Industrial network security protection method, system, terminal and storage medium

Through the dynamic graph neural network fusion self-supervised learning method, detect abnormal behavior and verify device identity in industrial networks, and combine dynamic key generation mechanism and encryption algorithm adjustment, the problem of insufficient security protection capabilities in the existing technology is solved, and the implementation of the More efficient data transmission security and system security.

CN118827247BActive Publication Date: 2024-11-29NINGBO WATER ENVIRONMENT GROUP CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202411312408.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-20
Publication Date
2024-11-29
Estimated Expiration
2044-09-20

AI Technical Summary

Technical Problem

The prior art is difficult to effectively detect and respond to complex network attacks in industrial networks, especially in the dynamic adaptability and accuracy of data information of different security risk assessments, resulting in insufficient security protection capabilities.

Method used

The abnormal identification model is constructed by dynamic graph neural network fusion self-supervised learning method, and the topology of industrial networks is based on real-time detection of abnormal behavior and verification of device identity legitimacy, and the dynamic key generation mechanism and encryption algorithm are adjusted to ensure data confidentiality and Integrity.

Benefits of technology

Real-time abnormal detection and dynamic encryption processing of industrial networks are realized, the security of data transmission and system security are improved, abnormal behaviors and device authentication failures in the network are promptly responded to, parameters of the abnormal identification model are optimized, and the Test accuracy and efficiency.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118827247B_ABST
    Figure CN118827247B_ABST
Patent Text Reader

Abstract

The present application provides an industrial network security protection method, system, terminal and storage medium, which relates to the field of communication technology. A graph structure is constructed based on the topological structure of the industrial network to represent the communication relationship and data flow between devices; an initial key is generated and distributed to each device in the network, and the initial key is used to encrypt the message transmitted between the devices; in order to detect potential abnormal behaviors and verify the identity of the device in real time, an abnormality recognition model based on a dynamic graph neural network fusion self-supervised learning method is adopted; once an abnormality is found, a dynamic key generation mechanism will be triggered, and an appropriate encryption algorithm will be selected according to the risk level to generate a new key, and the problematic message will be re-encrypted; in addition, the system also includes functional modules such as decryption and protocol conversion, security verification, etc., and a receiving report will be generated after the processing is completed and fed back to the source end. If there is a problem in the report, the encryption algorithm will be adjusted and the abnormality recognition model parameters will be optimized.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the field of communication technology, and in particular to an industrial network security protection method, system, terminal and storage medium. Background Art

[0002] In industrial automation control, it is necessary to use industrial networks to interconnect two or more control systems to transmit industrial control data such as images and voice. Since industrial networks use dedicated industrial control protocols, there are many types of industrial control protocols. Usually, different control systems run different industrial control protocols, which requires the deployment of protocol converters between different control systems to achieve data transmission between control systems with different industrial control protocols. During the transmission process of industrial networks, network attacks are often encountered, resulting in abnormal communication between industrial networks, which ultimately affects the production efficiency of industrial systems.

[0003] The document with the prior art publication number CN118199994A provides an industrial network security protection method, system, terminal and storage medium. By obtaining a message sent by any subsystem in the industrial network system or a message sent by an external network device, the verification data in the message is verified, and it is determined whether the verification key in the verification data matches the number of the target end in the industrial network system. The message data is decrypted using the decryption method corresponding to the verification key to obtain the decrypted message data, and the decrypted message data is converted into message data of the target end protocol type, and then encrypted according to the encryption method of the target end to obtain the encrypted message, and the encrypted message is sent to the target end, and a reception report fed back by the target end is obtained, and the encryption method pointing to the target end is changed according to the reception report. Although the above method has improved the security protection capability of the industrial network system to a certain extent, with the continuous evolution and complexity of network attack methods, passive protection methods that only rely on verification key matching and simple encryption and decryption processes have been unable to meet the growing security needs of industrial networks. In response to the above problems, the prior art CN114629674A discloses an industrial control network security risk assessment method based on an attention mechanism, which uses an attention mechanism to establish a risk assessment model for a variety of industrial control network scenarios, inputs data from a security risk assessment graph database, and obtains key elements and comprehensive risk scores of the industrial control network to be tested, thereby realizing security risk assessment of the industrial control network to be tested, and actively analyzing the security risks existing in the network. However, its anomaly detection capability is limited, and it is difficult to achieve dynamic adaptive adjustment and optimization of different security risk assessment data information, which makes it lack in accuracy and efficiency in risk detection.

[0004] In view of this, we propose an industrial network security protection method, system, terminal and storage medium. Summary of the invention

[0005] The purpose of the present invention is to provide an industrial network security protection method, which can detect abnormal behavior in message flows and verify the legitimacy of the identity of equipment in real time, and can adopt reasonable encryption processing methods according to the abnormal level to ensure the confidentiality and integrity of data at different security levels.

[0006] The present invention provides an industrial network security protection method, system, terminal and storage medium, and the technical solution includes the following aspects.

[0007] According to one aspect of an embodiment of the present application, an industrial network security protection method is provided, and the method includes the following steps.

[0008] S101. Based on the topological structure of the industrial network, a graph structure is constructed to represent communication and data flow between devices.

