A method for security detection of network-access edge devices based on boosting ensemble learning
By using the boosting ensemble learning method to perform security testing on grid-connected edge devices and constructing a scoring generator, the problem of differences among massive heterogeneous devices is solved, enabling accurate security assessment and abnormal device identification, thereby improving the security of the power grid system.
Patent Information
- Application Number
- CN202211592379.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-12-13
- Publication Date
- 2026-11-17
- Estimated Expiration
- 2042-12-13
AI Technical Summary
Existing technologies struggle to accurately assess the security of massive numbers of heterogeneous network-connected edge devices. They lack unified models and scoring standards, and the training set coverage is insufficient, making it impossible to effectively identify abnormal devices.
The boosting ensemble learning method is adopted. By performing ontological structure analysis on edge devices, a scoring parameter and a reliable scoring parameter vector matrix are constructed. An optimized boosting algorithm is used to train the generator, and security assessment is performed by combining device feature information.
It enables dynamic and continuous evaluation of grid-connected edge devices, accurately identifies abnormal devices and issues warnings, thereby improving the accuracy of power grid system security assessment.
Smart Images

Figure CN116012191B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of security detection of power system grid-connected edge equipment, specifically to a method for security detection of grid-connected edge equipment based on boosting ensemble learning. Background Technology
[0002] The rapid development of new power systems has brought about new characteristics such as a wider range of data acquisition and control objects, more source-grid-load-storage links, higher timeliness of business data, and stronger randomness on the power supply and load sides. These characteristics pose new requirements and challenges to the digital security protection of the power grid, including the continuous extension and blurring of traditional network boundaries, the lack of effective network security protection measures for large-scale social load aggregation systems, and insufficient protection of massive access and discretely distributed edge devices, which are prone to becoming attack springboards. Therefore, it is essential to focus on assessing the security issues of edge devices connected to the grid, and how to assess the security issues of edge devices connected to the grid has become an important research direction.
[0003] Due to the large number and diverse types of edge devices, accurately assessing the security of a network-connected edge device is a challenge. Existing research reveals several problems in assessing the security of network-connected edge devices: first, the variety and quantity of edge devices make it impossible to have a single model that perfectly fits these differences; second, there is no clear scoring standard to determine whether an edge device is within the security boundary; and third, the large amount of data makes it difficult to extract an effective and widely covered training set to train the model. Summary of the Invention
[0004] The purpose of this application is to provide a method for security detection of network access edge devices based on boosting ensemble learning, so as to realize dynamic and continuous evaluation and detection of the edge devices to be accessed.
[0005] To achieve the above objectives, this application provides the following technical solution:
[0006] This application provides a method for security detection of network-access edge devices based on boosting ensemble learning, including the following specific steps:
[0007] 1) Perform structural analysis on a large number of heterogeneous network-connected edge devices, extract feature information, and construct scoring parameters;
[0008] 2) Construction of the trusted scoring parameter vector matrix: Based on the scoring parameters obtained from various trusted edge devices collected in step 1), a scoring vector matrix is constructed using the scoring parameters of different trusted devices.
[0009] 3) Based on the above two steps, construct an edge device rating standard dataset, learn it using the optimized boosting algorithm, and build an edge device rating generator;
[0010] 4) Extract the feature information of the edge device to be connected and use it as the input of the edge device rating generator in step 3) to obtain the multi-dimensional rating and overall rating result of the device;
[0011] 5) Based on the output results of the edge device to be connected obtained in step 4), perform a static comparison with the results generated in step 3) to accurately identify abnormal devices and issue warnings.
[0012] The feature information in step 1) includes 13 features: the support status of four protocols in the data link layer (Ethernet, IEEE, PPP, HDLC), the support status of two protocols in the network layer (IP, ICMP), the support status of two protocols in the transport layer (TCP, UDP), the support status of two protocols in the application layer (HTTP, HTTPS), as well as the device serial number hash value, the device MAC address hash value, and the device communication port mapping value; wherein, the context network data packet is used to parse the above protocol support status features.
