AI-based asset identification and verification methods, devices, storage media, software products, and computer equipment

By using an AI-based asset identification and verification method, the asset characteristic data of power industrial control network equipment are compared and analyzed using artificial intelligence models. This solves the problems of accuracy and real-time performance in asset identification in complex networks, achieves efficient asset anomaly identification and verification, and improves the safety and stability of the power grid.

CN123069175APending Publication Date: 2026-11-10CHINA SOUTHERN POWER GRID COMPANY
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611094252.4
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-07-22
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing automated scanning tools struggle to accurately and quickly identify network assets in complex power control networks, resulting in insufficient accuracy and real-time performance in asset identification, which impacts the safety, stability, and reliability of the power grid.

Method used

An AI-based asset identification and verification method is adopted. By acquiring asset feature data of power industrial control network equipment, comparing and analyzing the data using an artificial intelligence model, asset feature matching information is generated. Based on a preset anomaly judgment priority, asset anomaly identification and verification are performed, and an asset verification report is generated.

Benefits of technology

It improves the accuracy and real-time performance of asset identification in complex power industrial control networks, enabling timely detection of high-risk asset anomalies, generation of accurate asset verification reports, and support for security policy configuration and fault location.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123069175A_ABST
    Figure CN123069175A_ABST
Patent Text Reader

Abstract

This application discloses an AI-based asset identification and verification method, device, storage medium, program product, and computer equipment. The method includes: acquiring asset feature data corresponding to power industrial control network equipment and determining asset feature information from the asset feature data, including traffic data; calling an AI-based power industrial control network asset anomaly identification model, comparing and analyzing the asset feature information with asset ledger benchmark information to generate asset feature matching information; at least based on the asset feature matching information, identifying asset anomalies in the power industrial control network where the equipment is located according to a preset anomaly judgment priority to obtain the asset anomaly type and the corresponding risk level of the power industrial control network; and generating an asset verification report based on the asset anomaly type and risk level. This improves the accuracy and real-time performance of asset identification in complex power industrial control networks.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of power industrial control network technology, and in particular to an AI-based asset identification and verification method, device, storage medium, program product and computer equipment. Background Technology

[0002] Power system control networks are core infrastructure supporting power system production control operations. The accuracy, integrity, and real-time performance of their network assets are crucial to the safety, stability, and reliability of power grid operation. Effective network asset identification and verification can improve the consistency between the equipment in the actual network and the asset ledger information registered in management systems such as CMDB (Configuration Management Database), thereby facilitating tasks such as security policy configuration, vulnerability management, fault location, and compliance auditing.

[0003] Among related technologies, automated scanning tools based on simple rules such as ICMP Ping (Internet Control Message Protocol Packet Internet Groper) and SNMP (Simple Network Management Protocol) polling can be used to complete network asset identification and verification.

[0004] However, the architecture of power industrial control networks is very complex. For example, there are often a large number of devices with different communication protocols, or even devices with proprietary extension protocols of power equipment manufacturers. The aforementioned automated scanning tools have difficulty accurately and quickly identifying relevant assets in complex networks. Summary of the Invention

[0005] To address the aforementioned technical issues, this application proposes an AI-based asset identification and verification method, apparatus, storage medium, program product, and computer equipment, which can improve the accuracy and real-time performance of asset identification in complex power industrial control networks.

[0006] In a first aspect, embodiments of this application provide an AI-based asset identification and verification method, including: Obtain asset characteristic data corresponding to power industrial control network equipment, and determine the asset characteristic information of the asset characteristic data, wherein the asset characteristic data includes traffic data; The system invokes an AI-based power industrial control network asset anomaly identification model, compares and analyzes the asset feature information with the asset ledger baseline information, and generates asset feature matching information. Based at least on the asset feature matching information, the power industrial control network where the power industrial control network equipment is located is identified as having asset anomalies according to a preset anomaly judgment priority, so as to obtain the asset anomaly type of the power industrial control network and the risk level corresponding to the asset anomaly type. An asset verification report is generated based on the described asset anomaly type and risk level.

[0007] Optionally, the asset characteristic information for determining the asset characteristic data includes: The continuous feature data in the asset feature data is discretized to obtain a discretized feature sequence, wherein the discretized feature sequence includes at least one dimension of discretized features. Determine the information entropy gain of each of the at least one dimension of the discretized features; Based on the information entropy gain, the discrete weights corresponding to the discrete features of each dimension are determined. Based on the discrete weights, the discrete features of at least one dimension are weighted, and the asset feature information is determined based on the weighted discrete features.

[0008] Optionally, the power industrial control network asset anomaly identification model includes a feature transformation unit and a feature attention layer; The process involves invoking an AI-based power control network asset anomaly identification model, comparing and analyzing the asset feature information with the asset ledger baseline information to generate asset feature matching information, including: The feature transformation unit is invoked to transform the asset feature information to obtain the transformed asset feature information. The feature attention layer is invoked to determine the first similarity information between the query vector and the key vector corresponding to the transformed asset feature information, and attention weight information is determined based on the first similarity information and the temperature adjustment factor, wherein the temperature adjustment factor is related to the equipment importance level of the power industrial control network equipment; Based on the attention weight information, the transformed asset feature information is weighted. The asset feature matching information is generated by comparing and analyzing the asset ledger baseline information with the weighted asset feature information.

[0009] Optionally, the step of comparing and analyzing the asset ledger baseline information with the weighted asset characteristic information to generate the asset characteristic matching information includes: The asset ledger baseline information is verified to extract valid asset ledger baseline information and the baseline feature information corresponding to the valid asset ledger baseline information is extracted. Determine the original similarity information between the baseline feature information and the weighted asset feature information; The original similarity information is normalized to obtain the asset feature matching information.

[0010] Optionally, the method further includes: Obtain the current load rate of the power industrial control network equipment and the cumulative number of core equipment anomalies in the most recent preset time period; Based on the current load rate, and the minimum load rate threshold and high load critical threshold corresponding to the power industrial control network equipment, determine the load rate influencing factor; An anomaly frequency impact factor is determined based on the ratio between the cumulative number of core anomalies of the equipment and the maximum core anomaly threshold, wherein the maximum core anomaly threshold is used to indicate the upper limit of the number of core anomalies of the power industrial control network equipment in the most recent preset time period. Based on the load rate influence factor, the abnormal frequency influence factor, and the previous period's dynamic threshold, determine the current period's dynamic threshold; The step of identifying asset anomalies in the power industrial control network where the power industrial control network equipment is located, based at least on the asset feature matching information and according to a preset anomaly judgment priority, includes: Based on the asset feature matching information and the current periodic dynamic threshold, the power industrial control network where the power industrial control network equipment is located is identified as having asset anomalies according to the preset anomaly judgment priority.

[0011] Optionally, the method further includes: The dynamic profile of power industrial control network assets is updated based on the continuous and / or discrete feature data in the asset feature data.

