An intelligent command and control network structure and key node prediction method

Abstract

The present invention discloses an intelligent command and control network structure and key node prediction method, which belongs to the field of communication reconnaissance and artificial intelligence. It aims to realize the prediction of changes in structure and key nodes in the command and control network. The method comprises the following steps: first, a data set of a specified accumulation time is extracted from the data obtained from the communication reconnaissance by using the sliding window technology, and the data set is divided into a training set and a test set; then, a deep neural network model based on a gated recurrent unit is constructed and trained, and the prediction performance of the model is optimized by continuously adjusting the accumulation time; finally, the trained network model is used to predict the new command and control network data, and the network structure number and the key node number are output. By traversing different accumulation time lengths, the present invention can effectively identify and utilize the laws in the communication reconnaissance data, thereby improving the prediction accuracy of the model. The present invention is suitable for the analysis and prediction of complex command and control networks, and has important application value.

Description

Technical Field

[0001] The invention discloses an intelligent command and control network structure and a key node prediction method, and belongs to the field of communication reconnaissance and artificial intelligence. Background Art

[0002] The command and control network is the nerve center of unmanned intelligent command and control. The changes in its identification structure and key nodes have an important impact on the intelligence acquisition and reconnaissance of the network. Traditional prediction methods are usually based on rules or simple statistical analysis, which makes it difficult to capture complex time dependencies and periodic changes. In the patent "A method for identifying key nodes in a command and control network based on deep learning" applied by the 54th Institute of China Electronics Technology Group Corporation, with the authorization announcement number CN11478474B, the identification of key nodes is realized based on an improved convolutional neural network, but the analysis of dynamic changes in the network is not realized. In the article "Reliability Analysis of Command and Control Network System Based on Generalized Continuous Time Bayesian Network" published by Li Yifan et al. in "System Engineering and Electronic Technology", the Bayesian network can be used to analyze node characteristics, thereby analyzing the reliability of the network. However, compared with convolutional neural networks and Bayesian networks, gated recurrent units, as a neural network structure that can process sequence data, have strong memory capabilities and can effectively capture and model long-term and short-term dependencies in time series data. By using a deep neural network based on gated recurrent units to predict the command and control network structure and key nodes, the responsiveness to network changes and the prediction accuracy can be improved. In the article "Research on the Importance of Dynamic Network Nodes Based on Spatiotemporal Graph Convolution" published by Zhang Kai et al. in "College Mathematics", the graph neural network and long short-term memory network are combined to extract the characteristics of dynamic changes in the network, thereby determining the importance of the nodes. However, the prediction of changes in network structure and key nodes has not been achieved. In addition, the change characteristics of network structure and key nodes are related to the length of observation time. The accumulation time of observations will affect the performance of the prediction model, so a method is needed to identify and utilize this feature to improve the prediction accuracy. Summary of the invention

[0003] The present invention provides an intelligent command and control network structure and key node prediction method, which aims to effectively capture and utilize the time dependency in the command and control network to improve the prediction accuracy. The method first generates a two-dimensional data set with periodic characteristics based on the communication reconnaissance results, including the network structure number and the key node number; then extracts data of a specified period length from the data set through the sliding window technology, and divides it into a training set and a test set. Next, the network model is used for training, and the prediction performance of the model under different accumulation times is optimized by traversing different accumulation times, and the optimal accumulation time and corresponding network parameters are saved. Finally, the newly acquired reconnaissance data is predicted using the trained network model, and the network structure number and the key node number are output.