[0013] In step 2), the construction of the trusted scoring parameter vector matrix is based on the scoring parameters of various trusted edge devices collected in step 1), constructing a parameter vector matrix for each edge device. Since the network protocol in the device is parsed through data packets, each data packet will differ at different times. Therefore, m data packets are captured (m represents the number of captured data packets). For each data packet, ten protocol support status features (Ethernet, IEEE, PPP, HDLC, IP, ICMP, TCP, UDP, HTTP, HTTPS) are extracted. Combined with the device serial number hash value, the device network card MAC address hash value, and the device open port mapping value, a total of 13 features are extracted to form an m*13 feature vector matrix.
[0014] The edge device rating generator generated in step 3) based on boosting reinforcement learning is trained through the following process:
[0015] Step 1. First, the training set is the dataset of all the rating criteria obtained in Step 2), and a weight W is initialized, where , i represents the i-th training sample, Wi represents the weight of the i-th training sample, and W represents the sample weight vector composed of each Wi;
[0016] Step 2. Let j represent the current iteration round number and M represent the total number of iteration rounds. The iterative algorithm is trained in the Mth round of enhancement as follows:
[0017] Training a weighted weak classifier involves: where is the feature vector matrix of the training samples, represents the label information corresponding to the training samples, represents the current iteration round, is the th iteration, represents the current weak classifier obtained, is the current sample weight vector, and is the training function that trains the weak classifier based on the sample weights, features, and labels.
[0018] b: Predicted category label: ; where predict() is the prediction function that calls the current weak classifier Cj to predict the classification of feature X, and y on the left side of the formula is the predicted classification label.
[0019] c: Calculate the weight error rate: ;
[0020] d: Calculation coefficient: ;
[0021] e: Update weights: ;
[0022] f: Normalize the weights so that their sum is 1: ; where, this formula means dividing the updated sample weight vector W by the sum of all sample weights Wi, so that the sum of the normalized Wi is 1;
[0023] Step 3. Determine if the prediction accuracy meets the requirements. If not, return to Step 2 to continue iterative training. If the accuracy meets the requirements, combine the weak classifiers as follows, where x is the feature vector of a single input edge device, M is the total number of weak classifiers, and m is the weak classifier index. Let x be the classification output of the m-th weak classifier. Let m be the coefficient of the m-th weak classifier. The weight error rate of the m-th weak classifier:
[0024]
[0025] in , In this formula, =x→{-1,1} represents the classification result of the m-th weak classifier mapping the input feature vector x to -1 or 1.
[0026] In step 5), based on the scoring results obtained in step 4), the individual scoring results are compared first. If any one of them does not meet the requirements, the connection to the power grid system is directly rejected. If all individual scoring results meet the requirements, the overall scoring of all the scoring results of the device is performed to see if it is higher than the minimum value to determine whether the device can finally be connected to the power grid system.
[0027] Compared with the prior art, the beneficial effects of the present invention are:
[0028] This invention, based on boosting reinforcement learning, solves the problem of large differences in training sets. Through the constructed scoring algorithm, it effectively integrates the model and can accurately and quickly make an effective security assessment of edge devices that are continuously waiting to join the network. Attached Figure Description
[0029] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments of this application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.
[0030] Figure 1 This is a diagram of the boosting algorithm in this invention;
[0031] Figure 2 This is a schematic diagram illustrating the principle of the security detection method for network access edge devices according to the present invention. Detailed Implementation
[0032] The technical solutions of the embodiments of this application will now be described with reference to the accompanying drawings. It should be noted that similar reference numerals and letters in the following drawings indicate similar items; therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.
[0033] The terms “comprising,” “including,” or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase “comprising one…” does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0034] The terms “first,” “second,” etc., are used only to distinguish one entity or operation from another, and should not be construed as indicating or implying relative importance, nor as requiring or implying any such actual relationship or order between these entities or operations.
[0035] like Figure 1-2 As shown, the method for security detection of network-access edge devices based on boosting ensemble learning according to the present invention includes the following steps:
[0036] Step S1: Perform ontological structure analysis on a massive number of heterogeneous edge devices, extract feature information, and construct scoring parameters. The feature information includes 13 features: Ethernet, IEEE, PPP, and HDLC in the data link layer; IP and ICMP in the network layer; TCP and UDP in the transport layer; HTTP and HTTPS in the application layer; and device serial number hash value, device MAC address hash value, and device communication port mapping value. Among them, the context network data packet is used to parse the above protocol support status. The format of the obtained feature information is as follows:
[0037] Data link layer protocols: Etherent / IEEE / PPP / HDLC;
[0038] Network layer protocols: IP / ICMP;
[0039] Transport layer protocols: TCP / UDP;
[0040] Application layer protocols: HTTP / HTTPS;
[0041] The hash value of the device serial number: an integer.