[0012] Secondly, embodiments of this application provide an AI-based asset identification and verification device, comprising: The feature processing module is used to acquire asset feature data corresponding to power industrial control network equipment and determine the asset feature information of the asset feature data, wherein the asset feature data includes traffic data; The matching module is used to call the AI-based power industrial control network asset anomaly identification model, compare and analyze the asset feature information with the asset ledger benchmark information, and generate asset feature matching information. An anomaly identification module is used to identify asset anomalies in the power industrial control network where the power industrial control network equipment is located, based at least on the asset feature matching information and according to a preset anomaly judgment priority, so as to obtain the asset anomaly type of the power industrial control network and the risk level corresponding to the asset anomaly type. The report generation module is used to generate an asset verification report based on the asset anomaly type and the risk level.

[0013] Thirdly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any of the above-mentioned embodiments.

[0014] Fourthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the method described in any of the above-described embodiments.

[0015] Fifthly, embodiments of this application provide a computer device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the steps of the method described in any of the preceding claims.

[0016] In summary, the embodiments of this application have at least the following beneficial effects: This application embodiment acquires asset characteristic data corresponding to power industrial control network equipment and determines the asset characteristic information of the asset characteristic data, wherein the asset characteristic data includes traffic data; it calls an AI-based power industrial control network asset anomaly identification model, compares and analyzes the asset characteristic information with the asset ledger benchmark information, and generates asset characteristic matching information; based at least on the asset characteristic matching information, it identifies asset anomalies in the power industrial control network where the power industrial control network equipment is located according to a preset anomaly judgment priority, so as to obtain the asset anomaly type of the power industrial control network and the risk level corresponding to the asset anomaly type; based on the asset anomaly type and the risk level, it generates an asset verification report. In this way, the AI ​​model can be used to achieve efficient and accurate comparison and analysis between the asset ledger benchmark information and the asset characteristic information of the collected asset characteristic data, so as to improve the accuracy of asset identification in complex power industrial control networks. The asset characteristic matching information generated by the analysis is combined with the anomaly judgment priority to identify asset anomalies in the power industrial control network, so as to prioritize the identification of asset anomalies of higher priority types, thereby improving the real-time identification of these types of asset anomalies, and generating an asset verification report for users to understand in a timely and accurate manner. Attached Figure Description

[0017] Figure 1 This is a flowchart illustrating the AI-based asset identification and verification method provided in the embodiments of this application; Figure 2 This is a schematic diagram of the structure of the AI-based asset identification and verification device provided in the embodiments of this application; Figure 3 This is a schematic diagram of the computer device provided in the embodiments of this application. Detailed Implementation

[0018] 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 / examples are only a part of the embodiments / examples of this application, and not all of the embodiments / examples. Based on the embodiments / examples in 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.

[0019] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "multiple" means two or more. In the description of this application, the term "comprising" and its variations are open-ended, meaning "including but not limited to." The term "based on" means "at least partially based on." The term "according to" means "at least partially according to." The term "one embodiment / example" means "at least one embodiment / example"; the term "another embodiment / example" means "at least one additional embodiment / example"; the term "some embodiments / examples" means "at least some embodiments / examples."

[0020] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0021] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this application have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this application is for the purpose of describing specific embodiments only and is not intended to limit the application. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0022] Firstly, see [the following] Figure 1The diagram shows a flowchart of an AI-based asset identification and verification method provided in an embodiment of this application. This AI-based asset identification and verification method can be applied to a computer device with data processing capabilities. The method includes S101-S104, as detailed below.

[0023] S101, acquire asset characteristic data corresponding to power industrial control network equipment, and determine the asset characteristic information of the asset characteristic data, wherein the asset characteristic data includes traffic data.

[0024] S102, invoke the AI-based power industrial control network asset anomaly identification model, compare and analyze the asset feature information with the asset ledger benchmark information, and generate asset feature matching information.

[0025] S103, at least based on the asset feature matching information, perform asset anomaly identification on the power industrial control network where the power industrial control network equipment is located according to the preset anomaly judgment priority, so as to obtain the asset anomaly type of the power industrial control network and the risk level corresponding to the asset anomaly type.

[0026] S104. Based on the asset anomaly type and the risk level, generate an asset verification report.

[0027] In some examples, traffic data can be obtained by collecting communication traffic from power control network devices at the network, transport, and / or application layers. This traffic can include interactions between the devices and internal authorized nodes and external networks. The collected traffic may include session tuples (source IP address, destination IP address, source port, destination port, transport layer protocol), packet length distribution, session duration, packet interaction frequency, and / or application layer protocol payload fragments. Specifically, at least some asset characteristic data of power control network devices can be collected by deploying an AI agent. The AI ​​agent can capture packets by deploying mirror ports to adapt to the full-duplex communication mode of the power control network. Traffic data can be aggregated according to the session dimension to retain the statistical characteristics of each session, including average packet length, maximum packet length, standard deviation of packet length, and / or session start timestamp. Duplicate and empty packets in each session can also be filtered to store valid session data in the traffic data.

[0028] In some examples, asset characteristic data may also include at least one of the following: passive asset characteristics, protocol fingerprint characteristics, active detection information from network security management nodes, etc.

[0029] For example, passive asset characteristics may include at least one of the following: device type, manufacturer name, device model, IP address, device unique identifier ID, protocol version number, hardware performance parameters, and operating status parameters. The device unique identifier ID may include at least one of the following: device unique serial number and MAC (Media Access Control) address; the hardware performance parameters may include at least one of the following: CPU clock speed, memory capacity, and operating years; the operating status parameters may include at least one of the following: CPU utilization, memory usage, voltage stability, and device temperature.

[0030] For example, the matching degree calculation for the device unique identifier ID can be performed as follows: The original character sequence of the device unique serial number is retained, and the Jaro-Winkler string similarity algorithm is used to calculate the similarity between the original character sequence and the sequence registered in the configuration management database; for the MAC address, separators (such as ":" or "-") are first removed, and then an exact match is performed with the MAC addresses registered in the configuration management database. (Device unique identifier ID matching degree) The following formula can be used for calculation:

[0031] in, This represents the exact match result for the MAC address; a value of 1 indicates a match, and a value of 0 indicates no match. This is a unique serial number for the pre-processed acquisition device; The preprocessed configuration management database registers the device's unique serial number; 0.4 is the serial number similarity weight coefficient; 0.6 is the MAC address matching weight coefficient; the parameters for the Jaro-Winkler string similarity algorithm are set as follows: prefix matching length threshold = 4, adjustment coefficient... .

[0032] For example, protocol fingerprint features may include standard protocol fingerprint features and proprietary extended protocol fingerprint features.

[0033] For standard protocols, fingerprint features of three power-specific protocols—IEC 60870-5-104, IEC 61850 MMS, and IEC 60870-5-101—can be extracted to form standard protocol fingerprint features. For example, the extraction order could be: control domain location, address domain extraction, and application service data unit parsing. Specifically, the control domain is located based on the protocol frame structure offset; the address domain is extracted according to device address encoding rules; and the application service data units can be obtained by filtering based on data identifier bits.

