluation indicators and each layer of the network protocol, this invention proposes a method, system, device, and medium for evaluating the network security verification effect of unmanned aerial vehicles. Based on the non-linear incentive learning and spatial information dynamic path iteration model, it is more applicable to the complex network protocol operation mechanism, and fully combines the model hierarchical structure to achieve more accurate and objective analysis, and achieve a more reliable and comprehensive evaluation effect, which is closer to the actual application scenario of the unmanned aerial vehicle network.

[0010] The technical solution adopted by the present invention is as follows: A method for evaluating the security verification effect of an unmanned aerial vehicle network, comprising: Preprocessing the index data related to the network security of the unmanned aerial vehicle and constructing an index matrix; Performing dimensionality reduction processing on the index matrix based on a non-linear excitation learning model, and extracting weight information based on a spatial information dynamic path iteration model to obtain an evaluation threshold vector and its weight vector for the network security of the unmanned aerial vehicle; Based on the evaluation threshold vector and its weight vector, calculating a normalized comprehensive threshold; calculating an evaluation vector for each node of the unmanned aerial vehicle network, and comparing it with the normalized comprehensive threshold to obtain the evaluation result of the network security of the unmanned aerial vehicle.

[0011] Further, the preprocessing of the index data related to the network security of the unmanned aerial vehicle includes: For data anomalies or missing values, cleaning the data to handle missing values, outliers, and duplicate data; among them, missing values are processed by deletion, filling, or interpolation, outliers are identified and processed by statistical methods or prior knowledge of the unmanned aerial vehicle network, and duplicate data are deleted or integrated into the same data; For the problem of index differences, normalizing the index data, that is, converting all index values into dimensionless numerical values between 0 and 1, and the expression is as follows:

[0012] Wherein, is the th index value of the th node, is the number of network nodes, is the result after normalization preprocessing.

[0013] Further, the construction of the index matrix includes: Constructing an index matrix based on the preprocessed index data; Reconstructing the index matrix, and the reconstruction method includes a loop operation and a stacking operation; The loop operation includes regarding the input matrix as two index vectors and changing the order in turn, that is, placing the first element at the last position of the vector and shifting the remaining elements forward; The stacking operation includes combining all the shifted column vectors obtained by the loop operation row by row to obtain a new index matrix.

[0014] Further, the dimensionality reduction processing of the index matrix based on the non-linear excitation learning model includes: Performing a convolution operation based on a non-linear function through the convolution layer of the non-linear excitation learning model to obtain a dimensionality-reduced index matrix:

[0015] Among them, is the index matrix after dimensionality reduction, is the input matrix, is the dimension of the input matrix; and are continuous univariate functions; The model output is realized through the activation function. The mean square error between the activation function and the obtained sample values is calculated to obtain the model loss function, and the model parameters are updated based on gradient descent:

[0016] Among them, is the model parameter before update, is the model parameter after update, is the step size of parameter update, is the mean square error function, is the number of samples, is the weight value in the sample, is the weight output value obtained based on the non-linear excitation learning model.

[0017] Furthermore, the spatial information dynamic path iteration model extracts weight information to obtain the evaluation threshold vector and its weight vector for UAV network security, including: After performing convolution operations through the convolution layer of the non-linear excitation learning model, the convolved index matrix is output to the first layer, i.e., the subunit layer, of the spatial information dynamic path iteration model. It is dynamically calculated and output to the second layer, i.e., the parent unit layer, and then a vector containing weight information is obtained through calculation by the parent unit layer; Based on the weight information extracted by the spatial information dynamic path iteration model, further dimensionality reduction and extraction processing are performed through the fully connected layer of the linear non-linear excitation learning model to obtain the evaluation threshold weight vector for UAV network security :

[0018] Among them, is the evaluation threshold weight for multiple indicators.

[0019] Furthermore, in the spatial information dynamic path iteration model, the total input of the processing unit is the weighted sum of the path vectors:

[0020] Among them, is the input vector, is the path weight, is the subunit matrix; The parent unit processes the input vector using an activation function to calculate the output vector :

[0021] wherein, is the activation function, is the norm calculation; The path weight is dynamically updated according to the proximity between the output of the parent unit and each path vector extracted based on the spatial information:

[0022] wherein, and are dynamic factors, and through the real-time change of the sub-unit vector, the path weight for transmitting data to the upper-level processing unit is dynamically updated; is the total number of paths, is the exponential function; The loss function is obtained by quantifying the difference between the actual weight and the sample:

[0023] wherein, is the calculation error for any data sample, is the sample label value, k is the output data vector optimized through dynamic path, is the sample serial number, is the positive offset constant of the output vector, is the negative offset constant of the output vector, is the dynamic coefficient factor, is the total error function of all samples, is the total number of samples.