[0042] Device MAC address hash value: Integer type;
[0043] Device communication ports: 0 / 1 / 2 / 3;
[0044] Step S2: Constructing the Trusted Scoring Parameter Vector Matrix. The hardware identification data extracted from the device's structure analysis mainly includes: device serial number, open ports, MAC address, various device communication protocols such as data link layer protocols (Ethernet / IEEE / PPP / HDLC), network layer protocols (IP / ICMP), transport layer protocols (TCP / UDP), and application layer protocols (HTTP / HTTPS), as well as context data packets. The device serial number is the device's unique machine code, the MAC address is the unique physical address on the device's network interface card, and the open communication ports are mapped to four levels based on their function: 0 (no port), 1 (port 0-1023), 2 (port 1024-49151), and 3 (port 49152-65535). When an edge device joins the network, it generates a large number of data packets. The context data packets generated by different devices in different environments will also differ. Based on the various communication protocols covered by the network data packets, common edge device communication protocol characteristics can be represented by binary 0 or 1, indicating whether the edge device supports the protocol. Since network protocols in a device are parsed via data packets, each data packet will differ at different times. Therefore, 13 data packets are captured, and for each data packet, ten protocol support status features (Ethernet, IEEE, PPP, HDLC, IP, ICMP, TCP, UDP, HTTP, HTTPS) are extracted. These features are then combined with the device serial number hash value, the device network card MAC address hash value, and the device open port mapping value to form a 13*13 feature vector matrix.
[0045]
[0046] Step S3: Based on the edge device rating standard dataset from steps S1-S2, use the optimized boosting algorithm to learn and construct an edge device rating generator. The edge device rating generator based on boosting reinforcement learning is trained through the following process:
[0047] 1) First, the training set is the dataset of all the scoring criteria obtained in step S2, and a weight W is initialized, where , i represents the i-th training sample, Wi represents the weight of the i-th training sample, and W represents the sample weight vector composed of each Wi;
[0048] 2) The iterative algorithm training is as follows:
[0049] For j in M rounds of enhancement (M is the total number of iterations, j is the current iteration round number):
[0050] a: Training a weighted weak classifier: (Where X is the feature vector matrix of the training samples, and y is the label information corresponding to the training samples,) Here, W is the current weak classifier obtained in the j-th round of training, W is the current sample weight vector, and train() is the training function that trains the weak classifier based on the sample weights W on the features X and the label y.
[0051] b: Predicted category label: ( The predicted classification label is the output of the `predict()` function, which is a prediction function that calls the current weak classifier Cj to predict the classification of feature X.
[0052] c: Calculate the weight error rate: ;
[0053] d: Calculation coefficient: ;
[0054] e: Update weights: ;
[0055] f: Normalize the weights so that their sum is 1: ; where, this formula means dividing the updated sample weight vector W by the sum of all sample weights Wi, so that the sum of the normalized Wi is 1;
[0056] 3) Determine if the prediction accuracy meets the requirements. If not, return to step 2) and continue iterative training. If the accuracy meets the requirements, combine the weak classifiers as follows (where... For the final strong classifier, x is the feature vector of a single input edge device, M is the total number of weak classifiers, and m is the index of the weak classifier. Let x be the classification output of the m-th weak classifier. Let m be the coefficient of the m-th weak classifier. (where the error rate is the weight of the m-th weak classifier)
[0057]
[0058] in , In this formula, Gm(x)=x→{-1,1} represents the classification result of the m-th weak classifier mapping the input feature vector x to -1 or 1.
[0059] Step S4: Based on the edge device rating generator generated in step S3, input the feature information of the edge device to be connected into the edge device rating generator, and obtain the various rating results and the overall rating result of the edge device according to the output results of the edge device rating generator;
[0060] Step S5: Based on the scoring results obtained in step S4, first compare the individual scoring results. If any one of them does not meet the requirements, the connection to the power grid system will be rejected directly. If all individual scoring results meet the requirements, the overall scoring of all the scoring results of the device will be performed to see if it is higher than the minimum value to determine whether the device can finally be connected to the power grid system.