[0034] For proprietary extension protocols, the type of proprietary extension protocol can be identified by the protocol frame header identifier and vendor ID field in the proprietary extension protocol information. Features in the proprietary extension protocol information are then extracted in the order of custom control fields, proprietary data units, and verification fields. Custom control fields can be extracted based on field definitions in the vendor's publicly available documentation; proprietary data units can be parsed according to rules regarding field length and data type.

[0035] For example, proactive probing information can be obtained by network security management nodes performing tiered probing. During probing, a dedicated power control system probing protocol (based on an extension of the IEC 62351 security standard) can be used. The IP addresses of the probing nodes are pre-added to the whitelist of power control system network security devices to ensure compliance of the probing behavior. The probing scope can cover commonly used ports corresponding to the dedicated power control protocol (such as port 2404 of the IEC 60870-5-104 protocol, port 102 of the IEC 61850 MMS protocol, and port 2401 of the IEC 60870-5-101 protocol) as well as common public network risk ports (ports 80, 443, and 23). The single-port probing timeout can be set to 30 seconds, and the probing interval for a single device can be set to no less than 60 seconds.

[0036] Among these features, differentiated data collection cycles can be set based on the hardware performance and operational age of the power industrial control network equipment. For high-performance core devices with a CPU clock speed of ≥1GHz, memory capacity of ≥4GB, and operating time of <5 years, the data acquisition cycle can be 1 minute / time. For conventional core devices with a CPU clock speed of ≥800MHz and <1GHz, memory capacity of ≥2GB and <4GB, and operating life of 5 to 8 years, the data acquisition cycle can be 2 minutes / time. For ordinary devices with a CPU clock speed >500MHz and <800MHz, memory capacity ≥1GB and <2GB, and operating years <8 years, the data acquisition cycle can be 5 minutes / time. For older, ordinary devices with a CPU clock speed of ≤500MHz, memory capacity of <1GB, and operating life of ≥8 years, the initial data collection cycle can be 10 minutes / time. If the CPU utilization rate is greater than 60% for three consecutive data collections, it can be automatically adjusted to 15 minutes / time to reduce the resource pressure on older devices.

[0037] For example, key events can be recorded synchronously during the collection of asset characteristic data, including protocol parsing failure events, port unresponsive events, probe anomaly events, and power-specific protocol field parsing error events. The recorded content may include the device's unique identifier ID, event occurrence time, event type, device type, and hardware performance parameters, in order to provide contextual support for subsequent asset anomaly identification.

[0038] In some examples, determining the asset characteristic information of the asset characteristic data may include: performing standardized preprocessing and feature fusion on the asset characteristic data to obtain the asset characteristic information.

[0039] For example, in the standardization preprocessing process, standardization preprocessing can be performed separately for numerical features and protocol fingerprint features in the asset feature data.

[0040] For the standardization preprocessing of numerical features, the numerical features may include the aforementioned operational status parameters from the asset feature data. The Z-score standardization algorithm can be used, and calculations can be performed for different scenarios based on the accumulated historical data of the equipment.

[0041] Scenario 1: When the device has collected historical data for six months, its standardized calculation formula can be:

[0042] in, For the first The standardized feature values ​​of dimensional numerical features; For the first The original feature values ​​of dimensional numerical features; For the first time this device The average value of historical data collected over the most recent 6 months for dimensional numerical features; For the first time this device Dimensional numerical features: standard deviation of historical data collected over the most recent 6 months; when At that time, it can be set .

[0043] Scenario 2: When the device is a new device (a new device can refer to a device that has been connected to the power control network and has collected historical data for less than six months), its standardized calculation formula can be:

[0044] in, For the first The standardized feature values ​​of dimensional numerical features; For the first The original feature values ​​of dimensional numerical features; To configure devices of the same manufacturer, model, and type in the management database The average value of historical data collected over the past 6 months for dimensional numerical features; To configure devices of the same manufacturer, model, and type in the management database The standard deviation of the historical data collected over the past 6 months for dimensional numerical features; when the new device collects enough historical data to meet the statistical reliability requirements of Z-score standardization (for example, a total of no less than 100 records collected and covering the entire runtime), it will automatically switch to the calculation method of scenario 1.

[0045] It should be noted that the operating status parameters may include at least one of the following: CPU utilization, memory usage, voltage stability, and device temperature, one of which can represent a numerical characteristic of one dimension.

[0046] For the standardization preprocessing of protocol fingerprint features, encoding mapping and normalization processing can be performed on the protocol fingerprint features.

[0047] Specifically, regarding the processing of standard protocol fields: the control field can be mapped sequentially to integers from 0 to 15 according to the sixteen instruction types defined in the IEC60870-5-104 standard; the address field can be segmented according to the device address, with each segment converted to an eight-bit binary number and then mapped to a decimal value; and the application service data unit can be mapped to integers from 0 to 100 according to the data type and value range.

[0048] Specifically, for the processing of private extended protocol fields: custom control fields can be mapped to integers from 16 to 31 according to the instruction type defined by the manufacturer, so as to clearly distinguish them from standard protocol fields; private data units can be mapped to integers from 0 to 100 according to "data type + value range"; and check fields can be mapped to integers from 0 to 32 according to the check bit length.

[0049] After completing the above mapping and encoding, all protocol field data can be padded to a uniform length of 256 bits: if the encoded length is less than 256 bits, zeros can be added to the end; if it exceeds 256 bits, the first 256 bits of the core field encoding data can be retained to form a fixed-length numerical sequence. The Min-Max normalization algorithm can then be used to normalize this numerical sequence to the [0,1] interval. The calculation formula is as follows:

[0050] in, These are the normalized feature values ​​of the protocol fingerprint. A single element in a numerical sequence after the protocol fingerprint features are encoded; The minimum value of the encoded features of the full protocol fingerprint features statistically analyzed during the pre-training phase; This represents the maximum value of the encoded features of the full protocol fingerprint features obtained during the pre-training phase.

[0051] In one optional implementation, the asset characteristic information for determining the asset characteristic data includes: The continuous feature data in the asset feature data is discretized to obtain a discretized feature sequence, wherein the discretized feature sequence includes at least one dimension of discretized features. Determine the information entropy gain of each of the at least one dimension of the discretized features; Based on the information entropy gain, the discrete weights corresponding to the discrete features of each dimension are determined. Based on the discrete weights, the discrete features of at least one dimension are weighted, and the asset feature information is determined based on the weighted discrete features.

[0052] In some examples, discretization may include binning discretization, where the binning method can be ordered equidistant binning based on business rules, and the number of bins can be set to 8 groups. Continuous features may include CPU utilization, protocol encoding normalized values, and standardized values ​​of runtime status parameters.

[0053] Furthermore, binning can be performed on ordered, equidistant intervals. For example, CPU utilization can be divided into intervals such as "0–12.5%, 12.5%–25%, 25%–37.5%, ..., 87.5%–100%"; voltage stability values ​​can be divided according to the gradient of power system safety operation thresholds to improve the matching degree between binning results and actual business logic and safety specifications. After binning, encoding mapping can be performed on the ordered intervals after binning. For example, eight ordered intervals can be mapped sequentially to integer encoding values ​​from 0 to 7 in ascending order of feature values, forming a discretized feature sequence. This sequence can contain at least one dimension of discretized features, and discretized features of the same dimension can correspond to the same type of continuous features.