[0009] S102: Generate an initial key and distribute it to each device in the network, and use the initial key to encrypt messages transmitted between the devices.

[0010] S103, using a dynamic graph neural network fused with a self-supervised learning method to build an anomaly recognition model, and performing feature pre-training in a self-supervised manner, inputting the graph structure to be tested into the trained anomaly recognition model to obtain anomaly recognition results.

[0011] S104. When an anomaly is detected, the corresponding message is marked and transmission is blocked, a dynamic key generation mechanism is triggered, an encryption algorithm is selected based on the anomaly risk level to generate a new dynamic key, and the anomaly message is re-encrypted.

[0012] S105: Decrypt the message using a decryption algorithm, and convert it into a protocol format that can be recognized and processed by the target end for transmission according to the protocol type of the detected message.

[0013] Furthermore, the industrial network security protection method also includes the following steps.

[0014] S106: The decrypted message is again subjected to security verification to ensure that there are no security loopholes in the decryption and format conversion process, and the verified message is sent to the target end.

[0015] S107. The target end feeds back a reception report to the source end to inform the status of message reception and processing. The reception report is transmitted in the form of a message.

[0016] S108. When a problem occurs in the received report, the specific details of the problem are recorded, feedback optimization is performed, the encryption algorithm is adjusted, and the information is fed back to the anomaly recognition model to optimize the parameters of the anomaly recognition model.

[0017] Preferably, in step S101, a graph structure is constructed to represent the communication relationship and data flow between devices. The graph consists of nodes (devices) and edges (communication connections). The graph structure can be represented as G=(V,E), where V is a node set representing each device in the network, and E is an edge set representing the communication line or data transmission path between devices. Each node v i ∈V represents a device. Nodes can have additional attributes, such as device type, IP address, sent message type, received message type, and message content summary. Each edge e ij ∈E represents device v i and device v j The communication connection between the edges can be attached with attributes, such as bandwidth, delay, message type, data flow, and message content. The adjacency matrix is ​​used to represent the graph structure. The adjacency matrix is ​​an n×n matrix A, where A ij Indicates device v i and v j The connection property between them is ij Existence, A ij For device v i and v j The connection property between them, otherwise, A ij =0.

[0018] Preferably, in step S102, an asymmetric encryption algorithm ECC is used to generate an initial key. The initial key generation process is to select an elliptic curve C and a base point P, generate a private key d and a public key Q, and distribute the generated public key Q to all devices in the network through an insecure channel. After the devices in the network receive the public key Q, they store the public key Q for data encryption.

[0019] Preferably, in step S103, a dynamic graph neural network fusion self-supervised learning (DGNN-SSL) method is used to build an anomaly recognition model. DGNN-SSL combines the characteristics of dynamic graph convolutional networks and self-supervised learning to better capture the dynamic change characteristics of nodes and edges, and performs feature pre-training in a self-supervised manner to improve the accuracy and generalization ability of anomaly detection. Dynamic graph neural network (DGCN) part input graph sequence G={G1,G2,…,G T}, where each graph G t=(V,E) represents the network state at time step t. The dynamic graph convolution layer uses a graph convolutional neural network (GCN) to aggregate neighborhood information, and uses a time convolution layer or a recurrent neural network to process the time series information of the graph, and enhances the modeling ability of the dynamic graph by learning the time series changes of node features. The self-supervised learning (SSL) part introduces two self-supervised learning tasks to improve the learning ability and anomaly detection effect of graph feature representation. The two self-supervised learning tasks are the node attribute reconstruction task and the graph structure comparison learning task. Among them, the node attribute reconstruction task is for each node v i , using its neighborhood information to predict its own features. The graph structure contrastive learning task uses the graph contrastive learning method to construct positive and negative sample pairs on the time series, maximize the similarity of the representation of the same node or edge features at different time steps, and minimize the representation similarity of unrelated nodes or edges. The final training goal is to minimize the weighted sum of the node attribute reconstruction loss and the graph structure contrastive learning loss. In the inference stage, the graph structure to be tested is used as the model input. For each node v i and edge ij , the anomaly score is determined by calculating the degree of deviation between its feature representation and the normal pattern learned during the self-supervised learning pre-training process. When the anomaly score of a node or edge exceeds the preset threshold, it is considered to have abnormal behavior, and the anomaly level can be identified based on the anomaly score, and the corresponding processing strategy can be adopted.

[0020] Preferably, in step S104, the anomaly recognition model detects that the anomalies existing in the message transmission process include two categories, namely, abnormal behavior in the message flow and device identity anomaly. For the abnormal behavior in the message flow, different processing is performed according to the level of recognition by the anomaly recognition model, specifically: for a safe message flow, the target end directly receives it and performs subsequent processing; for a low-risk message flow, the original asymmetric encryption algorithm is used to replace the key (Q′, R′), and the message is encrypted with the new public key Q′ and then retransmitted; for a medium-risk message flow, a symmetric encryption algorithm is added on the basis of replacing the key with the original asymmetric encryption algorithm, that is, the symmetric encryption algorithm is used to generate a symmetric key K AES , use the replaced public key Q′ to generate the symmetric key K AES Encrypt and get the encryption key E KAES =Encrypt(K AES ,Q′), using the symmetric key K AES Encrypt the message data M to obtain the encrypted message E M =Encrypt(M,K AES ), E KAES and E MFor high-risk message flows, the asymmetric encryption algorithm is replaced to generate a new asymmetric key pair (Q", R"), and the symmetric encryption strategy is combined to encrypt and transmit the message in accordance with the medium-risk processing method.