[0024] Furthermore, calculating the normalized comprehensive threshold based on the evaluation threshold vector and its weight vector includes:

[0025] wherein, is the normalized comprehensive threshold, is the evaluation threshold, is the evaluation threshold weight, is the normalized index vector; Calculating the evaluation vector of each node in the UAV network includes:

[0026] Among them, is the calculated value of the evaluation vector, is the value of the th group of indicators in the normalized index vector.

[0027] A UAV network security verification effect evaluation system includes: A preprocessing and index matrix construction module, configured to preprocess index data related to UAV network security and construct an index matrix; An evaluation threshold and weight vector calculation module, configured to perform dimensionality reduction processing on the index matrix based on a non-linear incentive learning model, and extract weight information based on a spatial information dynamic path iteration model to obtain an evaluation threshold vector and its weight vector for UAV network security; A UAV network security evaluation module, configured to calculate a normalized comprehensive threshold based on the evaluation threshold vector and its weight vector; calculate the evaluation vectors of each node in the UAV network, and compare them with the normalized comprehensive threshold to obtain the UAV network security evaluation result.

[0028] A computer device includes a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, the above-mentioned UAV network security verification effect evaluation method is implemented.

[0029] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned UAV network security verification effect evaluation method is implemented.

[0030] The beneficial effects of the present invention are as follows: (1) Since the UAV cluster has a complex network architecture and the traditional evaluation method does not consider the mutual influence between each layer of the network protocol and the security verification effect indicators, the accuracy of network security verification evaluation is relatively poor. The present invention conducts UAV network security verification effect evaluation based on non-linear incentive learning and spatial information dynamic path iteration models. Through the study of the influence relationship between network security verification evaluation indicators and each layer of the network protocol, the evaluation method can be more applicable to complex network protocol operation mechanisms, and fully combines the hierarchical structure of the network protocol model to achieve more accurate and objective analysis, and achieve a more reliable and comprehensive network security verification evaluation effect.

[0031] (2) Compared with traditional evaluation methods such as AHP and ANP, the fusion method based on non - linear incentive learning and spatial information dynamic path iteration model adopted by the present invention has stronger objectivity. By training the model through machine learning, the influence caused by subjective factors is reduced, making the evaluation less vulnerable to human interference and increasing the accuracy of the evaluation. Moreover, the present invention has low requirements for prior knowledge and has strong robustness in complex environments. The spatial information dynamic path iteration model proposed by the present invention can more fully reflect the connotative information of samples and also has good effects in the case of small - scale data sets. The calculation idea based on non - linear incentive learning can fit more complex non - linear relationships, enhance the calculation accuracy of the evaluation weight effect, and thus improve the accuracy of UAV network security evaluation. Description of the Drawings

[0032] Figure 1 It is a schematic diagram of the overall processing architecture and composition of the existing UAV network protocol.

[0033] Figure 2 It is one of the flowcharts of the UAV network security verification effect evaluation method in Embodiment 1 of the present invention.

[0034] Figure 3 It is the second flowchart of the UAV network security verification effect evaluation method in Embodiment 1 of the present invention.

[0035] Figure 4 It is a schematic diagram of the joint model structure in Embodiment 1 of the present invention.

[0036] Figure 5 It is a curve graph of the UAV network evaluation vector value result in Embodiment 1 of the present invention.

[0037] Figure 6 It is a performance curve graph of the UAV network security verification evaluation accuracy rate in Embodiment 1 of the present invention. Detailed Embodiments

[0038] In order to have a clearer understanding of the technical features, objectives, and effects of the present invention, the detailed embodiments of the present invention are now described. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention, that is, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0039] Embodiment 1 Since the UAV cluster has a complex network architecture and the traditional evaluation method does not consider the mutual influence between each layer of the network protocol and the security verification effect index, the accuracy of the network security verification evaluation is relatively poor.