[0054] Information gain (IG) can be used as an indicator to measure the ability of each dimension of discretized features to distinguish asset classes. For the i-th dimension of discretized features, its information gain... The calculation formula can be:

[0055] in, For the first Information gain of dimensionally discretized features (used to characterize feature importance); For the original dataset The information entropy, where the original dataset can refer to continuous feature data before discretization; In the known number of Under the condition of dimensional discretization features, the original dataset Conditional information entropy, for example, It can be used to characterize: according to For dataset After partitioning, the average value of the uncertainty (i.e. disorder) within each subset is obtained.

[0056] Information entropy The calculation formula is:

[0057] in, For dataset medium sample The probability of its occurrence.

[0058] Conditional information entropy The calculation formula is:

[0059] in, For the first All values ​​of the dimension discretized feature (integer encoding from 0 to 7); For the first Dimensional discretization feature extraction A subset of the values; For the first Dimensional discretization features.

[0060] In some examples, to quantify the contribution of each feature dimension to asset identification, the information gain can be normalized to obtain the discrete weights of each dimension. The calculation formula is as follows:

[0061] in, For the first Weights of dimensional discretized features; For all The sum of feature information gains in each dimension; This represents the total dimension of the features.

[0062] In some examples, asset characteristic information can be an asset feature vector. The asset ledger baseline information can be the ledger baseline feature vector. , The dimensional structure can be referenced from the configuration management database. The feature dimension specification ensures that the model input matches the baseline information in dimensions (which can be completely identical), reducing the comparison bias caused by dimension mismatch (such as dimension misalignment).

[0063] Among them, asset feature vector The calculation formula is:

[0064] in, For the first Integer encoded values ​​after discretization of dimensional features.

[0065] It should be noted that, The sources include two categories: Numerical features (such as CPU utilization, memory usage, etc.) are first processed by Z-score standardization, and then subjected to the above-mentioned ordered equal-distance binning and integer encoding. The protocol fingerprint features are first normalized to the [0,1] interval by Min-Max, and then binned and encoded in the same way.

[0066] In this embodiment, through the weighting mechanism described above, features with high information gain (such as CPU utilization ranges that can effectively distinguish between core devices and ordinary devices, specific protocol field encoding, etc.) can be given higher weights, thereby dominating asset feature information and improving the sensitivity and accuracy of subsequent anomaly identification.

[0067] In some examples, the power control network asset anomaly identification model can be trained in the following ways.

[0068] (1) Construct a comprehensive training dataset containing data samples of various operating scenarios of assets in the power monitoring system, including normal operation scenario data, illegal access simulation scenario data, IP drift simulation scenario data, external device access simulation scenario data, protocol parsing failure simulation scenario data, and port no response simulation scenario data. Each scenario has no less than 500 data entries to ensure coverage of core equipment and ordinary equipment, and covers different operating periods such as high load, low load, and start-stop switching, in order to improve the model's generalization ability.

[0069] (2) Preprocessing the training dataset may include: deleting duplicate records in the training dataset, filling missing values ​​with the mean, filtering minority class noise samples using the K nearest neighbor algorithm, using random undersampling to control the number of majority class samples to less than half of the total samples, and finally synthesizing new samples according to the sampling multiple n by synthetic minority oversampling technique (SMOTE), where n is the ratio of the number of majority class samples after undersampling to the number of minority class samples after denoising, so as to obtain a balanced dataset to avoid class bias affecting the model training effect.

[0070] (3) The standardized asset feature vector As input features for the model, the dimensions of the input features consist of the numerical feature dimensions plus the dimension after encoding the 256-bit protocol fingerprint features, so as to adapt to the feature requirements of multi-protocol and heterogeneous devices in the power industrial control environment.

[0071] (4) Model structure and parameter configuration: The model structure can be arranged in the order of data flow, including an input layer, a feature transformation layer, a protocol fingerprint feature attention layer, a fully connected layer, and an output layer. The specific configuration is as follows: Input layer: Receives asset feature vectors Its dimensions are consistent with the input features, and it is used to achieve standardized conversion of data formats.

[0072] Feature Transformation Layer: A hybrid structure of one-dimensional convolutional layer (Conv1D) and fully connected layer is employed. The Conv1D module contains two convolutional blocks, each containing 64 1×3 convolutional kernels, a ReLU activation function, and a 1×2 max-pooling layer. These convolutional blocks can be used to capture local correlations in the protocol sequence. The fully connected module contains a single fully connected layer with 64 neurons, preserving global features of the device's operating state. The output feature dimensions of the Conv1D and fully connected layers are concatenated and used as input to the feature attention layer.

[0073] Feature Attention Layer (i.e. Protocol Fingerprint Feature Attention Layer): By calculating the cosine similarity between the query vector and the key vector, and combining it with the temperature adjustment factor (λ=0.9 for core equipment and λ=1.1 for ordinary equipment) to allocate attention weights, the feature distinguishability of power-specific protocols and private extended protocols is enhanced.

[0074] Fully connected layer: Designed as a three-layer neural network structure with 512, 256 and 128 neurons respectively. Each layer is equipped with a Dropout layer to suppress overfitting.

[0075] Output layer: Outputs the original similarity information (such as original cosine similarity) between the asset feature vector and the ledger baseline feature vector, which is used for subsequent matching degree normalization calculation.

[0076] (5) Model Training Settings: The Adam optimizer and MSE (Mean Squared Error) are used during model training. The initial learning rate is set to 0.001, and it decays to 80% of the current value after every 5 iterations. The training batch size is set to 32. The training data is divided into a training set, a validation set, and a test set in a 7:2:1 ratio, which are used for updating model parameters, monitoring and verifying training effects, and final performance evaluation, respectively. In addition, an early stopping mechanism can be introduced, that is, training is stopped when the loss fluctuation on the validation set is less than 0.001 for 5 consecutive rounds or increases for 3 consecutive rounds, and the model parameters with the minimum loss are backtracked to prevent overfitting and improve training efficiency.

[0077] (6) Performance Verification and Result Output: After training, the model's performance is verified using a test set. The model is required to achieve an accuracy of no less than 95%, a recall of no less than 93%, and an F1 score of no less than 94% on the test set. The final output power industrial control network asset anomaly identification model can classify the input asset feature vector and identify six categories: normal operation, illegal access, IP migration, external device access, protocol parsing failure, and port unresponsiveness. A confidence score (range 0-1) is attached to each category of results. When the confidence score is lower than 0.8, it is marked as requiring manual review to ensure the accuracy of the identification results.

[0078] In one optional implementation, the power industrial control network asset anomaly identification model includes a feature transformation unit and a feature attention layer; exemplarily, the feature transformation unit may include the aforementioned input layer and feature transformation layer.