[0021] Preferably, security verification is performed again in step S106, including integrity check, data consistency check and security vulnerability detection, wherein the integrity check includes: 1) hash value verification, checking the hash value of the message to ensure that the data has not been tampered with during transmission and decryption; 2) digital signature verification, if the message contains a digital signature, the signature is verified using the public key to ensure the authenticity of the message source. The data consistency check uses structure verification to ensure that the decrypted and converted data conforms to the expected structure and format, such as checking whether the data field is complete and whether the data type is correct. Security vulnerability detection includes: 1) input verification, checking the data to prevent common security vulnerabilities and ensure that all data is valid and secure; 2) anomaly detection, by using anomaly detection models or rule engines, checking whether the data contains any abnormal or malicious content. After the security verification is passed, the message data is passed to the application layer for normal processing. This step usually includes business logic processing, data storage or further analysis.

[0022] Preferably, in step S108, the report sent back by the target end is received and decrypted, the problem details in the report are extracted, and the specific problems recorded in the report are analyzed. Classification and diagnosis are performed according to the problem type in the report, a new encryption algorithm is selected or the existing algorithm is adjusted to solve the problem pointed out in the report, the encryption system is reconfigured, and a new key pair is generated to ensure the security of the system. The problem details and encryption algorithm adjustment information are fed back to the anomaly recognition model for model updating and optimization, and the anomaly generation and recognition model is retrained using the feedback information, and the model parameters are adjusted to improve the detection accuracy and optimize the model parameters.

[0023] In one aspect of an embodiment of the present application, an industrial network security protection system is provided, and the system includes the following modules.

[0024] The topology building module builds an initial graph structure based on the topology of the industrial network, represents the communication and data flow between devices, and defines the connection relationship between devices in the network.

[0025] The key management module generates keys for asymmetric encryption algorithms and is responsible for distributing public keys to various devices in the network to ensure the initial security of communications.

[0026] The encryption communication module encrypts the message using the key generated by the key management module when transmitting the message between devices.

[0027] The anomaly recognition module is an anomaly recognition model built based on a dynamic graph neural network fused with a self-supervised learning method. It is used to detect abnormal behaviors in message flows in real time and verify the legitimacy of the device's identity.

[0028] The dynamic key update module, when abnormal behavior or device authentication failure is detected, marks the message and prevents its further transmission, triggers the dynamic key generation mechanism, selects the appropriate encryption algorithm based on the current network conditions and the detected risk level to generate a new dynamic key to re-encrypt the message with the abnormal problem.

[0029] The decryption and protocol conversion module uses a decryption algorithm to decrypt and convert the message into a format that can be recognized and processed by the target end according to the detected protocol type.

[0030] Security verification module: the decrypted message is again verified for security in this module to ensure that there are no security loopholes in the decryption and format conversion process, and the verified message is sent to the target end.

[0031] In the feedback optimization module, the target end will generate a reception report and send it back to the source end in the form of an encrypted message to inform the source end of the message reception and processing status. If there is a problem in the reception report, the specific details of the problem will be recorded, the encryption algorithm will be adjusted, and the parameters of the anomaly recognition model will be optimized.

[0032] According to one aspect of an embodiment of the present application, a terminal device is provided, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, and the processor implements the above-mentioned industrial network security protection method when executing the computer program.

[0033] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, wherein the computer-readable storage medium includes a stored computer program, wherein when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the above-mentioned industrial network security protection method.

[0034] Compared with the prior art, the present application has the following beneficial effects.

[0035] By combining asymmetric encryption and symmetric encryption strategies, the security of data transmission is improved, especially in medium-risk and high-risk message flows, different encryption methods are used to ensure the confidentiality and integrity of data at different security levels. At the same time, through the dynamic key update mechanism, abnormal behaviors and device identity authentication failures in the network can be responded to in a timely manner, further enhancing the security of the system.

[0036] By generating a reception report and feeding it back to the source, timely processing of abnormal situations and dynamic adjustment of encryption algorithms can be achieved. This feedback mechanism can not only detect and solve problems in the system in a timely manner, but also optimize the parameters of the abnormal identification model according to actual conditions, thereby improving the detection accuracy and efficiency of the system.

[0037] In summary, the present application has significant advantages in improving data transmission security, ensuring data integrity, promptly handling abnormal situations and system scalability, and can provide more secure and reliable protection measures for industrial networks. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 A flow chart of an industrial network security protection method provided for one embodiment of the present application.

[0039] Figure 2 A structural diagram of an industrial network security protection system provided for one embodiment of the present application. DETAILED DESCRIPTION

[0040] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0041] Please refer to Figure 1 , which shows a flow chart of an industrial network security protection method provided by an embodiment of the present application, including the following steps.

