An Adaptive Admission Method for Distributed Industrial Control Protocols Based on AI Large Models
By using an AI-based large-scale model-based distributed industrial control protocol adaptive admission method, the network security situation is generated by parsing industrial control protocol messages and updating admission parameters. This solves the problem of balancing security protection and business availability in existing technologies, and achieves efficient response and low bandwidth consumption in industrial control networks.
Patent Information
- Application Number
- CN202611619640.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-10-20
- Publication Date
- 2026-11-17
AI Technical Summary
Existing network access control mechanisms struggle to balance security strength and service availability in dynamically changing industrial control network environments, leading to delayed responses or excessive rate limiting, and the policy synchronization process consumes a significant amount of redundant bandwidth.
An AI-based large-scale model-based distributed industrial control protocol adaptive access method is adopted. By collecting network traffic and parsing industrial control protocol messages, semantic analysis and threat identification are performed to generate network security situation. Based on the situation and policy decision model, access risk scores are generated and mapped to the continuous response spectrum to update network access parameters.
It enables fine-tuning of access control parameters in dynamic network environments, improving the balance between security protection strength and service availability, and reducing network bandwidth consumption and response lag risks.
Smart Images

Figure CN123123035A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of cybersecurity technology for industrial control systems, and more specifically, to an adaptive admission method for distributed industrial control protocols based on a large AI model. Background Technology
[0002] Industrial control system cybersecurity refers to a comprehensive security protection technology that employs a series of technical measures to ensure the availability, integrity, and confidentiality of data acquisition, transmission, processing, and command execution within industrial control networks. This technology is widely used in critical infrastructure sectors such as power dispatching, petrochemicals, rail transportation, and intelligent manufacturing, serving as a crucial support for maintaining the continuous and stable operation of industrial production. The core value of industrial control system cybersecurity technology lies in effectively resisting external network attacks and internal abnormal operations that disrupt the production process, ensuring the accurate transmission and execution of industrial control commands, thereby maintaining the safe and stable operation of critical infrastructure.
[0003] However, existing network access control mechanisms mostly employ static rule configuration or coarse-grained adjustments based on discrete threat levels. Furthermore, policy distribution heavily relies on global broadcasting or fixed-path transmission, making it difficult to smoothly adjust access control parameters with fine-grained precision in response to real-time multi-dimensional threat intensity. In heterogeneous network environments, policy distribution is susceptible to link limitations, resulting in transmission delays and inconsistencies with node states. This leads to delayed security response or excessive rate limiting during high-risk periods, disrupting normal production operations. Simultaneously, policy synchronization consumes significant redundant bandwidth, thereby reducing the overall security effectiveness of industrial control systems and the continuity of production operations. Therefore, how to dynamically and adaptively adjust network access control parameters to improve the timeliness of balancing security strength and service availability is a challenge facing the industry. Summary of the Invention
[0004] This application provides an adaptive admission method for distributed industrial control protocols based on AI large models, which can dynamically and adaptively adjust network admission control parameters to improve the timeliness of balancing security protection strength and service availability.
[0005] Firstly, this application provides an adaptive admission method for distributed industrial control protocols based on a large AI model, the method comprising the following steps: Collect network traffic in the industrial control network area and parse the industrial control protocol messages on the edge analysis nodes; The locally deployed protocol access control analysis model is invoked to perform semantic analysis on the industrial control protocol messages, outputting a semantic description sequence. The semantic description sequence and the abnormal behavior records of the current device operation are then used to identify the response and obtain the network security situation of the current network area. By co-evaluating the historical decision feedback information in the strategy decision model and the behavioral attack characteristics of the industrial control network in the current operation strategy, the access risk score of the current network area is obtained. The access risk score is then mapped to the continuous response spectrum of the industrial control network from detection to blocking to obtain access control instructions. Based on the network security situation and the access control instructions, the network access parameters of the current device are updated in conjunction with the network access control instructions, and the updated network access parameters are distributed to each edge analysis node.
[0006] In this embodiment, the locally deployed protocol admission analysis model is invoked to perform semantic analysis on the industrial control protocol messages, and the output semantic description sequence specifically includes: Extract the field features of the industrial control protocol messages and construct a protocol-level semantic query vector; Extract the equipment process stages and parameter constraints, align them with the protocol-level semantic query vector, and generate a fused semantic input sequence; The fused semantic input sequence is input into the protocol admission analysis model for cross-modal reasoning, and a semantic description sequence is output.
[0007] In this embodiment, the response identification of the semantic description sequence and the abnormal behavior records of the current device operation to obtain the network security situation of the current network area specifically includes: The protocol intent features and threat level data are extracted from the semantic description sequence, and a threat association graph is constructed by combining the temporal attack traces in the abnormal behavior records. Based on the threat association graph, calculate the similarity score between each abnormal behavior node and the known attack pattern, and mark the confirmed threat events; A comprehensive threat index is obtained by weighted summation of the topological propagation paths and impact range weights in the confirmed threat events. A threat level index is determined based on the comprehensive threat index. Attack frequency statistics are obtained by statistically analyzing the confirmed threat events. The threat level index and the attack frequency statistics are written into the situational data, and a network security situational data is generated by aggregating and generating a network security situational data that includes a security level identifier, situational description text, the threat level index, and the attack frequency statistics.
[0008] In this embodiment, the historical decision feedback information in the strategy decision-making model and the behavioral attack characteristics of the industrial control network in the current operating strategy are jointly extrapolated to obtain the access risk score of the current network area, specifically including: Extract strategy execution effect parameters and misjudgment rate data from the historical decision feedback information to construct a decision performance evaluation matrix; From the behavioral attack features detected under the current operating strategy, attack path sequences and target asset nodes are extracted to generate an attack impact surface topology map; By integrating the decision effectiveness evaluation matrix with the attack impact surface topology map, the access risk score of the current network region is obtained through weighted aggregation calculation.