[0079] The process involves invoking an AI-based power control network asset anomaly identification model, comparing and analyzing the asset feature information with the asset ledger baseline information to generate asset feature matching information, including: The feature transformation unit is invoked to transform the asset feature information to obtain the transformed asset feature information; for example, the transformed asset feature information can be used to indicate the output of the feature transformation layer. The feature attention layer is invoked to determine the first similarity information between the query vector and the key vector corresponding to the transformed asset feature information, and attention weight information is determined based on the first similarity information and the temperature adjustment factor, wherein the temperature adjustment factor is related to the equipment importance level of the power industrial control network equipment; Based on the attention weight information, the transformed asset feature information is weighted. The asset feature matching information is generated by comparing and analyzing the asset ledger baseline information with the weighted asset feature information.

[0080] In some examples, the initial similarity information can be obtained by calculating the cosine similarity between the query vector and the key vector corresponding to the transformed asset feature information. The calculation formula is as follows:

[0081] in, For the first query vectors and the Key vectors Cosine similarity; For the first One query vector; For the first One key vector; for and The dot product; for The L2 norm; for The L2 norm.

[0082] In some examples, a temperature regulation factor can be set. : Core equipment (including high-performance core equipment and conventional core equipment) scenarios In the scenario of ordinary equipment (including old ordinary equipment) ; In some examples, the attention weight information includes attention weights. Calculate attention weights The calculation formula is:

[0083] in, For the first The query vector and the first Attention weights corresponding to each key vector; It is a natural exponential function; Temperature regulation factor; For the first Each query vector and all The sum of the similarity of each key vector after exponential operation; This represents the total number of protocol fingerprint features. For the first A number of key vectors.

[0084] In one optional implementation, the step of comparing and analyzing the asset ledger baseline information with the weighted asset characteristic information to generate the asset characteristic matching information includes: The asset ledger baseline information is verified to extract valid asset ledger baseline information and the baseline feature information corresponding to the valid asset ledger baseline information is extracted. Determine the original similarity information between the baseline feature information and the weighted asset feature information; The original similarity information is normalized to obtain the asset feature matching information.

[0085] In some examples, the verification process may include ledger consistency verification. When performing this verification, the fields to be compared may include key fields such as IP address, device type, manufacturer name, device model, device unique identifier (ID), and hardware performance parameters. Each ledger record in the asset ledger baseline information can be verified one by one, eliminating ledger data with conflicts in any key field, retaining only valid asset ledger data where key fields are completely consistent. Then, based on this valid asset ledger data, a ledger baseline feature vector can be constructed. The ledger's baseline feature vector The baseline feature vector corresponding to the effective asset ledger is constructed in the same way as the asset feature vector. The construction methods can be similar, making... and The dimensional structure and feature encoding specifications can be fully aligned, ensuring the consistency and accuracy of subsequent comparisons.

[0086] In some examples, asset feature matching information may include the degree of asset feature matching. Asset feature matching degree The original cosine similarity output by the model can be normalized to the [0,1] interval. The normalization formula is:

[0087] in, The asset feature matching degree (value range [0,1]); For asset feature vectors; This is the baseline feature vector for the ledger; for and The dot product; for The L2 norm; for The L2 norm; This represents the original cosine similarity in the original similarity information (with a value range of [-1, 1]).

[0088] In an optional implementation, the method further includes: The current load rate of the power industrial control network equipment and the cumulative number of core device anomalies in the most recent preset time period are obtained. For example, the most recent preset time period can be the past hour. The cumulative number of core device anomalies can be used to indicate the total number of core anomaly events that have occurred on the device. Core anomaly events can include critical security events such as protocol parsing failure, port unresponsiveness, and unauthorized access. The current load rate can be used to indicate the resource consumption level of the device during real-time operation, such as the real-time CPU utilization of a single device. Based on the current load rate, and the minimum load rate threshold and high load critical threshold corresponding to the power industrial control network equipment, determine the load rate influencing factor; An anomaly frequency impact factor is determined based on the ratio between the cumulative number of core anomalies of the equipment and the maximum core anomaly threshold, wherein the maximum core anomaly threshold is used to indicate the upper limit of the number of core anomalies of the power industrial control network equipment in the most recent preset time period. Based on the load rate influence factor, the abnormal frequency influence factor, and the previous period's dynamic threshold, determine the current period's dynamic threshold; The step of identifying asset anomalies in the power industrial control network where the power industrial control network equipment is located, based at least on the asset feature matching information and according to a preset anomaly judgment priority, includes: Based on the asset feature matching information and the current periodic dynamic threshold, the power industrial control network where the power industrial control network equipment is located is identified as having asset anomalies according to the preset anomaly judgment priority.

[0089] In some examples, it is assumed that the minimum load rate threshold for normal operation of power control network equipment is... High load critical threshold for normal operation of power industrial control network equipment Maximum number of core anomalies threshold This can represent the maximum threshold for the number of core anomalies of a single device within the most recent preset time period. .

[0090] Current load rate The real-time CPU utilization of a single device can be obtained through the above-described operating status parameter acquisition process, based on the extended version of the power industrial control network using the device's SNMP protocol.

[0091] Load factor The calculation formula is:

[0092] in, The load factor (value range [0,1]) This is a function to find the maximum value. The function is for finding the minimum value; This is the minimum load rate threshold; This is the high load critical threshold.

[0093] In some examples, the anomaly frequency impact factor is calculated. The calculation formula is:

[0094] in, The abnormal frequency impact factor (value range [0, 0.2]); This refers to the cumulative number of core device anomalies within the most recent preset time period. This is the threshold for the maximum number of core anomalies.

[0095] In some examples, the original dynamic threshold for the current period The calculation formula is:

[0096] Furthermore, it is possible to... Applying a clamping constraint within the [0.6, 0.9] interval yields the final dynamic threshold. The clamping calculation formula is as follows:

[0097] in, This is the dynamic threshold for the current period (value range [0.6, 0.9]).

[0098] In some examples, the default anomaly detection priority can be sorted from high to low based on the severity of security risks, in the following order: suspected unauthorized external connection anomaly, unauthorized access to unregistered devices anomaly, asset information change (IP address drift) anomaly, registered but not online anomaly, and legitimate unregistered device anomaly.

[0099] The logic for determining the above-mentioned exception types is as follows: (1) Abnormal external connection equipment suspected of being unauthorized: When the following conditions are met simultaneously, the external connection device is considered to be malfunctioning and potentially unauthorized: The target ports are common public network risk ports, including but not limited to port 80 (HTTP), port 443 (HTTPS), and port 23 (Telnet). The target port is in a listening state, and the detection criteria are: the port status is "open"; No valid external access authorization: After querying the configuration management database, there is no external access authorization record for this device; or after querying the network configuration, the port corresponding to this device has an external access configuration that is not in the authorized access list of the power industrial control network boundary router. External network session exists: TCP / UDP session interaction from an authorized network segment outside the power industrial control network was detected, and the destination address of the packet is not in the operation and maintenance IP whitelist.