[0061] The method for security detection of network access edge devices of this invention first performs on-structure analysis on a large number of edge devices that have already been securely connected to the network, such as optical transceivers, electrical transceivers, and data communication network devices. Thirteen features of all devices are extracted to construct scoring parameters, and a security device scoring standard dataset is generated. This dataset is then input into an optimized boosting algorithm for model training, resulting in an edge device scoring generator. As a specific application example, the method analyzes a switch device (actually a device already verified as secure) awaiting network access. Feature information of the switch device is extracted and mapped to a scoring parameter vector: if the switch supports Ethernet, IP, TCP, and HTTP protocols, the corresponding protocol feature positions are set to 1, and other protocol feature positions are set to 0; the device serial number hash value is 888, the device MAC address hash value is 128, and the device port number 64 is mapped to 1. After inputting the obtained 13-dimensional scoring vector into the scoring generator, both the output individual scores and the overall score fall within the range of security device scores. Finally, it is determined that the switch meets the category of secure devices and can be connected to the power system. This judgment result is consistent with the actual labels of known secure devices, thus verifying the feasibility of the network access device security monitoring method.
[0062] Compared with conventional edge device network access detection methods, this invention takes into account the differences of massive heterogeneous devices and the precise feature information of edge devices, enabling more granular and multi-faceted dynamic and continuous evaluation of the credibility of unknown edge devices.
[0063] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application.
Claims
1. A method for security detection of network-access edge devices based on boosting ensemble learning, characterized in that, The specific steps include the following: 1) Perform on-structure analysis on a large number of heterogeneous network-connected edge devices, extract feature information, and construct scoring parameters; the feature information in step 1) includes: data link layer protocol, network layer protocol, transport layer protocol, application layer protocol, device serial number, device open port, device MAC address, and context network data packet; among which, the data link layer protocols include: Etherent, IEEE, PPP, and HDLC; the network layer protocols include: IP and ICMP; the transport layer protocols include: TCP and UDP; and the application layer protocols include: HTTP and HTTPS; 2) Construction of the trusted scoring parameter vector matrix: Based on the scoring parameters obtained from various trusted edge devices collected in step 1), a scoring vector matrix is constructed using the scoring parameters of different trusted devices. 3) Based on the above two steps, construct an edge device rating standard dataset, learn it using the optimized boosting algorithm, and build an edge device rating generator; 4) Extract the feature information of the edge device to be connected and use it as the input of the edge device rating generator in step 3) to obtain the multi-dimensional rating and overall rating result of the device; 5) Based on the output results of the edge device to be connected obtained in step 4), perform a static comparison with the results generated in step 3) to accurately identify abnormal devices and issue warnings. In step 2), the construction of the trusted scoring parameter vector matrix involves building a parameter vector matrix for each edge device based on the scoring parameters of various trusted edge devices collected in step 1. Since the network protocol in the device is parsed through data packets, each data packet will differ at different times. Therefore, capturing... For each data packet, ten protocol support status features (Ethernet, IEEE, PPP, HDLC, IP, ICMP, TCP, UDP, HTTP, HTTPS) are extracted. These features are then combined with the device serial number hash, the device network card MAC address hash, and the device open port mapping value to extract a total of 13 features. eigenvector matrix: ; The edge device rating generator generated in step 3) based on boosting ensemble learning is trained through the following process: Step 1. First, the training set is the dataset of all the rating criteria obtained in Step 2), and a weight W is initialized, where ; Step 2. The iterative algorithm training is as follows: a: Training a weighted weak classifier: ; b: Predicted category label: ; c: Calculate the weight error rate ; d: Calculation coefficient: ; e: Update weights ; f: Normalize the weights so that their sum is ; Step 3. Determine if the prediction accuracy meets the requirements. If not, return to Step 2 to continue iterative training. If the accuracy meets the requirements, combine the weak classifiers as follows: , in , ; In step 5), based on the scoring results obtained in step 4), the individual scoring results are compared first. If any one of them does not meet the requirements, the connection to the power grid system is directly rejected. If all individual scoring results meet the requirements, the overall score of all individual scoring results of the device is calculated to see if it is higher than the minimum value to determine whether the device can ultimately be connected to the power grid system.
Citation Information
Patent Citations
Adaboost-based mobile intelligent terminal security level classification method
CN109344848A
Precise identification technology for terminal equipment in power grid
CN114611576A