[0042] S101. Based on the topological structure of the industrial network, a graph structure is constructed to represent communication and data flow between devices.

[0043] S102. Generate an initial key using an asymmetric encryption algorithm and distribute it to each device in the network to ensure the initial security of communication. When transmitting messages between devices, use the initial key to encrypt the messages to ensure the confidentiality of data during transmission.

[0044] S103. An anomaly recognition model is constructed using the dynamic graph neural network fusion self-supervised learning (DGNN-SSL) method to capture the dynamic change characteristics of nodes and edges, and perform feature pre-training through self-supervision to improve the accuracy and generalization ability of anomaly detection. The graph structure to be tested is input into the trained anomaly recognition model to obtain the anomaly recognition result.

[0045] S104. When abnormal behavior is detected in a message or device identity authentication fails, the message is marked and prevented from further transmission. At the same time, a dynamic key generation mechanism is triggered. Based on the detected risk level, a suitable encryption algorithm is selected to generate a new dynamic key, and the message with the abnormal problem is re-encrypted.

[0046] S105: Decrypt the message using a decryption algorithm, and convert it into a protocol format that can be recognized and processed by the target end according to the protocol type of the detected message.

[0047] S106. The decrypted message is again subjected to security verification to ensure that there are no security holes in the decryption and format conversion process. After the verification is passed, the message is processed normally.

[0048] S107: The target end generates a reception report and sends it back to the source end to inform the status of message reception and processing. The reception report is transmitted in the form of a message.

[0049] S108. When a problem occurs in the received reception report, the specific details of the problem are recorded, feedback optimization is performed, the encryption algorithm is adjusted, and the information is fed back to the anomaly recognition model to optimize the parameters of the anomaly recognition model.

[0050] In industrial networks, graph structures are used to represent the communication relationship and data flow between devices. Graphs are composed of nodes (devices) and edges (communication connections). In this embodiment, the graph structure can be represented as G=(V,E), where V is a node set and V={v1,v2,…,v n}, represents each device in the network, E is the edge set, E={e ij ∣e ij Connect node v i and v j}, indicating the communication line or data transmission path between devices. Each node v i ∈V represents a device. Nodes can have additional attributes, such as device type, IP address, sent message type, received message type, and message content summary. The sent message type represents the node (device) v i The type of message sent, such as control instructions, status updates, etc. The type of message received indicates the node (device) v i The received message type and message content summary indicate the content summary or characteristics of the message, such as data packet size, data format, etc. ij ∈E represents device v i and device v j The communication connection between the edges can be attached with attributes, such as bandwidth, delay, message type, data flow, and message content. The message type is indicated on the edge. ij The data flow rate indicates the data flow rate transmitted through the edge, and the message content describes the content characteristics of the message, such as keywords, data fields, etc. For example, given a device v i and v j The attributes of the edges between can be expressed as e ij =(v i ,v j, bandwidth, delay, message type, data flow, message content), use the adjacency matrix to represent the graph structure. The adjacency matrix is ​​an n×n matrix A, where A ij Indicates device v i and v j The connection property between them is ij Existence, A ij For device v i and v j The connection property between them, otherwise, A ij =0.

[0051] Further, in step S102, an initial key is generated using an asymmetric encryption algorithm ECC, the key length is 256 bits, and the initial key generation process is: select an elliptic curve C and a base point P (usually defined by a standard, such as secp256k1), generate a private key d, d∈Z n , is a random integer in the range [1, n-1], n is the order of the curve, and the public key Q is generated, which is the product of the private key and the base point, Q=d×G. The generated public key Q is distributed to all devices in the network through an insecure channel. Due to the characteristics of asymmetric encryption, even if the public key is intercepted, the attacker cannot deduce the private key d. After receiving the public key Q, the devices in the network store the public key Q for subsequent data encryption.

[0052] Furthermore, in step S103, a dynamic graph neural network fusion self-supervised learning (DGNN-SSL) method is used to build an anomaly recognition model. DGNN-SSL combines the characteristics of dynamic graph convolutional network (DGCN) and self-supervised learning (SSL) to better capture the dynamic change characteristics of nodes and edges, and perform feature pre-training in a self-supervised manner to improve the accuracy and generalization ability of anomaly detection. The dynamic graph neural network (DGCN) part inputs the graph sequence G={G1,G2,…,G T}, where each graph G t =(V,E) represents the network state at time step t, V is the node set, E is the edge set, and each node v i ∈V has a feature vector h i t ∈R d , each edge e ij ∈E has a feature vector e ij t ∈R d .

[0053] Specifically, the dynamic graph convolution layer uses a graph convolutional neural network (GCN) to aggregate neighborhood information, and uses a temporal convolutional layer (such as TCN) or a recurrent neural network (such as GRU or LSTM) to process the time series information of the graph. At each time step t, the node v i The feature update formula is:

[0054] h i (l+1),t =σ(∑ j∈N(i) 1 / c ij W (l) h j (l),t +W0 (l) h i (l),t ).