[0040] Based on this, this embodiment provides a method for evaluating the security verification effect of a drone network, as Figure 2 shown, including: Preprocess the index data related to the drone network security and construct an index matrix; Perform dimensionality reduction processing on the index matrix based on a non-linear incentive learning model, and extract weight information based on a spatial information dynamic path iteration model to obtain an evaluation threshold vector and its weight vector for the drone network security; Based on the evaluation threshold vector and its weight vector, calculate the normalized comprehensive threshold; calculate the evaluation vectors of each node in the drone network and compare them with the normalized comprehensive threshold to obtain the drone network security evaluation result.

[0041] As Figure 3 shown, the method for evaluating the security verification effect of the drone network in this embodiment can be implemented by the following steps: Step 1. Index data preprocessing; Step 2. Construct a network security evaluation index matrix; Step 3. Dimensionality reduction processing of the index matrix based on a non-linear incentive learning model; Step 4. Extract weight information based on a spatial information dynamic path iteration model; Step 5. Calculate the security evaluation threshold weight vector based on the joint model; Step 6. Output the network security evaluation result.

[0042] Specifically, the above steps are described in detail as follows.

[0043] Step 1. Index data preprocessing The optimization of the drone network weight calculation is a complex combination composed of many indicators. To evaluate whether the network is secure, it ultimately comes down to the evaluation of the indicators. Therefore, the first step in establishing an evaluation model is to define the indicators.

[0044] Preferably, the index data used in this embodiment is network transmission delay, network packet loss rate, and network throughput, which are specifically described as follows.

[0045] The network transmission delay is the time required to send a data frame. The network transmission delay consists of the transmission delays of each node. For each node, the time from the start of sending the first bit of the data frame to the completion of sending the last bit of the frame is the transmission delay of that node.

[0046] The network packet loss rate is the ratio of the number of lost data packets in the network to the number of data groups sent.

[0047] The network throughput is the amount of data successfully transmitted per unit time, reflecting the actual available bandwidth.

[0048] As important indicators reflecting the security status of the UAV network, network transmission delay, network packet loss rate, and network throughput are indispensable in the optimization of UAV network weight calculation. It is of typical significance to use the transmission delay and packet loss rate of network nodes as the index basis for UAV network weight calculation optimization.

[0049] Since the UAV network security index data is obtained from the UAV network, there may be problems such as data anomalies or missing data caused by noise. Moreover, there are usually large differences among the UAV network security index data. For example, there are significant differences in the value range and dimension of the transmission delay and packet loss rate. Therefore, before evaluating the UAV network security indicators, it is necessary to perform preprocessing of the indicators to reduce the errors brought by the abnormal values of the indicators themselves and the differences of the indicators to the evaluation.

[0050] For data anomalies or missing data problems, data cleaning methods are usually used for processing. Data cleaning can handle missing values, outliers, and duplicate data. Missing values are processed by deletion, filling, or interpolation, outliers can be identified and processed through statistical methods or prior knowledge of the UAV network, and duplicate data is deleted or integrated into the same data.

[0051] For the processing of indicator differences, choosing to convert the indicator data into the same dimension and removing the dimension can also reduce the number of network trainings and accelerate network convergence. In this embodiment, the preprocessing of the evaluation indicators is mainly indicator normalization, that is, converting all indicator values into dimensionless numerical values between 0 and 1. The expression is as follows:

[0052] Among them, is the th indicator value of the th node, is the number of network nodes, is the result after normalization preprocessing.

[0053] After preprocessing, construct the network security evaluation index matrix according to the obtained data.

[0054] Step 2: Construct the network security evaluation index matrix For each network node, obtain the sample data of the transmission delay and packet loss rate indicators, and construct the network security evaluation index matrix. Reconstruct the obtained UAV network security evaluation index matrix to better fit the actual network security evaluation application scenario.

[0055] Input the evaluation sample data:

[0056] Among them, is the number of samples, is the th index of the

[0057] 1) Circular repetition: The circular operation regards the input matrix as two index vectors and changes the order in turn, that is, puts the first element at the last position of the vector, and the remaining elements are shifted forward. The expression is as follows:

[0058] Among them, represents the vector obtained by circularly shifting the initial vector bits, is the index among them. Since the elements of each vector are the same and only the element order is different, it is a circular repetition operation.