[0009] In this embodiment, mapping the access risk score to the continuous response spectrum of the industrial control network from the detection of obstruction to obtain access control instructions specifically includes: The access risk score is mapped to the risk level range in the continuous response spectrum to determine the target response level that matches the access risk score. Based on the target response level, extract the corresponding control action sequence and execution timing constraints from the continuous response spectrum; Based on the control action sequence and the execution timing constraints, an access control instruction containing blocking strategies and access rules is generated.
[0010] In this embodiment, extracting the corresponding control action sequence and execution timing constraints from the continuous response spectrum based on the target response level specifically includes: Retrieve a set of candidate control actions associated with the target response level from the continuous response spectrum, and extract the control action identifier for each control action; The candidate control action set is sorted according to the priority field and dependency field contained in the control action identifier to generate a control action sequence containing execution order and dependency relationship; Based on the dependencies between adjacent actions in the control action sequence, and combined with the standard execution time of each control action, network status confirmation delay, and minimum safe time interval, the execution interval threshold is calculated to generate execution timing constraints.
[0011] In this embodiment, updating the network access parameters of the current device in conjunction with the network security situation and the access control command specifically includes: The threat level index and attack frequency statistics of the current network area are extracted from the network security situation to generate a situation quantification feature vector; Based on the blocking strategy and allowing rules in the access control command, extract the access permission adjustment parameters corresponding to the current device; The situation quantization feature vector is associated and mapped with the access permission adjustment parameters to generate updated network access parameters.
[0012] In this embodiment, distributing the updated network admission parameters to each edge analysis node specifically includes: The node identifier and network topology location of each edge analysis node are parsed to match the target route distribution path for the updated network admission parameters; Based on the target routing distribution path, a secure transmission channel is constructed with each edge analysis node, and the updated network access parameters are processed by fragmented encryption. The encrypted network access parameters are sent to each edge analysis node through the secure transmission channel, triggering each node to synchronously update the access policy.
[0013] Secondly, this application provides a computer device, the computer device including a memory and a processor, the memory for storing a computer program, and the processor for calling and running the computer program from the memory, so that the computer device executes the above-mentioned adaptive admission method for a distributed industrial control protocol based on an AI large model.
[0014] Thirdly, this application provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the aforementioned adaptive admission method for a distributed industrial control protocol based on a large AI model.
[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: The system collects network traffic from the industrial control network area and parses the industrial control protocol messages on the edge analysis nodes. It then uses a locally deployed protocol access control analysis model to perform semantic analysis on the industrial control protocol messages, outputting a semantic description sequence. The system responds to and identifies the semantic description sequence and abnormal behavior records of the current device operation to obtain the network security situation of the current network area. It also performs collaborative deduction of historical decision feedback information from the policy decision model and the behavioral attack characteristics of the industrial control network in the current operating strategy to obtain the access risk score of the current network area. This access risk score is mapped to the continuous response spectrum of the industrial control network from detection to blocking to obtain access control instructions. Based on the network security situation and the access control instructions, the system updates the network access parameters of the current device and distributes the updated network access parameters to each edge analysis node.
[0016] Therefore, this application demonstrates that the network access parameters of the current device can be updated in conjunction with the network security situation and the access control command, and the updated network access parameters can be distributed to each edge analysis node. Determining the network security situation allows for the acquisition of structured and quantitative characteristics reflecting the real-time threat intensity and attack distribution of the current network area, thereby overcoming the shortcomings of existing access control mechanisms that rely solely on discrete threat level labels, resulting in coarse control granularity and delayed response. This feature transforms the macro-security environment into a continuous variable that can participate in numerical calculations, enabling the adjustment of access parameters to accurately perceive subtle changes in threat intensity, avoiding insufficient protection during high-risk periods or excessive rate limiting during low-risk periods. This achieves an adaptive and fine balance between security protection strength and business availability in a dynamically changing industrial control network environment, significantly improving overall protection effectiveness and production continuity. Determining the access control command provides a set of permission adjustment parameters precisely bound to the current device and carrying an effective priority, thereby solving the problems of parameter mismatch, redundant transmission, and inconsistent node states caused by traditional global broadcast policy distribution. This parameter transforms the fine-grained global response decision into a single-device-level executable configuration, ensuring that the access permission changes of each edge analytics node strictly follow the overall security intent. At the same time, it provides accurate baseline input for subsequent dynamic corrections based on situational characteristics, thereby significantly reducing network bandwidth consumption and improving the timeliness of access policy synchronization updates while ensuring policy consistency.
[0017] In summary, the technical solution adopted in this application can dynamically and adaptively adjust network access control parameters to improve the timeliness of balancing security protection strength and service availability. Attached Figure Description
[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0019] Figure 1 This is an exemplary flowchart of an adaptive admission method for a distributed industrial control protocol based on an AI large model, provided in this application. Figure 2 This is a flowchart illustrating the process of determining access control instructions provided in this application; Figure 3 This is a schematic diagram of the structure of a computer device according to an adaptive admission method for a distributed industrial control protocol based on an AI large model, as provided in this application. Detailed Implementation
[0020] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.
[0021] This application provides a distributed industrial control protocol adaptive admission method based on an AI large model. Its core is to collect network traffic in an industrial control network area and parse the industrial control protocol messages on edge analysis nodes; call a locally deployed protocol admission analysis model to perform semantic analysis on the industrial control protocol messages, outputting a semantic description sequence; and identify responses to the semantic description sequence and abnormal behavior records of the current device operation to obtain the network security situation of the current network area; collaboratively extrapolate the historical decision feedback information in the policy decision model and the behavioral attack characteristics of the industrial control network in the current operating strategy to obtain the admission risk score of the current network area; map the admission risk score to the continuous response spectrum of the industrial control network from detection to blocking to obtain admission control instructions; and update the network admission parameters of the current device in conjunction with the network security situation and the admission control instructions, and distribute the updated network admission parameters to each edge analysis node.
[0022] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is an exemplary flowchart of an adaptive admission method for a distributed industrial control protocol based on an AI large model, according to this embodiment of the application. The method includes the following steps: In step S1, network traffic in the industrial control network area is collected, and industrial control protocol messages on the edge analysis nodes are parsed out.