[0055] Among them, h i (l),t represents the feature representation of node i at layer l and time step t, N(i) is the set of neighbor nodes of node i, c ij is the normalization constant, W (l) and W0 (l) is a learnable weight matrix, σ is an activation function (such as ReLU), and a temporal convolutional network (TCN) or a gated recurrent unit (GRU) is used to capture the changes in node features over time. For example, using the TCN approach, for each node v i , whose features are updated through the temporal convolutional layer, h i (l+1) =TCN(h i (l),1:T ); The way to use GRU is, h i (l +1),t =GRU(h i (l),t ,h i (l+1),t-1 ), TCN and GRU represent temporal convolutional network and gated recurrent unit, respectively, which enhance the modeling ability of dynamic graphs by learning the time series changes of node features.

[0056] The self-supervised learning (SSL) part introduces two self-supervised learning tasks to improve the learning ability of graph feature representation and the anomaly detection effect. The two self-supervised learning tasks are node attribute reconstruction task and graph structure comparison learning task. Among them, the node attribute reconstruction task is for each node v i , use its neighborhood information to predict its own features, and define the loss function as the mean square error (MSE) of node feature reconstruction, L recon =∑ vi∈V ||h i -h^ i ||2 , where h^ i is the node v obtained by aggregating neighborhood information through the graph convolution layer i The graph structure contrastive learning task uses the graph contrastive learning method to construct positive and negative sample pairs on the time series, maximize the similarity of the feature representations of the same node or edge at different time steps, and minimize the similarity of the representations of irrelevant nodes or edges. The loss function can be defined as the contrastive loss (such as InfoNCE), L contrast = -log[exp(sim(h i t ,h i t′ )) / ∑ j=1 N exp(sim(h i t ,h j t′ ))], where sim(a,b) represents the similarity between two feature vectors a and b (such as dot product or cosine similarity). The final training goal is to minimize the weighted sum of node attribute reconstruction loss and graph structure contrast learning loss, L=αL recon +βL contrast , where α and β are hyperparameters used to balance the weights of the two losses. In the inference phase, for each node v i and edge ij , the anomaly score is determined by calculating the degree of deviation of its feature representation from the normal pattern learned during the self-supervised learning pre-training process, such as for node v i The anomaly score s i ,s i =||h i -h^ i || 2 , for edge e ij The anomaly score of can be calculated similarly. When the anomaly score of a node or edge exceeds a preset threshold, it is considered to have abnormal behavior, and the detected risk level is further divided accordingly, and the corresponding processing strategy is adopted. For example, a threshold τ is set to judge the anomaly. If the anomaly score s of a node or edge exceeds the preset threshold τ, it is considered to have abnormal behavior.

[0057] Furthermore, in step S104, the anomaly recognition model detects that the anomalies in the message transmission process include two categories, namely, abnormal behavior in the message flow and device identity anomaly. For the abnormal behavior in the message flow, progressive encryption processing is performed according to the level identified by the anomaly recognition model, specifically: 1) For safe message flows, the target end directly receives and performs subsequent processing; 2) For low-risk message flows, the original asymmetric encryption algorithm is used to replace the key (Q′, R′), and the message is encrypted with the new public key Q′ and then retransmitted; 3) For medium-risk message flows, the original asymmetric encryption algorithm is used to replace the key, and a symmetric encryption algorithm is added, that is, the symmetric encryption algorithm is used to generate a symmetric key K AES , use the replaced public key Q′ to generate the symmetric key K AES Encrypt and get the encryption key E KAES =Encrypt(K AES ,Q′), using the symmetric key K AES Encrypt the message data M to obtain the encrypted message E M =Encrypt(M,K AES ), E KAES and E M Transmitted to the target end together, 4) For high-risk message flows, replace the asymmetric encryption algorithm, generate a new asymmetric key pair (Q", R"), and combine the symmetric encryption strategy to encrypt and transmit the message according to the medium-risk processing method.

[0058] For device identity anomalies, that is, device identity authentication failure, further transmission of the message is blocked, and a device identity anomaly warning is issued to notify staff to verify. If abnormal behavior and device identity anomalies occur at the same time in the message flow, the handling of device identity anomalies is given a higher priority.

[0059] Furthermore, in step S106, security verification is performed again, including integrity check, data consistency check and security vulnerability detection, wherein the integrity check includes: 1) Hash value verification, checking the hash value of the message to ensure that the data has not been tampered with during transmission and decryption. received =Hash(M parsed ), Verify(H received ,H expected ), 2) Digital signature verification, if the message contains a digital signature, use the public key to verify the signature to ensure the authenticity of the message source, Verify_Signature(M parsed,Signature,Public_Key). Data consistency check uses structure validation to ensure that the decrypted and converted data conforms to the expected structure and format, such as checking whether the data fields are complete and the data type is correct, Validate_Structure (M parsed ,Schema). Security vulnerability detection includes: 1) Input validation, checking data to prevent common security vulnerabilities, such as SQL injection, cross-site scripting (XSS), etc., to ensure that all data is valid and secure, Sanitize (M parsed ), 2) Anomaly detection, by using anomaly detection models or rule engines to check whether the data contains any abnormal or malicious content. After the security verification is passed, the message data M parsed Passed to the application layer for normal processing. This step usually includes business logic processing, data storage, or further analysis.