[0004] The technical method adopted in the present invention is:

[0005] An intelligent command and control network structure and key node prediction method, comprising the following steps:

[0006] Step 1: Obtain the number of structural types and key nodes of the command and control network, and number them. The structural type numbers and key node numbers obtained within a set time period are combined into a reconnaissance data set;

[0007] Step 2: Using the set accumulation time as the window length, perform sliding window extraction on the reconnaissance data set to generate the data set and label set required for training and testing, and split the data set and label set according to the set split ratio to form a training data set, a test data set, a training label set, and a test label set;

[0008] Step 3: Build a network, including a gated recurrent unit and two fully connected layers. The gated recurrent unit is used to extract data features, and the two fully connected layers are used to map the features into network structure numbers and key node numbers respectively; and use the training data set and training label set to train the network, and use the test data set and test label set to test the network to obtain the trained network parameters;

[0009] Step 4: Change the value of the accumulation time and return to step 2 until the set number of times is reached to obtain the optimal accumulation time and optimal network parameters;

[0010] Step 5: Based on the optimal accumulation time association, the current reconnaissance data is obtained, and the optimal network parameters are loaded. The current reconnaissance data is input into the network to obtain the estimated network structure number and key node number.

[0011] Furthermore, the specific process of step 1 is:

[0012] The communication reconnaissance equipment is used to obtain the number of structural types N and the number of key nodes M of the command and control network, and the structural types and key nodes of the command and control network are numbered in the order of reconnaissance, and L continuous time reconnaissance data sets are obtained in the format of [structural type serial number, key node serial number]; where L is the set value.

[0013] Furthermore, the specific process of step 2 is:

[0014] Step 2-1: Set the accumulation time p as the window length, perform sliding window extraction on the generated reconnaissance data set, and generate data X t = [ d t , ⋯ , d t + p − 1 ] p × 2 and tags ; and summarize the data and tags , generate the dataset X and label set Y required for training and testing, where, t ∈ [ 1 , L − p ] , the dimension of the dataset X is , the dimension of the label set Y is ;

[0015] Step 2-2: Set the split ratio , split the data set X and label set Y into dimensions The training dataset , The test dataset , The training label set and The test label set .

[0016] Furthermore, the specific process of step 3 is as follows:

[0017] Step 3-1: Build a network. The network structure includes a gated recurrent unit and two fully connected layers. The gated recurrent unit is used to extract data features. The two fully connected layers are used to map features to network structure numbers and key node numbers, respectively. The output dimensions correspond to N and M respectively.

[0018] Step 3-2: From the training dataset and training label set Extract some data and use the network to estimate:

[0019] Step 3-3: Calculate the loss using the cross entropy function for the network estimate and training labels;

[0020] Step 3-4: Set the learning rate, use the Adam optimizer, and use the loss Optimize network parameters;

[0021] Step 3-5: Repeat steps 3-2 to 3-4 until the training data set and training label set are extracted;

[0022] Step 3-6: Repeat steps 3-2 to 3-5 for a set number of times;

[0023] Step 3-7: Set the network to disable gradient calculation and input the test data set Get the estimated label set ;

[0024] Step 3-8: Using the accuracy score, calculate the estimated label set and the test label set The correlation , and save the corresponding network parameters .

[0025] Furthermore, the specific process of step 4 is as follows:

[0026] Step 4-1: Change the value of the accumulated time p, return to step 2 to set the number of times, and obtain the corresponding correlation of the accumulated time Store to List middle;

[0027] Step 4-2: List Use the elbow method to obtain the optimal accumulation time , the optimal accumulation time The corresponding network parameters are taken as the optimal network parameters .

[0028] The advantages of the present invention compared with the prior art are:

[0029] 1. Strong adaptability: It can dynamically grasp the law of network changes based on the historical communication reconnaissance data obtained, and can detect real-time changes in network structure and key nodes.

[0030] 2. Low computational complexity: It can accurately extract the accumulation time required to obtain the optimal prediction accuracy, avoiding the burden of excessive data accumulation on the network structure, scale, training and operation. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The present invention provides a flowchart of an intelligent command and control network structure and a key node prediction method.

[0032] Figure 2 This is a graph showing the relationship between the accumulation time and accuracy in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] The present invention will be further described below in conjunction with the accompanying drawings and specific embodiments.