[0023] In practice, the process begins with deploying traffic acquisition probes at critical communication links or core switching devices in the industrial control network. These probes capture all network data packets losslessly using port mirroring or network splitting mechanisms, and then transmit the raw data stream, with added timestamps and node identifiers, to the edge analysis node. Upon receiving this data stream, the edge analysis node sequentially strips the data link layer and network layer encapsulation headers, extracts the transport layer port number and payload signature, and matches them against a pre-built industrial control protocol fingerprint database to accurately identify the industrial control protocol type to which the data packet belongs. Subsequently, based on the protocol structure specifications determined by the matching, the application layer payload undergoes deep, field-by-field parsing to extract core control fields such as function codes, register addresses, and operational data values. These are then reassembled and encapsulated into a structured industrial control protocol message.
[0024] It should be noted that, in this application, network traffic refers to the complete set of network data packets transmitted in the industrial control network communication link, which carries equipment status data and control interaction commands; edge analysis node refers to a distributed processing unit deployed within the physical boundary or subdomain of the industrial control network, which has the ability to collect local data in real time, parse protocols, and perform edge computing; and industrial control protocol message refers to a structured data transmission unit that carries equipment status monitoring data and control command information.
[0025] In step S2, the locally deployed protocol access analysis model is invoked to perform semantic analysis on the industrial control protocol messages, output a semantic description sequence, and respond to and identify the semantic description sequence and the abnormal behavior records of the current device during operation to obtain the network security situation of the current network area.
[0026] In this embodiment, the semantic analysis of the industrial control protocol messages is performed by calling the locally deployed protocol admission analysis model, and the semantic description sequence is output. This can be achieved by the following steps: Extract the field features of the industrial control protocol messages and construct a protocol-level semantic query vector; Extract the equipment process stages and parameter constraints, align them with the protocol-level semantic query vector, and generate a fused semantic input sequence; The fused semantic input sequence is input into the protocol admission analysis model for cross-modal reasoning, and a semantic description sequence is output.
[0027] In practice, the edge analysis node first performs field-level structured extraction on the parsed industrial control protocol message to obtain key field features such as function codes, register addresses, data lengths, and payload values. Each field feature is then converted into a fixed-dimensional numerical field embedding representation according to a pre-defined protocol field semantic mapping table. This mapping table is trained based on the statistical distribution of field values and semantic relationships from historical normal industrial control communication samples. Subsequently, a self-attention mechanism is used to weighted aggregate the field embedding representations, calculating the semantic dependency weights between fields to generate a protocol-level semantic query vector representing the overall operational intent of the current message. Next, the edge analysis node obtains the current device's process stage identifier and parameter constraint information from the local device status monitoring module. The process stage identifier is obtained by decoding the device's operating status time-series data using a Hidden Markov Model (HMM). The model's state transition probability matrix is trained based on historical normal operating condition samples of the device, outputting the most likely process stage category at the current moment. The parameter constraint information includes the maximum allowed communication frequency, the set of legal function codes, and the register read / write range under the current process stage, retrieved from a pre-defined process-parameter constraint configuration table by indexing the process stage identifier. The process stage identifier and parameter constraint information are encoded into process embedding vectors and constraint embedding vectors, respectively. A cross-attention mechanism is used to align and fuse these vectors with the protocol-level semantic query vector as the query key and the process embedding vector and constraint embedding vector as the query keys, generating a fused semantic input sequence. Finally, this fused semantic input sequence is input into a locally deployed protocol admission analysis model, which employs an encoder-decoder structure based on the Transformer architecture. The encoder layer performs multi-layer self-attention computation and feedforward network transformation on the fused semantic input sequence to extract a joint high-order feature representation of protocol semantics and equipment operating conditions. The decoder layer uses this joint high-order feature representation as conditional input and generates semantic labels step-by-step through a masked self-attention mechanism. The output of each time step is normalized by softmax, and the label with the highest probability is selected as the semantic description unit for the current time step. The semantic description units output from all time steps are sequentially concatenated to form a semantic description sequence.
[0028] It should be noted that, in this application, the protocol access analysis model refers to a localized deep learning inference model that performs semantic parsing on industrial control protocol messages and outputs compliance semantic descriptions; semantic analysis refers to a structured parsing process that identifies operational intent and annotates compliance semantics in industrial control protocol messages; the semantic description sequence refers to a set of protocol semantic annotations arranged in chronological order, containing operation types and compliance judgment results; field features refer to structured protocol attribute information such as function codes, register addresses, and payload data extracted from industrial control protocol messages; the protocol-level semantic query vector refers to a fixed-dimensional continuous dense semantic embedding representation of the overall operational intent of the industrial control protocol message; the equipment process stage refers to the category identifier of the current production process determined by decoding the timing data of the equipment's operating status; parameter constraints refer to the limiting conditions for the communication frequency, legal function codes, and register operation range allowed by the equipment in the current process stage; the fused semantic input sequence refers to the joint semantic representation sequence formed by aligning and fusing the protocol semantic vector and the equipment operating condition embedding vector; and cross-modal reasoning refers to a deep learning inference operation process that jointly encodes and decodes two types of heterogeneous information: protocol semantics and equipment operating conditions.
[0029] In this embodiment, the network security posture of the current network area can be obtained by responding to and identifying the semantic description sequence and the abnormal behavior records of the current device during operation through the following steps: The protocol intent features and threat level data are extracted from the semantic description sequence, and a threat association graph is constructed by combining the temporal attack traces in the abnormal behavior records. Based on the threat association graph, calculate the similarity score between each abnormal behavior node and the known attack pattern, and mark the confirmed threat events; A comprehensive threat index is obtained by weighted summation of the topological propagation paths and impact range weights in the confirmed threat events. A threat level index is determined based on the comprehensive threat index. Attack frequency statistics are obtained by statistically analyzing the confirmed threat events. The threat level index and the attack frequency statistics are written into the situational data, and a network security situational data is generated by aggregating and generating a network security situational data that includes a security level identifier, situational description text, the threat level index, and the attack frequency statistics.