[0059] 2) Stacking: The stacking operation combines all the shifted column vectors obtained from the previous circular operation row by row to obtain a new network security evaluation index matrix. The operation expression is as follows:

[0060] Among them, is the combined evaluation index matrix, represents the th vector The overall transformation effect of the circular repetition stacking operation on the index matrix is as follows:

[0061] Among them, is the network security evaluation index matrix before circular repetition stacking, is the index among them. Any column vector of the right matrix and have the same elements but different orders; any column vector of the right matrix and contain the same elements but different orders.

[0062] After completing the construction of the UAV network security evaluation index matrix, the index matrix is obtained as the input of the subsequent model.

[0063] Step 3. Dimensionality reduction processing of the index matrix based on the non - linear incentive learning model Input the preprocessed and constructed evaluation index matrix into the evaluation model. Through the convolutional layer of the non-linear excitation learning model, perform convolutional operations based on non-linear functions to obtain a network security evaluation index matrix after dimensionality reduction, which reflects the characteristics of network security evaluation indexes. In the non-linear excitation learning model, the first layer after input is the convolutional layer. Through the convolution operation of the convolution kernel and the input matrix, a data matrix after dimensionality reduction is obtained:

[0064] Among them, and are continuous univariate functions, is the dimension of the input matrix. Each function can be expressed as a linear combination of multiple basic functions such as trigonometric functions and power functions.

[0065] The output of the model is realized through the activation function. Calculate the mean square error between the activation function and the obtained sample values to obtain the model loss function, and update the model parameters based on gradient descent, as shown in the following formula:

[0066] Among them, is the model parameter before update, is the model parameter after update, is the mean square error function, are the weight values in the sample and the weight output values obtained based on the new method model respectively, is the model parameter, is the step size of parameter update, is the number of samples.

[0067] After training, a non-linear excitation learning model is obtained to perform dimensionality reduction processing on the index matrix, and the obtained vector is output to the spatial information dynamic path iteration model to realize the extraction of weight information.

[0068] Step 4: Extract weight information based on the spatial information dynamic path iteration model In the non-linear excitation learning model, the first layer after input is the convolutional layer. Through the convolution operation of the convolution kernel and the input matrix, a data matrix after dimensionality reduction is obtained. Since the non-linear excitation learning model changes the linear operation of the original deep learning model to a non-linear excitation function, it can fit more complex non-linear relationships, capture the correlation between some network security index data, and is conducive to obtaining weights.

[0069] After performing convolution operations through the first convolutional layer of the non - linear excitation learning model, the convolved matrix is output to the first layer of the spatial information dynamic path iteration model, that is, the sub - unit layer, and the output is dynamically calculated through the spatial information dynamic path iteration model. Then, through the calculation of the parent - unit layer, a vector containing weight information is obtained.

[0070] After processing the data samples based on the non - linear excitation learning model, the obtained features are input into the spatial information dynamic path iteration model. Each lower - level sub - unit uses a learnable weight matrix to convert its output vector into a dynamic path information vector pointing to the parent - unit, that is, a higher - level dynamic path information vector in the spatial information dynamic path iteration model.

[0071] The total input of the model processing unit is calculated as the weighted sum of the path vectors:

[0072] where, is the input vector, is the path weight, is the sub - unit matrix.

[0073] The parent - unit processes the input vector using an activation function , and calculates the output vector :

[0074] where, is the norm calculation.

[0075] The path weights are dynamically updated according to the proximity between the output of the parent - unit extracted from the spatial information and each path vector:

[0076] where, is the dynamic factor, which realizes the dynamic update of the path weight from the model processing unit to the upper - level model processing unit through the real - time change of the sub - unit vector; is the total number of paths, is the exponential function.

[0077] After determining the path between the sub - unit and the parent - unit, an activation function is used to calculate the vector output.

[0078] In the spatial information dynamic path iteration model, the loss function is obtained by quantifying the difference between the actual weight and the sample, as shown in the following formula:

[0079] where, is the calculation error for any data sample, is the total error function for all samples, is the total number of samples, is the output data vector optimized through dynamic path, is the sample label value, is the dynamic coefficient factor, which can be updated in real time according to the loss function. k is the sample serial number, is the positive offset constant of the output vector, is the negative offset constant of the output vector. Preferably, and take values of 0.9 and 0.1 respectively.