[0034] An intelligent command and control network structure and key node prediction method, the overall process is as follows Figure 1 As shown, the specific steps include:

[0035] Step 1: Reconnaissance dataset construction;

[0036] Using communication reconnaissance equipment, obtain the number of command and control network structure types N=10 and the number of key nodes M=10, number the command and control network structure types and key nodes in the order of reconnaissance, and obtain L=36 continuous time reconnaissance data sets in the format of reconnaissance data d=[structure type serial number, key node serial number], the dimension of the reconnaissance data is 2, and is stored in the form of a list;

[0037] [ [2, 8], [3, 9], [2, 8], [3, 9], [4, 6], [2, 8], [2, 8], [3, 9], [2,8], [3, 9], [4, 6], [2, 8], [2, 8], [3, 9], [2, 8], [3, 9], [4, 6], [2, 8], [2, 8], [3, 9], [2, 8], [3, 9], [4, 6], [2, 8], [2, 8], [3, 9], [2, 8], [3, 9], [4, 6], [2, 8], [2, 8], [3, 9], [2, 8], [3, 9], [4, 6], [2, 8],] In order to verify the effectiveness of the algorithm, the data at the next moment is saved for verification, and its value is [2, 8].

[0038] Step 2: Sliding window extraction and segmentation of the reconnaissance dataset;

[0039] Step 2-1: Sliding window extraction of the reconnaissance dataset

[0040] The generated reconnaissance data is extracted by sliding window, with the accumulation time p=1 as the window length, and the data is generated X t = [ d t , ⋯ , d t + p − 1 ] p × 2 and tags ,in t ∈ [ 1 , L − p ] ; and summarize the data and tags , generate the dataset X and label set Y required for training and testing, where the dimension of dataset X is , the dimension of the label set Y is .

[0041] Step 2-2: Scouting dataset segmentation;

[0042] According to the division ratio = 0.8, split the reconnaissance data set into dimensions The training dataset , The test dataset , The training label set and The test label set .

[0043] Step 3: Build, train and test the network;

[0044] Step 3-1: Build the network;

[0045] network It consists of a gated recurrent unit and two fully connected layers. The gated recurrent unit is used to extract data features. The input dimension is input_dim=2, and the feature dimension of the hidden state is hidden_dim=64. The two fully connected layers are used to map the features to the network structure number and the key node number respectively. The input dimension is hidden_dim, and the output dimensions are N and M respectively.

[0046] Step 3-2: From the training dataset and training label set Extract batch=32 data for estimation:

[0047] .

[0048] Step 3-3: The loss function is the cross entropy function In this embodiment, the CrossEntropyLoss function provided by the pytorch framework is used, and the calculation process is:

[0049] .

[0050] Step 3-4: Use the Adam optimizer and set the learning rate to =0.001, using Optimize network parameters.

[0051] Step 3-5: Repeat steps 3-2 to 3-4 until the training data set and training label set are extracted.

[0052] Step 3-6: Repeat steps 3-2 to 3-5 for a total of E = 500 times.

[0053] Step 3-7: Set the network to disable gradient calculation:

[0054] with torch.no_grad(]

[0055] Input test data set Get the estimated label set :

[0056] .

[0057] Step 3-8: Use the accuracy score. In this example, we use the accuracy_score function provided by the sklearn library to calculate the estimated label set. and the test label set The correlation , and save the corresponding network parameters .

[0058] Step 4: Traverse the accumulation time to obtain the optimal accumulation time and network model;

[0059] Step 4-1: Set the value range of p accumulation time to [1, 10], return to step 2 to set the number of times, and obtain the corresponding correlation of the accumulation time Store to List The relationship between accumulation time and accuracy is as follows: Figure 2 shown.

[0060] Step 4-2: List Use the elbow method to obtain the optimal accumulation time =3, the optimal accumulation time The corresponding network parameters are taken as the optimal network parameters .

[0061] Step 5: Predict the network structure and key nodes;

[0062] Step 5-1: Load the optimal network parameters .