[0030] In practice, the edge analysis node first performs structured parsing on the semantic description sequence, extracting protocol intent features representing operation types and target objects, and reading the threat level data embedded in each semantic annotation. Simultaneously, it extracts the timestamps, durations, and indicator offsets of each abnormal event from the abnormal behavior records, forming a temporal attack trace. Using protocol intent features and temporal attack traces as node entities and threat level data as edge weights, a time window alignment mechanism is used to associate protocol semantic nodes and abnormal behavior nodes within the same window, constructing a threat association graph. Then, the edge analysis node loads a pre-built library of known attack patterns, containing standard attack subgraph templates verified through historical attack and defense exercises and public threat intelligence. For each abnormal behavior node in the threat association graph, its local neighborhood subgraph structure is extracted, and a graph edit distance algorithm is used to calculate the structural similarity score between the subgraph and each known attack pattern template. The node replacement cost and edge replacement cost of the graph edit distance are dynamically weighted based on the threat level data, with values ranging from zero to one. When the highest similarity score corresponding to a node exhibiting abnormal behavior exceeds a preset threshold, the node and its associated protocol intent features are identified as confirmed threat events. Finally, edge analysis nodes extract the topology propagation paths of each confirmed threat event from the threat association graph. These paths are a set of directed paths extending from the initial abnormal behavior node along associated edges to all reachable nodes. For each topology propagation path, an impact range weight is assigned based on the asset importance and business relevance of the devices corresponding to each node on the path. The asset importance and business relevance are jointly determined by the industrial control network asset ledger and the process flow dependency table. The topology propagation path lengths, the number of nodes involved, and the corresponding impact range weights of all confirmed threat events are weighted and summed to obtain the comprehensive threat index for the current network area. The comprehensive threat index is compared with a preset situational awareness mapping table to output a network security situational awareness that includes a security level identifier and situational description text.
[0031] It should be noted that, in this application, abnormal behavior records refer to the recorded data formed by time-series monitoring and log storage of operational events that deviate from the normal baseline of equipment operation; response identification refers to the structured processing process of performing time-series correlation analysis on protocol semantics and equipment anomalies to determine the causal relationship of threats; protocol intent features refer to the structured semantic attributes extracted from the semantic description sequence that characterize the type of industrial control protocol operation and the target object; threat level data refers to the numerical annotation information embedded in the semantic description sequence that quantifies the potential harm of protocol operations; time-series attack traces refer to the time-series abnormal features extracted from abnormal behavior records that include timestamps, duration, and indicator offsets; threat association graph refers to the structured threat relationship topology representation with protocol semantics and abnormal behavior as nodes and threat level as edge weights; abnormal behavior nodes refer to the nodes in the threat association graph that characterize a single The graph structure of anomaly events and their associated context information is defined as follows: Known attack patterns refer to a set of pre-defined threat pattern templates that have been historically verified and include standard attack subgraph structures and node attributes; similarity score refers to the structural matching degree between the anomalous behavior subgraph calculated using a graph editing distance algorithm and the known attack pattern templates; confirmed threat events refer to the set of labeled anomalous behavior nodes and their associated protocol intent features whose similarity scores exceed a preset threshold; topology propagation path refers to the set of directed paths in the threat association graph that extend from the initial anomalous node along associated edges to all reachable nodes; impact range weight refers to a numerical parameter that quantifies the business impact of threat events, allocated based on the importance of equipment assets and their business relevance; network security posture refers to the network status assessment result, including security level and descriptive text, output after mapping the comprehensive threat index and posture level.
[0032] In step S3, the historical decision feedback information in the strategy decision model and the behavioral attack characteristics of the industrial control network in the current operating strategy are jointly deduced to obtain the access risk score of the current network area. The access risk score is then mapped to the continuous response spectrum of the industrial control network from detection to blocking to obtain access control instructions.
[0033] In this embodiment, the access risk score of the current network area can be obtained by co-deducing the historical decision feedback information in the strategy decision model and the behavioral attack characteristics of the industrial control network in the current operating strategy: Extract strategy execution effect parameters and misjudgment rate data from the historical decision feedback information to construct a decision performance evaluation matrix; From the behavioral attack features detected under the current operating strategy, attack path sequences and target asset nodes are extracted to generate an attack impact surface topology map; By integrating the decision effectiveness evaluation matrix with the attack impact surface topology map, the access risk score of the current network region is obtained through weighted aggregation calculation.
[0034] In practice, the edge analysis node first extracts policy execution effect parameters and false positive rate data from the historical decision logs of the local Strategy Decision Model for past decision cycles. Policy execution effect parameters include threat interception success rate after the issuance of access control commands, service interruption duration, and device communication recovery time. These parameters are obtained from the network status monitoring module and device operation logs, indexed by decision timestamp. False positive rate data is calculated by statistically analyzing the proportion of decisions manually reviewed and marked as false positives or false negatives to the total number of decisions made during the same period. The policy execution effect parameters are arranged by column, and the false positive rate data by row, to construct a decision performance evaluation matrix. Then, the edge analysis node performs structured analysis on the behavioral attack characteristics detected by the industrial control network under the current operating strategy, extracting the attack path sequence and target asset nodes. The attack path sequence is an ordered address chain consisting of the source address, relay address, and destination address that the attack behavior sequentially passes through in the industrial control network, obtained hop-by-hop from the source and destination address fields of the industrial control protocol messages and routing records. The target asset node is identified as the industrial control equipment or critical business system pointed to by the end of the attack path sequence, and is determined by address index matching from the industrial control network asset ledger. Using the target asset node as the core node and each hop address in the attack path sequence as the edge node, a directed topology structure is constructed based on the actual network physical connection relationship, forming an attack impact surface topology map. Finally, the edge analysis nodes perform cross-modal fusion calculations between the decision effectiveness evaluation matrix and the attack impact surface topology map. First, a graph convolutional network is used to extract features from the attack impact surface topology map. The graph convolutional network aggregates the topological features of each node and its neighboring nodes through two layers of graph convolution operations, outputting a topological feature vector with the same column dimensions as the decision effectiveness evaluation matrix. Then, the decision effectiveness evaluation matrix is flattened row-wise into an effectiveness feature vector. The effectiveness feature vector is concatenated with the topological feature vector and input into a multilayer perceptron for weighted aggregation calculation. The multilayer perceptron contains two hidden layers, using a modified linear unit (MLU) as the activation function. The output layer uses a sigmoid activation function to normalize the calculation results to the zero-to-one interval, and the output value is the access risk score for the current network region.