[0080] Calculating the loss function can obtain the spatial information loss value of the path between the parent unit and the sub-unit for the currently extracted spatial information. Thus, based on the output of the loss function of the spatial information dynamic path iteration model, the network model is trained, and the corresponding weights of the network are gradually updated. When the loss function converges, an optimized model for calculating the weights of the UAV network based on the non-linear incentive learning and spatial information dynamic path iteration model is obtained. According to the optimized path obtained by iteration, efficient extraction and analysis of spatial information are realized, which is used to calculate the weight coefficient of the UAV network security assessment index threshold.

[0081] After being calculated by the spatial information dynamic path iteration model, replacing the original fully connected neural network model with the fully connected layer of the non-linear optimization deep learning model can reduce the memory consumption caused by the number of parameters, and thus obtain the output of the threshold weight faster.

[0082] Step 5: Calculation of the security assessment threshold weight vector based on the joint model After completing the preprocessing of the index matrix, a non-linear incentive learning model and a spatial information dynamic path iteration model are used to jointly train the optimized model for calculating the UAV network weights. The structure of the non-linear incentive learning and spatial information dynamic path iteration joint model is as Figure 4 shown.

[0083] In Figure 4 , the convolutional layer of the non-linear incentive learning model performs dimensionality reduction processing on the input matrix and outputs it to the spatial information dynamic path iteration model. After the sequential analysis and extraction of the sub-unit and the parent unit, weight information is obtained, and the network security assessment threshold weight vector output is calculated through the fully connected layer. The sub-unit refers to a module aggregated by a relatively small number of deep learning processing units, which is used to process local information. The parent unit is a complex unit aggregated by a relatively large number of modules, which can perform global processing and analysis of information. The path between the two is determined by the weight coefficients between each sub-unit and the parent unit and is in dynamic change.

[0084] Through the convolutional non - linear excitation learning model, perform convolutional operations based on non - linear functions to obtain a network security evaluation index matrix after dimensionality reduction, in which the correlation between network security evaluation indexes is better reflected. Further, input this matrix into the processing unit layer of the spatial information dynamic path iteration model. Through the layer - by - layer calculation of the model processing unit layer, a vector containing the importance information of network security evaluation indexes can be extracted. , the vector containing weight information is output to the linear non - linear excitation learning layer for the next dimensionality reduction and extraction processing, and a weight vector of the network security evaluation threshold can be obtained:

[0085] Among them, are the evaluation threshold weights of the three indexes of transmission delay, packet loss rate and throughput.

[0086] Construct an index matrix based on the drone network security index data and perform pre - processing operations. The pre - processed index matrix is input into the joint model and calculated and processed successively through the convolutional non - linear excitation learning layer, the spatial information dynamic path iteration model layer, and the non - linear excitation learning layer, and a weight vector of the network security evaluation threshold is output.

[0087] Step 6, output the network security evaluation result Extract the network security evaluation threshold through non - linear excitation learning and the spatial information dynamic path iteration model, and further obtain a network security evaluation threshold vector:

[0088] Among them, are the network security evaluation thresholds corresponding to network transmission delay, network packet loss rate and network throughput.

[0089] Based on the calculated threshold weight vector and threshold vector, perform network security evaluation on the drone network. Let the index vector obtained by normalizing the index data information of the drone network nodes be , and the normalized comprehensive threshold is:

[0090] Among them, are the network security evaluation thresholds of the three indexes, is the weight.

[0091] And, the calculation result of the evaluation vector of each network node is

[0092] Among them, is the calculated value of the evaluation vector, is the value of the group of metrics in the normalized metric vector. Compare the calculated value of the evaluation vector with the normalized comprehensive evaluation threshold to obtain the UAV network security evaluation result.

[0093] The actual security of the network is measured by metrics such as packet loss rate, transmission delay, and throughput. Specifically, in this embodiment, according to international network security standards such as ISO / IEC 27001 and ISO / IEC 24392:2023, as well as engineering experience, the corresponding range of network transmission delay is 500 - 2000 milliseconds; the range of network packet loss rate is 1% - 3%; the range of network throughput is 1 - 5 Mbps. By comparing the network security evaluation result with the actual security of the network, the accuracy of the network security evaluation result can be judged, and thus the network security evaluation accuracy rate can be obtained.