[0060] Furthermore, a reception report is generated in step S107. The report includes the status of the received message, such as successful reception or a problem, and records the processing of the message, including data parsing, format conversion, and any security verification performed. If an exception occurs during the processing, the report will include the specific details of these exceptions. The report generation process is to extract key information from the processed message, such as processing results, exception details, etc., and organize the extracted information into a structured report, which usually includes fields such as status code, description information, and timestamp. After the reception report is generated, the report is encrypted using the same encryption method as the message, and is encapsulated into a message format to ensure transmission security, and the encrypted report is sent back to the source end through the network.

[0061] Further, in step S108, the report sent back by the source is received and decrypted, the problem details in the report are extracted, and the specific problems recorded in the report are analyzed, including data processing failure, unresolved anomalies, etc. Classification and diagnosis are performed according to the type of problem in the report (such as key management problems, encryption algorithm inapplicability, etc.), and a new encryption algorithm is selected or the existing algorithm is adjusted to solve the problem pointed out in the report. For example, it may be necessary to select a stronger encryption algorithm or improve the key management strategy, reconfigure the encryption system, and generate a new key pair to ensure the security of the system. Feedback the problem details and encryption algorithm adjustment information to the anomaly recognition model for model update and optimization, use the feedback information to retrain the anomaly recognition model, adjust the model parameters to improve detection accuracy, and optimize the model parameters.

[0062] Exemplarily, assuming that message data is transmitted between nodes (devices) V1 and V2, the specific steps are as follows.

[0063] 1. Build a graph structure (S101): Build a graph structure that represents the communication between nodes (devices) V1 and V2. The graph has two nodes V1 and V2, and an edge e 12 Represents the communication connection between V1 and V2.

[0064] 2. Generate an initial key (S102): Generate a 256-bit initial key using ECC. Devices V1 and V2 each generate a public key and a private key, and distribute the public key to ensure the initial security of communication. V1 uses V2's public key to encrypt and send messages.

[0065] 3. Anomaly Identification (S103): Input the graph structure between V1 and V2 into the trained DGNN-SSL model. The model identifies abnormal behavior by detecting dynamic features and self-supervised learning. For example, suppose that in time step t, the model detects that the message flow sent from V1 to V2 is abnormal. Calculated.

[0066] Node V1 anomaly score s V1 is 0.08, node V2 anomaly score s V2 is 0.06, and the anomaly score of edge e_{12} is 0.20.

[0067] Assuming that the threshold τ is set to 0.10, and an anomaly score greater than τ indicates abnormal behavior, the anomaly scores of V1 and V2 are both greater than the threshold, indicating that they may be abnormal, and the anomaly score of edge e_{12} significantly exceeds the threshold, indicating that there may be serious problems with the communication link.

[0068] And further divide the risk levels.

[0069] There is no abnormality in the identities of V1 and V2 devices, and edge e_{12} may be high risk.

[0070] 4. Handle abnormal behavior (S104).

[0071] High risk (edge ​​e_{12}): If the anomaly recognition model marks the message flow as high risk, it means that the anomaly level of the message flow is very high and may pose a serious threat to system security. At this time, stricter encryption measures need to be taken. Execute the high-risk processing strategy, that is, replace the asymmetric encryption algorithm and combine it with the symmetric encryption strategy, use the new asymmetric encryption algorithm to generate a new asymmetric key pair (Q", R"), and generate the symmetric key K AES , use the replaced public key Q" to encrypt the symmetric key K AES , get the encryption key E KAES =Encrypt(K AES ,Q"), using the symmetric key K AES Encrypt the message data M to obtain the encrypted message E M=Encrypt(M,K AES ), E KAES and E M Transfer to the target end.

[0072] 5. Message decryption and format conversion (S105): After receiving the encrypted message, V2 uses the public key Q" to decrypt E KAES Get K AES , use K AES DecryptionE M Get the original message. V2 converts the message into a processable protocol format.

[0073] 6. Security Verification (S106): V2 performs integrity checks (such as hash value verification and digital signature verification), data consistency checks, and security vulnerability detection. Confirm that the message has not been tampered with during transmission and decryption, and conforms to the expected structure and format.

[0074] 7. Generate a reception report (S107): V2 generates a report containing the processing status, such as the details of the successful reception and processing of the message. The report includes the exception handling situation and any security verification results.

[0075] 8. Feedback optimization (S108): After receiving the reception report from V2, V1 analyzes the problem details in the report (such as anomaly detection failure). If the report indicates that the encryption algorithm is not applicable, V1 adjusts the encryption algorithm or key management strategy, and uses the feedback information to optimize the anomaly recognition model to improve detection accuracy.

[0076] 9. Results and treatment: Assume that V2 detects abnormal traffic in the message and adopts a high-risk treatment strategy to encrypt the message. After verification, the message has not been tampered with during transmission. V1 adjusts the encryption algorithm based on the received report to ensure the security of future transmissions.

[0077] Please refer to Figure 2 , which shows a structural diagram of an industrial network security protection system provided by an embodiment of the present application, including the following modules.