[0100] The port status, network segment affiliation, and session information mentioned above are collected and cross-verified by the AI ​​agent and network security management node to ensure the reliability of the judgment results.

[0101] (2) Abnormal access to unregistered devices: When the asset feature matching score ∈ [0, 0.5), and suspected unauthorized external connection devices have been excluded, and the following conditions are met, it is determined to be an unauthorized access to an unregistered device: There is no asset ledger record corresponding to the unique identifier ID of this device in the configuration management database; The device has no valid access approval record; No matching vendor registration information was found for the protocol fingerprint feature.

[0102] (3) Abnormal asset information change (IP address drift): When an asset feature matching score ∈ [0.5, Th) (Th is the dynamic threshold for the current period) and the following conditions are met simultaneously, it is determined to be an asset information change (IP address drift) anomaly: Matching degree of device unique identifier ID It was confirmed to be the same device already registered in the configuration management database; The actual IP address of the device does not match the IP address registered in the configuration management database; The IP addresses collected in two consecutive data collections are different according to the device's own data collection cycle; There are no legitimate IP address change operation and maintenance records.

[0103] The core logic behind this determination is that if the device hardware identification information has not changed, but the IP address has undergone an unauthorized change, it may reflect a configuration error, address conflict, or potential security incident.

[0104] (4) Registered but not online, abnormal: When an asset's feature matching score ∈ [0, 0.5), and the above three types of anomalies have been excluded, and the following conditions are met, it is judged as an anomaly of "registered but not yet online": The configuration management database contains an asset ledger record corresponding to the device's unique identifier ID; There are no valid maintenance work orders covering the current data collection period; No device operation signals (such as heartbeat packets, protocol interactions, CPU utilization, and other operating status parameters) were detected for three consecutive acquisition cycles.

[0105] (5) Abnormalities in legally registered but unregistered equipment: When the asset feature matching score ∈ [0, 0.5), and the above four types of abnormal situations have been excluded, and the following conditions are met, it is judged as an abnormal legitimate unregistered device: There is no asset ledger record corresponding to the unique identifier ID of this device in the configuration management database; The device has a valid access approval record; The protocol fingerprint features are matched with the manufacturer's registration information.

[0106] Such anomalies typically occur when newly deployed devices have been approved for access but have not yet completed the ledger entry process. In such cases, a prompt can be made to update the ledger synchronously.

[0107] To support the above-mentioned anomaly detection logic, this embodiment can further define the detection methods for several key auxiliary information: The determination method for the maintenance IP whitelist: The Power Control Network Security Management Center is the main updater, employing a mechanism combining quarterly batch updates and temporary additions. Quarterly batch updates synchronize regular IP address change records for maintenance personnel; temporary additions are used to adapt IP addresses for temporary scenarios such as mobile office and emergency maintenance. All whitelist data is stored in a dedicated field in the configuration management database and is verified in real time during anomaly detection.

[0108] The criteria for determining a valid maintenance work order are as follows: A maintenance work order must include four core fields: unique device identifier (ID), maintenance item, execution time, and approver; the work order status must be "approved" and the execution time must cover the current data collection period for it to be considered valid.

[0109] The criteria for determining a legitimate IP address change maintenance record are as follows: The record must include five core fields: device unique identifier ID, IP address before change, IP address after change, reason for change, and approver; the record status must be "effective".

[0110] The criteria for determining a legitimate access approval record are as follows: The record must include five core fields: device unique identifier ID, access reason, access period, approval department, and approver; the record status must be "approved" and the access period must cover the current collection cycle.

[0111] Among them, valid maintenance work orders and legitimate IP address change maintenance records are independent record structures. They share the "device unique identifier ID" and "approver" fields. The configuration management database can realize related queries through the device unique identifier ID to ensure information consistency and traceability.

[0112] In this embodiment, "core anomalies" can refer to the following four types of events described above: protocol parsing failure events, port unresponsive events, probe anomaly events, and power-specific protocol (including proprietary extended protocols) field parsing error events. These events, as important indicators of equipment malfunction, are included in the anomaly frequency statistics and risk assessment system.

[0113] In summary, this embodiment can achieve comprehensive, accurate, and interpretable verification of the status of power industrial control network assets by constructing a refined anomaly judgment rule system and combining multi-source ledger data, dynamic matching degree thresholds, and equipment operating context. This embodiment not only improves the real-time performance and completeness of asset ledgers but also provides a solid data foundation and decision support for power system security protection.

[0114] In some examples, the generated asset verification report may include the following: Basic device information includes the device's unique identifier (ID), IP address, device type, manufacturer name, device model, and hardware performance parameters. Anomaly types: such as suspected unauthorized external connection devices, unauthorized access to unregistered devices, asset information changes (IP address migration), registered but not online devices, and legitimate but unregistered devices. Judgment criteria include: asset feature matching score, feature comparison results (such as device unique identifier ID matching score Match_hard, protocol fingerprint consistency, etc.), and key event records (such as protocol parsing failure events, port unresponsive events, etc.). Risk levels are categorized based on the degree of security impact as follows: High risk: suspected unauthorized external connection devices or unauthorized access to unregistered devices; Medium risk: asset information changes (IP address migration); Low risk: registered but not online devices or legitimate but unregistered devices.

[0115] In some examples, standardized processing work orders can be generated based on asset verification reports and pushed to the relevant responsible personnel. The automatically generated standardized processing work orders can include a work order number, anomaly information, handling requirements, and the responsible department. Differential processing time limits can also be set based on risk level: 1 hour for high-risk work orders; 4 hours for medium-risk work orders; and 24 hours for low-risk work orders.

[0116] Work orders are pushed to the responsible department's intranet maintenance terminal via a dedicated protocol for the power industrial control network. At the same time, the power industrial control network security management platform can send SMS reminders to the department head to ensure timely response and closed-loop handling of abnormal events.

[0117] In an optional implementation, the method further includes: The dynamic profile of power industrial control network assets is updated based on the continuous and / or discrete feature data in the asset feature data.

[0118] In some examples, the dynamic profile of the power industrial control network assets can include both static and dynamic attributes. Static attributes are factory-fixed information, including unique device identifiers (IDs), manufacturer names, device models, and hardware performance parameters. These attributes can be refreshed synchronously only when the configuration management database is updated, and remain unchanged during daily operation.

[0119] Dynamic attributes are divided into continuous dynamic attributes and discrete dynamic attributes, and their update mechanisms are as follows.

[0120] (1) Continuous dynamic attribute update Continuous dynamic attributes include CPU utilization, memory usage, voltage stability, and device temperature. The system uses an exponentially weighted moving average to calculate the updated characteristic values. The formula is:

[0121] in, For adaptive weights; This is the previously collected and stored value of this dynamic attribute; This represents the current collected value of the dynamic attribute; This represents the weighting coefficient of the currently collected value.