[0035] It should be noted that, in this application, the strategy decision-making model refers to an intelligent decision-making model that performs inference calculations based on historical decision data and current network behavior characteristics and outputs access risk assessment results; historical decision feedback information refers to the set of decision results generated by the strategy decision-making model in past operating cycles and their corresponding actual execution effect assessment data; behavioral attack characteristics refer to the set of behavioral identifiers of the industrial control network that match known attack patterns under the current operating strategy; collaborative inference refers to the joint inference process of performing cross-modal fusion calculations on historical decision effectiveness information and current attack characteristics to output risk assessment results; strategy execution effect parameters refer to the set of multi-dimensional numerical indicators that quantify the threat interception success rate and business impact after the access control command is issued; and false positive rate data refers to... The statistical data on the proportion of decisions marked as false alarms or omissions in historical decision-making to the total number of decisions during the same period; the decision effectiveness evaluation matrix refers to a structured numerical matrix formed by arranging strategy execution effect parameters and false alarm rate data in rows and columns and then normalizing them; the attack path sequence refers to an ordered address chain consisting of the source address, relay address, and destination address that the attack behavior passes through in the industrial control network in sequence; the target asset node refers to the unique identifier of the industrial control equipment or critical business system pointed to by the end of the attack path sequence; the attack impact surface topology diagram refers to the directed network topology structure constructed with the target asset node as the core and the hop addresses of each hop of the attack path as edge nodes; the access risk score refers to the value between zero and one representing the degree of illegal access threat faced by the current network area.
[0036] Preferably, in this embodiment, the access risk score is mapped to the continuous response spectrum of the industrial control network from the detection of obstruction to obtain access control instructions, with reference to... Figure 2 As shown in the figure, this is a flowchart illustrating the process of determining access control instructions in some embodiments of this application. In this embodiment, determining access control instructions can be achieved through the following steps: In step S31, the admission risk score is mapped to the risk level range in the continuous response spectrum to determine the target response level that matches the admission risk score; In step S32, the corresponding control action sequence and execution timing constraints are extracted from the continuous response spectrum according to the target response level; In step S33, an access control instruction containing blocking strategies and access rules is generated based on the control action sequence and the execution timing constraints.
[0037] In practice, the edge analysis node first loads a pre-defined continuous response spectrum configuration table. This spectrum divides the zero-to-one numerical space into multiple continuous and non-overlapping risk level intervals, each corresponding to a unique target response level. The interval boundary values are determined by analyzing historical admission risk score samples and corresponding actual treatment effects using subject operating characteristic curves, selecting the optimal cutoff point between each level that combines the false positive and false negative rates. The currently calculated admission risk score is compared one by one with the upper and lower thresholds of each risk level interval. When the score falls within a certain interval, the response level identifier bound to that interval is output as the target response level. If the score is exactly equal to the interval boundary value, it is assigned to the response level corresponding to the higher risk level according to a pre-defined rounding rule. Then, the edge analysis node uses the target response level as the index key to retrieve the control action sequence and execution timing constraints bound to that level from the continuous response spectrum configuration table. The control action sequence is an ordered list of atomic security operations that must be executed sequentially at this response level, including operation types and parameter configurations such as traffic rate limiting, protocol field filtering, session reset, and port closure. Execution timing constraints define the minimum interval, maximum allowable delay, and timeout fallback strategy between each atomic operation. These constraint parameters are statistically calibrated based on the real-time requirements of industrial control networks and measured data of equipment communication cycles to ensure that control actions are executed without interfering with normal production rhythms. Finally, the edge analysis node encapsulates the control action sequence and execution timing constraints into instructions to generate access control instructions. The instruction body contains two parts: blocking policies and allowing rules. The blocking policies consist of filtering, rate limiting, and session termination operations for illegal traffic in the control action sequence, arranged into executable instruction blocks according to the order and intervals defined in the execution timing constraints. The allowing rules consist of whitelist matching, protocol verification, and rate guarantee operations for legitimate traffic in the control action sequence, also arranged according to timing constraints. The command header includes a target response level identifier, a generation timestamp, and a command validity period field, while the tail includes a digital signature to ensure command integrity and source credibility. The generated access control command is sent to the execution device at the industrial control network boundary through a secure channel.
[0038] It should be noted that, in this application, the continuous response spectrum refers to a continuous strategy configuration system that divides the access risk score space into multiple level intervals and binds corresponding response measures; the risk level interval refers to the range of access risk scores uniquely bound to a specific target response level within the continuous response spectrum; the target response level refers to the structured hierarchical identifier that characterizes the required response intensity after the access risk score is mapped through intervals; the control action sequence refers to the ordered list of atomic security operations and their parameter configurations that need to be executed sequentially under the target response level; the execution timing constraints refer to the set of timing limit parameters for the minimum interval, maximum delay, and timeout rollback between each operation in the control action sequence; the blocking strategy refers to the set of strategy rules in the access control command that perform filtering, rate limiting, and session termination operations on illegal traffic; the allowing rule refers to the set of strategy rules in the access control command that perform whitelist matching, protocol verification, and rate guarantee on legitimate traffic; and the access control command refers to the executable security policy distribution data unit that encapsulates the blocking strategy and the allowing rule and attaches timing constraints.
[0039] Furthermore, in this embodiment, extracting the corresponding control action sequence and execution timing constraints from the continuous response spectrum based on the target response level can be achieved using the following steps: Retrieve a set of candidate control actions associated with the target response level from the continuous response spectrum, and extract the control action identifier for each control action; The candidate control action set is sorted according to the priority field and dependency field contained in the control action identifier to generate a control action sequence containing execution order and dependency relationship; Based on the dependencies between adjacent actions in the control action sequence, and combined with the standard execution time of each control action, network status confirmation delay, and minimum safe time interval, the execution interval threshold is calculated to generate execution timing constraints.