[0094] Now, perform a simulation test on the performance of the method in this embodiment. The simulation scenario uses a UAV network composed of 100 UAV nodes. The evaluation vector includes 3 performance parameters: transmission delay, packet loss rate, and throughput. The maximum number of training times in this embodiment, and the step size of parameter update are set. By setting the above simulation parameters and training the network, the curve of the UAV network security verification evaluation vector result is as Figure 5 shown.

[0095] In Figure 5 the evaluation result curve shown, the horizontal axis represents the number of UAV network nodes, and the vertical axis represents the UAV network security evaluation vector result value (i.e., the score value, with a value of 1 being a full score). From Figure 5 it can be seen that the evaluation result value of the method in this embodiment is above 0.8, with a better effect, meeting the needs of the engineering actual scenario, and it is an effective method for evaluating the UAV network security verification effect.

[0096] Moreover, in this embodiment, a simulation experiment on the evaluation accuracy rate of the network security verification effect evaluation method is carried out. By statistically calculating the ratio of the number of nodes with correct network security verification evaluation results to the total number of nodes in the target network, the evaluation accuracy rate of the network security verification evaluation method is obtained. Simulations are performed on the evaluation method based on AHP, the evaluation method based on BP neural network, and the method in this embodiment to obtain the corresponding evaluation accuracy rate curves, as Figure 6 shown. From Figure 6 it can be seen that the evaluation accuracy rate of the method in this embodiment reaches more than 95%, which is higher than the traditional methods and has a better evaluation effect.

[0097] Embodiment 2 This embodiment is based on Embodiment 1: This embodiment provides a system for evaluating the effect of UAV network security, including: A preprocessing and index matrix construction module, configured to preprocess index data related to UAV network security and construct an index matrix; An evaluation threshold and weight vector calculation module, configured to perform dimensionality reduction processing on the index matrix based on a non-linear incentive learning model, and extract weight information based on a spatial information dynamic path iteration model to obtain an evaluation threshold vector and its weight vector for UAV network security; A UAV network security evaluation module, configured to calculate a normalized comprehensive threshold based on the evaluation threshold vector and its weight vector; calculate an evaluation vector for each node in the UAV network, and compare it with the normalized comprehensive threshold to obtain the UAV network security evaluation result.

[0098] Embodiment 3 Based on Embodiment 1, this embodiment: This embodiment provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the UAV network security verification effect evaluation method of Embodiment 1. Among them, the computer program can be in the form of source code, object code, executable file, or some intermediate form, etc.

[0099] Embodiment 4 Based on Embodiment 1, this embodiment: This embodiment provides a computer-readable storage medium, storing a computer program, and when the computer program is executed by a processor, it implements the UAV network security verification effect evaluation method of Embodiment 1. Among them, the computer program can be in the form of source code, object code, executable file, or some intermediate form, etc. The storage medium includes: any entity or device capable of carrying computer program code, recording medium, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc. It should be noted that the content included in the storage medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the storage medium does not include electrical carrier signals and telecommunication signals.

[0100] It should be noted that for the foregoing method embodiments, for the sake of simplicity of description, they are expressed as a series of action combinations. However, those skilled in the art should know that this application is not limited by the described action sequence, because according to this application, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

Claims

1. A method for evaluating the network security verification effect of an unmanned aerial vehicle, characterized in that: include: Preprocess the indicator data related to drone network security and construct an indicator matrix; The indicator matrix is ​​reduced in dimension based on the nonlinear incentive learning model, and the weight information is extracted based on the spatial information dynamic path iteration model to obtain the evaluation threshold vector and weight vector of the UAV network security. Based on the evaluation threshold vector and its weight vector, the normalized comprehensive threshold is calculated; the evaluation vector of each node in the drone network is calculated and compared with the normalized comprehensive threshold to obtain the drone network security evaluation result.

2. The drone network security verification effect evaluation method according to claim 1 is characterized in that: The preprocessing of the indicator data related to the network security of the drone includes: In order to solve the data anomaly or missing problem, data cleaning is used to process missing values, outliers and duplicate data. Missing values ​​are processed by deletion, filling or interpolation, outliers are identified and processed by statistical methods or prior knowledge of drone networks, and duplicate data are deleted or integrated into the same data. To solve the problem of indicator differences, the indicator data is normalized, that is, all indicator values ​​are converted into dimensionless values ​​between 0 and 1. The expression is as follows: in, For the The node The index value, is the number of network nodes, This is the result after normalization preprocessing.