[0122] Among them, adaptive weights Based on the characteristic variation coefficient of this attribute over the past 30 days Dynamic adjustment. Characteristic coefficient of variation. The calculation formula is:

[0123] in, This represents the standard deviation of the data collected over the past 30 days for this continuous dynamic attribute. This is the average value of the data collected over the past 30 days for this continuous dynamic attribute.

[0124] Based on statistical analysis of nearly one year's data collected from various power industrial control network devices and multiple types of continuous dynamic attributes, the following settings are established. The mapping relationship between threshold and weight is as follows: when At that time, the feature stability is high, and the value is taken ; when At that time, the characteristics are moderately stable, and the following is taken. ; when At that time, the characteristic fluctuation is large, and the value is taken as follows: .

[0125] Thus, this embodiment can ensure that historical trends are preserved when the equipment is running smoothly, and respond more sensitively to the current state when there are sharp fluctuations.

[0126] (2) Discrete dynamic attribute update Discrete dynamic attributes include IP address and protocol version number. Their update logic is as follows: If there are legitimate maintenance change records (such as legitimate IP address change records), then the current collected value will be used directly. Replace the previously stored value This is considered a normal change; If no legitimate changes to the operation and maintenance records exist, it will be marked as abnormal and will still be... Update the asset dynamic profile and simultaneously synchronize the change information to the asset verification report for manual review and processing.

[0127] To further improve the stability of anomaly detection, a 24-hour moving average matching degree can be calculated. The calculation formula is as follows:

[0128] in, This represents the total number of data collection cycles actually completed in the past 24 hours (the past 24 hours is a time window that goes back 24 hours from the current data collection time; only data collection cycles successfully completed within this window are counted, and missing cycles due to data collection failures or device offline are not included). For all within the last 24 hours Asset feature matching degree for each collection period ( The summation operation of the results; For the first Asset feature matching degree for each collection period ( )result.

[0129] Secondly, correspondingly, this application also provides an AI-based asset identification and verification device, which can implement all the processes of the AI-based asset identification and verification method provided in the above embodiments.

[0130] See Figure 2 The diagram shows a structural schematic of an AI-based asset identification and verification device 200 provided in an embodiment of this application. The AI-based asset identification and verification device 200 includes: The feature processing module 201 is used to acquire asset feature data corresponding to power industrial control network equipment and determine the asset feature information of the asset feature data, wherein the asset feature data includes traffic data; Matching module 202 is used to call the AI-based power industrial control network asset anomaly identification model, compare and analyze the asset feature information with the asset ledger benchmark information, and generate asset feature matching information. The anomaly identification module 203 is used to identify asset anomalies in the power industrial control network where the power industrial control network equipment is located, based at least on the asset feature matching information and according to a preset anomaly judgment priority, so as to obtain the asset anomaly type of the power industrial control network and the risk level corresponding to the asset anomaly type. The report generation module 204 is used to generate an asset verification report based on the asset anomaly type and the risk level.

[0131] In one optional implementation, the asset characteristic information for determining the asset characteristic data includes: The continuous feature data in the asset feature data is discretized to obtain a discretized feature sequence, wherein the discretized feature sequence includes at least one dimension of discretized features. Determine the information entropy gain of each of the at least one dimension of the discretized features; Based on the information entropy gain, the discrete weights corresponding to the discrete features of each dimension are determined. Based on the discrete weights, the discrete features of at least one dimension are weighted, and the asset feature information is determined based on the weighted discrete features.

[0132] In one optional implementation, the power industrial control network asset anomaly identification model includes a feature transformation unit and a feature attention layer; The process involves invoking an AI-based power control network asset anomaly identification model, comparing and analyzing the asset feature information with the asset ledger baseline information to generate asset feature matching information, including: The feature transformation unit is invoked to transform the asset feature information to obtain the transformed asset feature information. The feature attention layer is invoked to determine the first similarity information between the query vector and the key vector corresponding to the transformed asset feature information, and attention weight information is determined based on the first similarity information and the temperature adjustment factor, wherein the temperature adjustment factor is related to the equipment importance level of the power industrial control network equipment; Based on the attention weight information, the transformed asset feature information is weighted. The asset feature matching information is generated by comparing and analyzing the asset ledger baseline information with the weighted asset feature information.

[0133] In one optional implementation, the step of comparing and analyzing the asset ledger baseline information with the weighted asset characteristic information to generate the asset characteristic matching information includes: The asset ledger baseline information is verified to extract valid asset ledger baseline information and the baseline feature information corresponding to the valid asset ledger baseline information is extracted. Determine the original similarity information between the baseline feature information and the weighted asset feature information; The original similarity information is normalized to obtain the asset feature matching information.

[0134] In an optional embodiment, the apparatus further includes a dynamic threshold determination module, which is used to: Obtain the current load rate of the power industrial control network equipment and the cumulative number of core equipment anomalies in the most recent preset time period; Based on the current load rate, and the minimum load rate threshold and high load critical threshold corresponding to the power industrial control network equipment, determine the load rate influencing factor; An anomaly frequency impact factor is determined based on the ratio between the cumulative number of core anomalies of the equipment and the maximum core anomaly threshold, wherein the maximum core anomaly threshold is used to indicate the upper limit of the number of core anomalies of the power industrial control network equipment in the most recent preset time period. Based on the load rate influence factor, the abnormal frequency influence factor, and the previous period's dynamic threshold, determine the current period's dynamic threshold; The step of identifying asset anomalies in the power industrial control network where the power industrial control network equipment is located, based at least on the asset feature matching information and according to a preset anomaly judgment priority, includes: Based on the asset feature matching information and the current periodic dynamic threshold, the power industrial control network where the power industrial control network equipment is located is identified as having asset anomalies according to the preset anomaly judgment priority.

[0135] In one alternative embodiment, the device further includes: The dynamic profile update module is used to update the dynamic profile of power industrial control network assets based on continuous and / or discrete feature data in the asset feature data.

[0136] Thirdly, embodiments of this application provide a non-transitory computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the method described in any of the above-mentioned embodiments.

[0137] Fourthly, embodiments of this application provide a computer program product, including computer instructions that, when executed by a processor, implement the steps of the method described in any of the above-described embodiments.

[0138] Fifthly, embodiments of this application provide a computer device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the steps of the method described in any of the preceding claims.

[0139] See Figure 3 The computer device in this embodiment includes a processor 301, a memory 302, and a computer program stored in the memory 302 and executable on the processor 301, such as an AI-based asset identification and verification program. When the processor 301 executes the computer program, it implements the steps in the various AI-based asset identification and verification method embodiments described above, for example... Figure 1 The steps S101-S104 are shown.

[0140] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 302 and executed by the processor 301 to complete this application. The one or more modules / units may be a series of computer program instruction segments capable of performing a specific function, which describe the execution process of the computer program in the computer device.

[0141] The computer device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device may include, but is not limited to, a processor 301 and a memory 302. Those skilled in the art will understand that the schematic diagram is merely an example of a computer device and does not constitute a limitation on the computer device. It may include more or fewer components than shown, or combine certain components, or different components. For example, the computer device may also include input / output devices, network access devices, buses, etc.