[0040] In practice, the edge analysis node first receives the target response level and then accesses a locally stored continuous response hierarchy configuration library. This library uses a key-value pair structure to store the mapping relationship between each response level and available security measures. The node uses the target response level as the query key to retrieve all bound security measures, forming a candidate control action set. It then traverses this set, reading the pre-assigned control action identifier for each control action. Next, the edge analysis node parses the priority and dependency fields in each control action identifier. The dependency field indicates the prerequisite action identifier that must be completed before the current action can be executed. The node uses a topological sorting algorithm to traverse the candidate control action set in a directed acyclic graph, placing actions without prerequisite dependencies at the beginning of the sequence and arranging subsequent actions according to their priority values from smallest to largest, generating a control action sequence. Finally, the edge analysis node traverses the control action sequence and extracts the dependency types between adjacent actions. Dependency types are divided into strong dependencies and weak dependencies. Strong dependencies require the status of the prerequisite action to be confirmed before the subsequent action can be executed, while weak dependencies only require a fixed time delay. The node reads the preset standard execution time for each action. For strong dependencies, it adds the standard execution time of the preceding action to the network status confirmation delay to obtain the execution interval threshold. For weak dependencies, it takes the preset minimum safe time interval as the execution interval threshold. All execution interval thresholds are encapsulated as execution timing constraints and output.
[0041] It should be noted that, in this application, the candidate control action set refers to an unordered set of basic security operation measures that are associated with a specific target response level and can be combined and invoked; the control action identifier refers to a structured code used to uniquely distinguish control actions and to carry their type, scope, priority, and dependencies; the control action sequence refers to an ordered list of security operations that has a strict linear execution order and conflict-free dependencies after topological sorting; and the execution interval threshold refers to the minimum waiting time parameter between adjacent actions in the control action sequence, calculated based on dependency type and execution time.
[0042] In step S4, the network access parameters of the current device are updated in conjunction with the network security situation and the access control command, and the updated network access parameters are distributed to each edge analysis node.
[0043] In this embodiment, updating the network access parameters of the current device in conjunction with the network security situation and the access control command can be achieved through the following steps: The threat level index and attack frequency statistics of the current network area are extracted from the network security situation to generate a situation quantification feature vector; Based on the blocking strategy and allowing rules in the access control command, extract the access permission adjustment parameters corresponding to the current device; The situation quantization feature vector is associated and mapped with the access permission adjustment parameters to generate updated network access parameters.
[0044] In practice, firstly, the edge analysis node parses the received network security situation data and extracts a threat level indicator that represents the overall danger level of the current network area. This indicator is a continuous value normalized to the range of zero to one, output by the preceding collaborative inference module. Simultaneously, it reads the attack frequency statistics carried in the situation data, which counts the total number of confirmed threat events detected in the most recent assessment period and their temporal distribution density. The threat level indicator is used as the first dimension, and the attack frequency statistics, after logarithmic normalization, are used as the second dimension; these are concatenated to form a two-dimensional situation quantification feature vector. Next, the edge analysis node parses the blocking policies and allowing rules contained in the access control instructions, performs matching and filtering based on the target device identifier, and selects policy entries directly related to the current device. From the matched blocking policies, it extracts restriction adjustment items, including the maximum allowed concurrent connections, protocol field verification strictness level, and traffic rate limit; from the matched allowing rules, it extracts permission adjustment items, including whitelist address ranges, exempt protocol types, and priority bandwidth quotas. These restriction and permission adjustment items are then assembled into access permission adjustment parameters according to a preset parameter template. Finally, the edge analysis node inputs the situation quantification feature vector and access permission adjustment parameters into a preset parameter mapping model. This model is an adaptive regulator based on multiple linear regression, using the situation quantification feature vector as the independent variable and the fields of the access permission adjustment parameters as the dependent coefficients to calculate the dynamic correction factor for each access parameter. The correction factor ranges from 0.5 to 1.5. When the threat level index increases, the correction factor for restrictive parameters approaches 1.5 to tighten access conditions, while the correction factor for permissive parameters approaches 0.5 to reduce the range of permissions granted. The original access permission adjustment parameters are multiplied by their corresponding correction factors and rounded to generate updated network access parameters, which are then written to the access control module configuration register of the current device. The coefficient matrix of the regression model is obtained through offline training using historical situation data and optimal parameter configuration samples.
[0045] It should be noted that, in this application, the threat level index refers to a normalized continuous value that quantitatively represents the overall danger level of the current network area in the network security situation; the attack frequency statistics refer to the statistics of the total number of confirmed threat events detected in the most recent assessment period and their time distribution density; the situation quantification feature vector refers to a structured feature representation used for numerical calculation, which is composed of the threat level index and the normalized attack frequency; the access permission adjustment parameters refer to the set of restriction and permission configuration adjustment items extracted from the access control instructions and bound to the current device; and the network access parameters refer to the set of final effective access control configuration values written into the device configuration register after dynamic correction of the situation features.
[0046] In this embodiment, the updated network admission parameters can be distributed to each edge analysis node using the following steps: The node identifier and network topology location of each edge analysis node are parsed to match the target route distribution path for the updated network admission parameters; Based on the target routing distribution path, a secure transmission channel is constructed with each edge analysis node, and the updated network access parameters are processed by fragmented encryption. The encrypted network access parameters are sent to each edge analysis node through the secure transmission channel, triggering each node to synchronously update the access policy.