3. The drone network security verification effect evaluation method according to claim 1 is characterized in that: The construction of the indicator matrix includes: Construct an indicator matrix based on the preprocessed indicator data; Reconstruct the indicator matrix, and the reconstruction method includes loop operation and stacking operation; The loop operation includes treating the input matrix as two index vectors and changing the order in sequence, that is, placing the first element at the last position of the vector and shifting the remaining elements forward; The stacking operation includes combining all shifted column vectors obtained by the loop operation row by row to obtain a new indicator matrix.

4. The drone network security verification effect evaluation method according to claim 1 is characterized in that: The dimension reduction process of the indicator matrix based on the nonlinear incentive learning model includes: Through the convolution layer of the nonlinear excitation learning model, a convolution operation based on a nonlinear function is performed to obtain the indicator matrix after dimensionality reduction: in, is the indicator matrix after dimension reduction, is the input matrix, is the dimension of the input matrix; and is a continuous univariate function; The model output is realized through the activation function, the mean square error between the activation function and the acquired sample value is calculated to obtain the model loss function, and the model parameters are updated based on gradient descent: in, are the model parameters before updating, are the updated model parameters, is the step size of parameter update, is the mean square error function, is the sample size, is the weight value in the sample, is the weight output value obtained based on the nonlinear incentive learning model.

5. The drone network security verification effect evaluation method according to claim 1 is characterized in that: The spatial information-based dynamic path iteration model extracts weight information to obtain the evaluation threshold vector and weight vector of drone network security, including: After the convolution operation is performed through the convolution layer of the nonlinear excitation learning model, the convolution indicator matrix is ​​output to the first layer, i.e., the sub-unit layer, of the spatial information dynamic path iteration model, and is dynamically calculated through the sub-unit layer and output to the second layer, i.e., the parent unit layer, and then the vector containing the weight information is calculated through the parent unit layer; Based on the weight information extracted by the spatial information dynamic path iteration model, the next step of dimensionality reduction and extraction processing is performed through the fully connected layer of the linear nonlinear excitation learning model to obtain the evaluation threshold weight vector of the drone network security : in, It is the evaluation threshold weight of multiple indicators.

6. The drone network security verification effect evaluation method according to claim 5 is characterized in that: In the spatial information dynamic path iteration model, the total input of the processing unit is the weighted sum of the path vectors: in, is the input vector, is the path weight, is the subunit matrix; The parent unit processes the input vector using the activation function , calculate the output vector : in, is the activation function, is the norm calculation; Path Weight Dynamically update the proximity between the parent unit output and each path vector based on the spatial information: in, and It is a dynamic factor, which is realized by the real-time change of the sub-unit vector through the processing unit Upward processing unit Path weight for transmitting data Dynamic updates; is the total number of paths, is an exponential function; The loss function is obtained by quantifying the difference between the actual weight and the sample: in, is the calculation error for any data sample, is the sample label value, To optimize the output data vector through dynamic path, k is the sample serial number, is the positive offset constant of the output vector, is the negative offset constant of the output vector, is the dynamic coefficient factor, is the total error function for all samples, is the total number of samples.

7. The drone network security verification effect evaluation method according to claim 1 is characterized in that: The step of calculating a normalized comprehensive threshold based on the evaluation threshold vector and its weight vector includes: in, is the normalized comprehensive threshold, To evaluate the threshold, To evaluate the threshold weight, is the normalized index vector; The calculation of the evaluation vector of each node of the drone network includes: in, Compute values ​​for the evaluation vector, is the normalized index vector The value of the group indicator.

8. A drone network security verification effect evaluation system, characterized in that: include: The preprocessing and indicator matrix construction module is configured to preprocess the indicator data related to the network security of drones and construct an indicator matrix; The evaluation threshold and weight vector calculation module is configured to perform dimensionality reduction processing on the indicator matrix based on a nonlinear incentive learning model, and extract weight information based on a spatial information dynamic path iteration model to obtain an evaluation threshold vector and its weight vector for drone network security; The drone network security assessment module is configured to calculate a normalized comprehensive threshold based on an assessment threshold vector and its weight vector; calculate the assessment vector of each node in the drone network and compare it with the normalized comprehensive threshold to obtain a drone network security assessment result.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the drone network security verification effect evaluation method described in any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for evaluating the network security verification effect of a drone as described in any one of claims 1 to 7 is implemented.

Images