[0063] Step 5-2: Based on the optimal accumulation time , associate to get the current reconnaissance data X t ′ = [ d t ′ , ⋯ , d t ′ − p ′ + 1 ] p ′ × 2 , that is, [[3, 9], [4, 6], [2, 8]].

[0064] Step 5-3: Input the network to get the estimated value , that is, [2, 8], The first dimension is the structure type numbered 3 and the second dimension is the key node numbered 6.

[0065] The prediction results are consistent with the collected data, which fully demonstrates that this method can predict the command and control network structure and key nodes.

Claims

1. An intelligent command and control network structure and key node prediction method, characterized in that: The following steps are involved: Step 1: Use communication reconnaissance equipment to obtain the number of structural types N and the number of key nodes M of the command and control network, number the structural types and key nodes of the command and control network in the order of reconnaissance, and obtain L continuous time reconnaissance data sets in the format of [structural type serial number, key node serial number]; where the reconnaissance data set is a two-dimensional data set with periodic characteristics, which can capture and utilize the time dependency in the command and control network, and L is a set value; Step 2: Using the set accumulation time as the window length, perform sliding window extraction on the reconnaissance data set to generate the data set and label set required for training and testing, and split the data set and label set according to the set split ratio to form a training data set, a test data set, a training label set, and a test label set; Step 3: Build a network, including a gated recurrent unit and two fully connected layers. The gated recurrent unit is used to extract data features, and the two fully connected layers are used to map the features into network structure numbers and key node numbers respectively; and use the training data set and training label set to train the network, and use the test data set and test label set to test the network to obtain the trained network parameters; Step 4: Change the value of the accumulation time and return to step 2 until the set number of times is reached to obtain the optimal accumulation time and optimal network parameters; Step 5: Based on the optimal accumulation time association, the current reconnaissance data is obtained, and the optimal network parameters are loaded. The current reconnaissance data is input into the network to obtain the estimated network structure number and key node number.

2. The intelligent command and control network structure and key node prediction method according to claim 1 is characterized in that: The specific process of step 2 is as follows: Step 2-1: Set the accumulation time p as the window length, perform sliding window extraction on the generated reconnaissance data set, and generate data and tags ; and summarize the data and tags , generate the dataset X and label set Y required for training and testing, where, , the dimension of the dataset X is , the dimension of the label set Y is ; Step 2-2: Set the split ratio , split the data set X and label set Y into dimensions The training dataset , The test dataset , The training label set and The test label set .

3. The intelligent command and control network structure and key node prediction method according to claim 2 is characterized in that: The specific process of step 3 is as follows: Step 3-1: Build a network. The network structure includes a gated recurrent unit and two fully connected layers. The gated recurrent unit is used to extract data features. The two fully connected layers are used to map features to network structure numbers and key node numbers, respectively. The output dimensions correspond to N and M respectively. Step 3-2: From the training dataset and training label set Extract some data and use the network to estimate: Step 3-3: Calculate the loss using the cross entropy function for the network estimate and training labels; Step 3-4: Set the learning rate, use the Adam optimizer, and use the loss Optimize network parameters; Step 3-5: Repeat steps 3-2 to 3-4 until the training data set and training label set are extracted; Step 3-6: Repeat steps 3-2 to 3-5 for a set number of times; Step 3-7: Set the network to disable gradient calculation and input the test data set Get the estimated label set ; Step 3-8: Using the accuracy score, calculate the estimated label set and the test label set The correlation , and save the corresponding network parameters .

4. The intelligent command and control network structure and key node prediction method according to claim 3 is characterized in that: The specific process of step 4 is as follows: Step 4-1: Change the value of the accumulated time p, return to step 2 to set the number of times, and obtain the corresponding correlation of the accumulated time Store to List middle; Step 4-2: List Use the elbow method to obtain the optimal accumulation time , the optimal accumulation time The corresponding network parameters are taken as the optimal network parameters .

Images