[0047] In practice, the central control platform first reads the industrial control network asset ledger and parses the node identifier and network topology location of each edge analysis node. The node identifier is a globally unique device code, and the network topology location includes the subnet address to which the node belongs, the port number of the upstream aggregation switch, and the physical layer depth. Using the target node identifier list carried in the updated network access parameters as an index, the platform executes a shortest path search algorithm in the pre-stored topology adjacency matrix to calculate the hop-by-hop forwarding sequence from the central control platform to each target edge analysis node. When multiple equivalent paths exist, the link with the lowest historical average transmission delay is selected as the target routing distribution path. Then, based on the address information of each hop node in the target routing distribution path, the central control platform negotiates with the target edge analysis node to establish a secure transmission channel. The channel construction adopts a transport layer security protocol handshake process, verifying the digital certificates of both parties and generating a session key during the negotiation. Subsequently, the updated network access parameters are fragmented and encrypted at fixed lengths. Each data block is encrypted using the session key and an advanced encryption standard algorithm, with a sequence number and message authentication code appended to the beginning of each fragment. The fragment length is determined by subtracting the protocol overhead from the maximum transmission unit of the smallest link in the target routing distribution path. Finally, the central control platform sends the encrypted fragmented network access parameters to each edge analysis node sequentially via a secure transmission channel. Upon receiving a complete data fragment, each edge analysis node decrypts it using its locally stored session key and verifies the message authentication code. After all verifications pass, the nodes reassemble the parameters according to their sequence numbers to restore the complete updated network access parameters. The nodes write the restored parameters to their local access policy configuration area, overwriting the original corresponding fields, and record the version number and effective timestamp of this update. After the configuration is written, the nodes return a confirmation message containing the update result status to the central control platform, completing the closed loop of access policy synchronization update.
[0048] It should be noted that, in this application, node identifier refers to the device code used to globally and uniquely identify edge analysis nodes in an industrial control network; network topology location refers to the spatial location information of the edge analysis node in the network, including its subnet, upstream port, and physical layer; target route distribution path refers to the hop-by-hop forwarding sequence determined by the shortest path search from the central management platform to the target edge analysis node; secure transmission channel refers to the encrypted communication link established between the central management platform and the edge analysis node through certificate verification and session key negotiation; fragmented encryption processing refers to the operation of fragmenting admission parameters according to link transmission units, encrypting them with session keys, and adding integrity verification; and admission policy synchronization update refers to the configuration refresh process in which the edge analysis node receives and applies new admission parameters to achieve network-wide policy consistency.
[0049] Therefore, this application demonstrates that the network access parameters of the current device can be updated in conjunction with the network security situation and the access control command, and the updated network access parameters can be distributed to each edge analysis node. Determining the network security situation allows for the acquisition of structured and quantitative characteristics reflecting the real-time threat intensity and attack distribution of the current network area, thereby overcoming the shortcomings of existing access control mechanisms that rely solely on discrete threat level labels, resulting in coarse control granularity and delayed response. This feature transforms the macro-security environment into a continuous variable that can participate in numerical calculations, enabling the adjustment of access parameters to accurately perceive subtle changes in threat intensity, avoiding insufficient protection during high-risk periods or excessive rate limiting during low-risk periods. This achieves an adaptive and fine balance between security protection strength and business availability in a dynamically changing industrial control network environment, significantly improving overall protection effectiveness and production continuity. Determining the access control command provides a set of permission adjustment parameters precisely bound to the current device and carrying an effective priority, thereby solving the problems of parameter mismatch, redundant transmission, and inconsistent node states caused by traditional global broadcast policy distribution. This parameter transforms the fine-grained global response decision into a single-device-level executable configuration, ensuring that the access permission changes of each edge analytics node strictly follow the overall security intent. At the same time, it provides accurate baseline input for subsequent dynamic corrections based on situational characteristics, thereby significantly reducing network bandwidth consumption and improving the timeliness of access policy synchronization updates while ensuring policy consistency.
[0050] In summary, the technical solution adopted in this application can dynamically and adaptively adjust network access control parameters to improve the timeliness of balancing security protection strength and service availability.
[0051] The foregoing detailed an example of an adaptive admission method for a distributed industrial control protocol based on a large AI model, as provided in this application. It is understood that the corresponding apparatus, in order to achieve the above functions, includes hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.
[0052] In some embodiments, this application also provides a computer device, the computer device including a memory and a processor, the memory for storing a computer program, and the processor for calling and running the computer program from the memory, so that the computer device executes the above-described adaptive admission method for a distributed industrial control protocol based on an AI large model.
[0053] In some embodiments, reference Figure 3 The dashed lines in the figure indicate that the unit or module is optional. This figure is a structural schematic diagram of a computer device implementing an AI-based large-scale model-based adaptive admission method for distributed industrial control protocols, according to an embodiment of this application. The AI-based large-scale model-based adaptive admission method for distributed industrial control protocols described in the above embodiment can... Figure 3 The computer device shown is used to implement this, and the computer device includes at least one processor 301, a memory 302 and at least one communication unit 305. The computer device may be a terminal device, a server or a chip.
[0054] Processor 301 can be a general-purpose processor or a special-purpose processor. For example, processor 301 can be a central processing unit (CPU), which can be used to control computer devices, execute software programs, and process data from software programs. The computer device may also include a communication unit 305 for inputting (receiving) and outputting (transmitting) signals.
[0055] For example, the computer device may be a chip, and the communication unit 305 may be the input and / or output circuit of the chip, or the communication unit 305 may be the communication interface of the chip, which may be a component of a terminal device, network device or other device.
[0056] For example, the computer device may be a terminal device or a server, and the communication unit 305 may be a transceiver of the terminal device or the server, or the communication unit 305 may be a transceiver circuit of the terminal device or the server.
[0057] The computer device may include one or more memories 302 storing a program 304. The program 304 can be executed by a processor 301 to generate instructions 303, causing the processor 301 to execute the method described in the above method embodiments according to the instructions 303. Optionally, the memory 302 may also store data (such as a target audit model). Optionally, the processor 301 may also read data stored in the memory 302, which may be stored at the same storage address as the program 304, or it may be stored at a different storage address than the program 304.
[0058] The processor 301 and memory 302 can be configured separately or integrated together, for example, integrated on the system on chip (SOC) of the terminal device.
[0059] It should be understood that each step of the above method embodiment can be completed by hardware logic circuits or software instructions in the processor 301. The processor 301 can be a CPU, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, such as discrete gate, transistor logic devices, or discrete hardware components.