[0078] The topology building module builds an initial graph structure based on the topology of the industrial network, represents the communication and data flow between devices, and defines the connection relationship between devices in the network.

[0079] The key management module generates keys for asymmetric encryption algorithms and is responsible for distributing public keys to various devices in the network to ensure the initial security of communications.

[0080] The encryption communication module encrypts the message using the key generated by the key management module when transmitting the message between devices.

[0081] The anomaly recognition module is an anomaly recognition model built based on the dynamic graph neural network fusion self-supervised learning (DGNN-SSL) method, which is used to detect abnormal behaviors in message flows in real time and verify the legitimacy of the device's identity.

[0082] The dynamic key update module, when abnormal behavior or device authentication failure is detected, marks the message and prevents its further transmission, triggers the dynamic key generation mechanism, selects the appropriate encryption algorithm based on the current network conditions and the detected risk level to generate a new dynamic key to re-encrypt the message with the abnormal problem.

[0083] The decryption and protocol conversion module uses a decryption algorithm to decrypt and convert the message into a format that can be recognized and processed by the target end according to the detected protocol type.

[0084] Security verification module: the decrypted message is again verified for security in this module to ensure that there are no security loopholes in the decryption and format conversion process, and the verified message is sent to the target end.

[0085] In the feedback optimization module, the target end will generate a reception report and send it back to the source end in the form of an encrypted message to inform the source end of the message reception and processing status. If there is a problem in the reception report, the specific details of the problem will be recorded, the encryption algorithm will be adjusted, and the parameters of the anomaly recognition model will be optimized.

[0086] The present invention provides a terminal device. The terminal device includes a central processing unit (CPU), a memory, a network interface card (NIC) and a security chip. The CPU is responsible for performing various computing tasks, the memory is used to store data and program codes, the NIC is used to connect to the industrial network and transmit data, and the security chip is specifically used to process security-related operations. The security chip has built-in multiple encryption algorithms and can support asymmetric encryption and dynamic key generation mechanisms. It works closely with the key distribution module to ensure the secure generation, storage and distribution of keys. In addition, the security chip also has a hardware-level message integrity verification function, which can verify at the hardware level whether the message has been tampered with during transmission.

[0087] The present invention also provides a storage medium, in which a computer program can be stored in a computer-readable storage medium, and the medium may include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

Claims

1. An industrial network security protection method, characterized in that: The following steps are involved: S101. Based on the topological structure of the industrial network, a graph structure is constructed to represent the communication and data flow between devices; S102, generating an initial key and distributing it to each device in the network, and using the initial key to encrypt messages transmitted between the devices; S103, using dynamic graph neural network fusion self-supervised learning method to build an anomaly recognition model, the dynamic graph neural network part input graph sequence G = {G1, G2, ..., G T }, the dynamic graph convolution layer uses a graph convolutional neural network to aggregate neighborhood information, and uses a temporal convolutional layer or a recurrent neural network to process the time series information of the graph, and performs feature pre-training in a self-supervised manner. The self-supervised learning part introduces node attribute reconstruction tasks and graph structure comparison learning tasks to improve the learning ability of graph feature representation and the anomaly detection effect. Among them, The node attribute reconstruction task is performed on each node v i , use its neighborhood information to predict its own features, and define the loss function as the mean square error of node feature reconstruction, L recon =∑ vi∈V ||h i -h^ i || 2 , where h^ i is the node v obtained by aggregating neighborhood information through the graph convolution layer i The reconstructed feature representation of The graph structure contrastive learning task uses the graph contrastive learning method to construct positive and negative sample pairs on the time series, maximize the similarity of the feature representations of the same node or edge at different time steps, and minimize the similarity of the representations of unrelated nodes or edges. The loss function is defined as the contrastive loss, L contrast = -log[exp(sim(h i t ,h i t′ )) / ∑ j=1 N exp(sim(h i t ,h j t′ ))], where sim(a,b) represents the similarity between two feature vectors a and b, and h i t and h i t′ denote the feature representation of node i at time step t and time step t′, respectively, and h j t′ represents the feature representation of node j at time step t′, The final training goal is to minimize the weighted sum of node attribute reconstruction loss and graph structure contrast learning loss, L = αL recon +βL contrast , where α and β are hyperparameters used to balance the weights of the two losses. The image structure to be tested is input into the trained anomaly recognition model, and the anomaly score is determined by calculating the degree of deviation between its feature representation and the normal pattern learned in the self-supervised learning pre-training process. The anomaly score is used to determine whether there is abnormal behavior and obtain the anomaly recognition result. S104, when an anomaly is detected, the corresponding message is marked and transmission is blocked, a dynamic key generation mechanism is triggered, an encryption algorithm is selected based on the anomaly risk level to generate a new dynamic key, and the anomaly message is re-encrypted; S105: Decrypt the message using a decryption algorithm, and convert it into a protocol format that can be recognized and processed by the target end for transmission according to the protocol type of the detected message.

2. The industrial network security protection method according to claim 1, characterized in that: The following steps are also included: S106, the decrypted message is again subjected to security verification to ensure that there are no security loopholes in the decryption and format conversion process, and the verified message is sent to the target end; S107, the target end feeds back a reception report to the source end, informing the status of message reception and processing, and the reception report is transmitted in the form of a message; S108. When a problem occurs in the received report, the specific details of the problem are recorded, feedback optimization is performed, the encryption algorithm is adjusted, and feedback is given to the anomaly recognition model to optimize the parameters of the anomaly recognition model.