[0142] The processor 301 can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or the processor 301 can be any conventional processor. The processor 301 is the control center of the computer device, connecting various parts of the entire computer device through various interfaces and lines.

[0143] The memory 302 can be used to store the computer programs and / or modules. The processor 301 implements various functions of the computer device by running or executing the computer programs and / or modules stored in the memory 302 and calling the data stored in the memory 302. The memory 302 may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created according to the use of the mobile phone (such as audio data, phonebook, etc.). In addition, the memory 302 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0144] Wherein, if the modules / units integrated into the computer device are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a non-transitory computer-readable storage medium. When the computer program is executed by the processor 301, it can implement the steps of the various method embodiments described above. Wherein, the computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form, etc. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0145] In summary, the embodiments of this application have at least the following beneficial effects: This application embodiment acquires asset characteristic data corresponding to power industrial control network equipment and determines the asset characteristic information of the asset characteristic data, wherein the asset characteristic data includes traffic data; it calls an AI-based power industrial control network asset anomaly identification model, compares and analyzes the asset characteristic information with the asset ledger benchmark information, and generates asset characteristic matching information; based at least on the asset characteristic matching information, it identifies asset anomalies in the power industrial control network where the power industrial control network equipment is located according to a preset anomaly judgment priority, so as to obtain the asset anomaly type of the power industrial control network and the risk level corresponding to the asset anomaly type; based on the asset anomaly type and the risk level, it generates an asset verification report. In this way, the AI ​​model can be used to achieve efficient and accurate comparison and analysis between the asset ledger benchmark information and the asset characteristic information of the collected asset characteristic data, so as to improve the accuracy of asset identification in complex power industrial control networks. The asset characteristic matching information generated by the analysis is combined with the anomaly judgment priority to identify asset anomalies in the power industrial control network, so as to prioritize the identification of asset anomalies of higher priority types, thereby improving the real-time identification of these types of asset anomalies, and generating an asset verification report for users to understand in a timely and accurate manner.

[0146] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary hardware platforms, or it can be implemented entirely by hardware. Based on this understanding, all or part of the technical solutions of this application that contribute to the background technology can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM (Read-Only Memory) / RAM (Random Access Memory), magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0147] The above description represents optional embodiments of this application. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of this application, and these improvements and modifications are also considered to be within the scope of protection of this application.

Claims

1. An AI-based asset identification and verification method, characterized in that, include: Obtain asset characteristic data corresponding to power industrial control network equipment, and determine the asset characteristic information of the asset characteristic data, wherein the asset characteristic data includes traffic data; The system invokes an AI-based power industrial control network asset anomaly identification model, compares and analyzes the asset feature information with the asset ledger baseline information, and generates asset feature matching information. Based at least on the asset feature matching information, the power industrial control network where the power industrial control network equipment is located is identified as having asset anomalies according to a preset anomaly judgment priority, so as to obtain the asset anomaly type of the power industrial control network and the risk level corresponding to the asset anomaly type. An asset verification report is generated based on the described asset anomaly type and risk level.

2. The method according to claim 1, characterized in that, The asset characteristic information used to determine the asset characteristic data includes: The continuous feature data in the asset feature data is discretized to obtain a discretized feature sequence, wherein the discretized feature sequence includes at least one dimension of discretized features. Determine the information entropy gain of each of the at least one dimension of the discretized features; Based on the information entropy gain, the discrete weights corresponding to the discrete features of each dimension are determined. Based on the discrete weights, the discrete features of at least one dimension are weighted, and the asset feature information is determined based on the weighted discrete features.

3. The method according to claim 1, characterized in that, The power industrial control network asset anomaly identification model includes a feature transformation unit and a feature attention layer; The process involves invoking an AI-based power control network asset anomaly identification model, comparing and analyzing the asset feature information with the asset ledger baseline information to generate asset feature matching information, including: The feature transformation unit is invoked to transform the asset feature information to obtain the transformed asset feature information. The feature attention layer is invoked to determine the first similarity information between the query vector and the key vector corresponding to the transformed asset feature information, and attention weight information is determined based on the first similarity information and the temperature adjustment factor, wherein the temperature adjustment factor is related to the equipment importance level of the power industrial control network equipment; Based on the attention weight information, the transformed asset feature information is weighted. The asset feature matching information is generated by comparing and analyzing the asset ledger baseline information with the weighted asset feature information.

4. The method according to claim 3, characterized in that, The step of comparing and analyzing the asset ledger baseline information with the weighted asset feature information to generate the asset feature matching information includes: The asset ledger baseline information is verified to extract valid asset ledger baseline information and the baseline feature information corresponding to the valid asset ledger baseline information is extracted. Determine the original similarity information between the baseline feature information and the weighted asset feature information; The original similarity information is normalized to obtain the asset feature matching information.

5. The method according to any one of claims 1-4, characterized in that, The method further includes: Obtain the current load rate of the power industrial control network equipment and the cumulative number of core equipment anomalies in the most recent preset time period; Based on the current load rate, and the minimum load rate threshold and high load critical threshold corresponding to the power industrial control network equipment, determine the load rate influencing factor; An anomaly frequency impact factor is determined based on the ratio between the cumulative number of core anomalies of the equipment and the maximum core anomaly threshold, wherein the maximum core anomaly threshold is used to indicate the upper limit of the number of core anomalies of the power industrial control network equipment in the most recent preset time period. Based on the load rate influence factor, the abnormal frequency influence factor, and the previous period's dynamic threshold, determine the current period's dynamic threshold; The step of identifying asset anomalies in the power industrial control network where the power industrial control network equipment is located, based at least on the asset feature matching information and according to a preset anomaly judgment priority, includes: Based on the asset feature matching information and the current periodic dynamic threshold, the power industrial control network where the power industrial control network equipment is located is identified as having asset anomalies according to the preset anomaly judgment priority.

6. The method according to any one of claims 1-4, characterized in that, The method further includes: The dynamic profile of power industrial control network assets is updated based on the continuous and / or discrete feature data in the asset feature data.

7. An AI-based asset identification and verification device, characterized in that, include: The feature processing module is used to acquire asset feature data corresponding to power industrial control network equipment and determine the asset feature information of the asset feature data, wherein the asset feature data includes traffic data; The matching module is used to call the AI-based power industrial control network asset anomaly identification model, compare and analyze the asset feature information with the asset ledger benchmark information, and generate asset feature matching information. An anomaly identification module is used to identify asset anomalies in the power industrial control network where the power industrial control network equipment is located, based at least on the asset feature matching information and according to a preset anomaly judgment priority, so as to obtain the asset anomaly type of the power industrial control network and the risk level corresponding to the asset anomaly type. The report generation module is used to generate an asset verification report based on the asset anomaly type and the risk level.

8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method described in any one of claims 1-6.

9. A computer program product comprising computer instructions, characterized in that, When the computer instructions are executed by the processor, they implement the method described in any one of claims 1-6.

10. A computer device, characterized in that, The method includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor, when executing the computer program, implements the method of any one of claims 1-6.