[0060] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0061] For example, in some embodiments, this application also provides a computer-readable storage medium storing instructions or code that, when executed on a computer, cause the computer to implement the above-described adaptive admission method for a distributed industrial control protocol based on a large AI model.
[0062] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.
[0063] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
Claims
1. An adaptive admission method for distributed industrial control protocols based on a large AI model, characterized in that, The method includes: Collect network traffic in the industrial control network area and parse the industrial control protocol messages on the edge analysis nodes; The locally deployed protocol access control analysis model is invoked to perform semantic analysis on the industrial control protocol messages, outputting a semantic description sequence. The semantic description sequence and the abnormal behavior records of the current device operation are then used to identify the response and obtain the network security situation of the current network area. By co-evaluating the historical decision feedback information in the strategy decision model and the behavioral attack characteristics of the industrial control network in the current operation strategy, the access risk score of the current network area is obtained. The access risk score is then mapped to the continuous response spectrum of the industrial control network from detection to blocking to obtain access control instructions. Based on the network security situation and the access control instructions, the network access parameters of the current device are updated in conjunction with the network access control instructions, and the updated network access parameters are distributed to each edge analysis node.
2. The adaptive admission method for distributed industrial control protocols based on a large AI model as described in claim 1, characterized in that, The locally deployed protocol admission analysis model is invoked to perform semantic analysis on the industrial control protocol messages, and the output semantic description sequence specifically includes: Extract the field features of the industrial control protocol messages and construct a protocol-level semantic query vector; Extract the equipment process stages and parameter constraints, align them with the protocol-level semantic query vector, and generate a fused semantic input sequence; The fused semantic input sequence is input into the protocol admission analysis model for cross-modal reasoning, and a semantic description sequence is output.
3. The adaptive admission method for distributed industrial control protocols based on a large AI model as described in claim 1, characterized in that, The response identification of the semantic description sequence and the abnormal behavior records of the current device operation to obtain the network security situation of the current network area specifically includes: The protocol intent features and threat level data are extracted from the semantic description sequence, and a threat association graph is constructed by combining the temporal attack traces in the abnormal behavior records. Based on the threat association graph, calculate the similarity score between each abnormal behavior node and the known attack pattern, and mark the confirmed threat events; A comprehensive threat index is obtained by weighted summation of the topological propagation paths and impact range weights in the confirmed threat events. A threat level index is determined based on the comprehensive threat index. Attack frequency statistics are obtained by statistically analyzing the confirmed threat events. The threat level index and the attack frequency statistics are written into the situational data, and a network security situational data is generated by aggregating and generating a network security situational data that includes a security level identifier, situational description text, the threat level index, and the attack frequency statistics.
4. The adaptive admission method for distributed industrial control protocols based on a large AI model as described in claim 1, characterized in that, By jointly extrapolating historical decision feedback information from the strategy decision-making model and behavioral attack characteristics of the industrial control network under the current operating strategy, the access risk score of the current network area is obtained, specifically including: Extract strategy execution effect parameters and misjudgment rate data from the historical decision feedback information to construct a decision performance evaluation matrix; From the behavioral attack features detected under the current operating strategy, attack path sequences and target asset nodes are extracted to generate an attack impact surface topology map; By integrating the decision effectiveness evaluation matrix with the attack impact surface topology map, the access risk score of the current network region is obtained through weighted aggregation calculation.
5. The adaptive admission method for distributed industrial control protocols based on a large AI model as described in claim 1, characterized in that, Mapping the access risk score to the continuous response spectrum of the industrial control network from the detection of blockage, the specific access control instructions include: The access risk score is mapped to the risk level range in the continuous response spectrum to determine the target response level that matches the access risk score. Based on the target response level, extract the corresponding control action sequence and execution timing constraints from the continuous response spectrum; Based on the control action sequence and the execution timing constraints, an access control instruction containing blocking strategies and access rules is generated.
6. The adaptive admission method for distributed industrial control protocols based on a large AI model as described in claim 5, characterized in that, Extracting the corresponding control action sequence and execution timing constraints from the continuous response spectrum based on the target response level specifically includes: Retrieve a set of candidate control actions associated with the target response level from the continuous response spectrum, and extract the control action identifier for each control action; The candidate control action set is sorted according to the priority field and dependency field contained in the control action identifier to generate a control action sequence containing execution order and dependency relationship; Based on the dependencies between adjacent actions in the control action sequence, and combined with the standard execution time of each control action, network status confirmation delay, and minimum safe time interval, the execution interval threshold is calculated to generate execution timing constraints.
7. The adaptive admission method for distributed industrial control protocols based on a large AI model as described in claim 1, characterized in that, The specific steps for updating the network access parameters of the current device in conjunction with the network security situation and the access control instructions include: The threat level index and attack frequency statistics of the current network area are extracted from the network security situation to generate a situation quantification feature vector; Based on the blocking strategy and allowing rules in the access control command, extract the access permission adjustment parameters corresponding to the current device; The situation quantization feature vector is associated and mapped with the access permission adjustment parameters to generate updated network access parameters.
8. The adaptive admission method for distributed industrial control protocols based on a large AI model as described in claim 1, characterized in that, Distributing the updated network admission parameters to each edge analytics node specifically includes: The node identifier and network topology location of each edge analysis node are parsed to match the target route distribution path for the updated network admission parameters; Based on the target routing distribution path, a secure transmission channel is constructed with each edge analysis node, and the updated network access parameters are processed by fragmented encryption. The encrypted network access parameters are sent to each edge analysis node through the secure transmission channel, triggering each node to synchronously update the access policy.
9. A computer device, characterized in that, The computer device includes a memory and a processor. The memory is used to store computer programs, and the processor is used to call and run the computer programs from the memory, so that the computer device executes the adaptive admission method for a distributed industrial control protocol based on an AI large model as described in any one of claims 1 to 8.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores instructions or code that, when executed on a computer, cause the computer to implement the adaptive admission method for a distributed industrial control protocol based on an AI large model as described in any one of claims 1 to 8.