3. The industrial network security protection method according to claim 1, characterized in that: In step S101, a graph structure is constructed for representing communication and data flow between devices, which is expressed as G = (V, E), where V is a set of nodes, representing each device in the network, and E is a set of edges, representing communication lines or data transmission paths between devices, and attributes are added to nodes and edges respectively; The adjacency matrix A is used to represent the graph structure, where A ij Indicates device v i and v j The connection property between them is ij Existence, A ij For device v i and v j The connection property between them, otherwise, A ij = 0, where edge e ij ∈E represents device v i and device v j The communication connection between them.

4. The industrial network security protection method according to claim 1, characterized in that: In step S102, an initial key is generated, specifically, using an asymmetric encryption algorithm ECC to generate the initial key. The initial key pair generation process is to select an elliptic curve C and a base point P, generate a private key d and a public key Q, and distribute the generated public key Q to all devices in the network through an insecure channel. After the devices in the network receive the public key Q, they store the public key Q for data encryption.

5. The industrial network security protection method according to claim 1, characterized in that: The dynamic key generation mechanism in step S104 performs progressive encryption processing on abnormal behaviors in the message flow according to the levels identified by the abnormal identification model.

6. The industrial network security protection method according to claim 5, characterized in that: The progressive encryption process is as follows: For secure message flows, the target end directly receives them and performs subsequent processing; For low-risk message flows, the original asymmetric encryption algorithm is used to replace the key (Q′, R′), and the message is encrypted with the new public key Q′ and then retransmitted; For medium-risk message flows, a symmetric encryption algorithm is added on the basis of replacing the key with the original asymmetric encryption algorithm. A symmetric key is generated using the symmetric encryption algorithm. The generated symmetric key is encrypted using the replaced public key Q′ to obtain an encryption key. The message data is encrypted using the symmetric key to obtain an encrypted message. The encryption key and the message encrypted with the symmetric key are transmitted to the target end together. For high-risk message flows, replace the asymmetric encryption algorithm, generate a new asymmetric key pair (Q", R"), and combine the symmetric encryption strategy to encrypt and transmit the message according to the medium-risk processing method.

7. The industrial network security protection method according to claim 2, characterized in that: In step S106, security verification is performed again including integrity check, data consistency check and security vulnerability detection. After the security verification passes, the message data is passed to the application layer for normal processing.

8. The industrial network security protection method according to claim 1, characterized in that: Feedback optimization in step S108: The problem details in the received report and the adjustment information of the encryption algorithm are fed back to the anomaly recognition model, and the anomaly recognition model is retrained using the feedback information to adjust and optimize the model parameters.

9. An industrial network security protection system, characterized in that: According to the industrial network security protection method described in claims 1-8, the system includes the following modules: The topology building module builds the initial graph structure based on the topology of the industrial network, represents the communication and data flow between devices, and defines the connection relationship between devices in the network; The key management module generates key pairs for asymmetric encryption algorithms and is responsible for distributing public keys to various devices in the network; The encryption communication module encrypts the message using the key generated by the key management module when transmitting the message between devices; The anomaly recognition module is an anomaly recognition model built based on a dynamic graph neural network and a self-supervised learning method. It is used to detect abnormal behaviors in message flows and verify the legitimacy of device identities in real time. The dynamic key update module, when abnormal behavior or device identity authentication failure is detected, marks the corresponding message and blocks the transmission, triggers the dynamic key generation mechanism, selects the encryption algorithm based on the current network conditions and the detected risk level to generate a new dynamic key to re-encrypt the abnormal message; The decryption and protocol conversion module uses a decryption algorithm to decrypt the message and converts the message into a format that can be recognized and processed by the target end according to the detected protocol type.

10. The industrial network security protection system according to claim 9, characterized in that: The system also includes the following modules: Security verification module: the decrypted message is again verified in this module to ensure that there are no security loopholes in the decryption and format conversion process, and the verified message is sent to the target end; In the feedback optimization module, the target end generates a reception report and sends it back to the source end in the form of an encrypted message to inform the source end of the message reception and processing status. If there is a problem in the reception report, the specific details of the problem are recorded, the encryption algorithm is adjusted, and the parameters of the anomaly recognition model are optimized.

11. A terminal device, characterized in that: The method comprises a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein when the processor executes the computer program, the industrial network security protection method according to any one of claims 1 to 8 is implemented.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium includes a stored computer program, and when the computer program is executed, the device where the computer-readable storage medium is located is controlled to execute the industrial network security protection method according to any one of claims 1 to 8.

Citation Information

Patent Citations

  • Industrial control network security risk assessment method based on attention mechanism

    CN114629674A

  • Industrial network security protection method and system, terminal and storage medium

    CN118199994A

  • Dynamic encryption method and system, computer equipment and storage medium

    CN117131484A

  • High-speed encryption method and device for business data of electric power internet of things

    CN117692257A