Abnormality detection method fusing knowledge distillation and group learning
By adopting anomaly detection methods that integrate knowledge distillation and group learning in 5G Internet of Things, using attention and graph convolution networks and multi-scale hollow convolution modules, combining group learning and smart contracts, traditional detection methods are difficult to cope with large-scale The problem of data changes has achieved efficient and secure abnormal detection and is suitable for 5G Internet of Things environment.
Patent Information
- Application Number
- CN202410988747.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-23
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-07-23
AI Technical Summary
When existing 5G IoT devices face large-scale, multi-dimensional and real-time changing data, traditional anomaly detection methods are difficult to effectively detect device failures and abnormal traffic, making it difficult to deal with network security threats.
Using anomaly detection methods that combine knowledge distillation and group learning, anomaly detection network based on attention and graph convolution is constructed. It is composed of modules of FastDTW, GCN, multi-scale hollow convolution and LSH attention, combining group learning network and smart contracts. , realize the optimization and update of the model to improve detection accuracy and efficiency.
It realizes distributed anomaly detection that is efficient, secure, scalable and privacy-protected in the 5G IoT environment, and can respond in real time and quickly adapt to dynamically changing network environments.
Smart Images

Figure CN119030743A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of Internet of Things and relates to an anomaly detection method integrating knowledge distillation and group learning. Background Art
[0002] 5G Internet of Things (5G IoT) is an IoT application based on the fifth generation of mobile communication technology. Due to the characteristics of high speed, low latency, large-scale connection, and high-density connection, 5G network can better meet the needs of IoT applications and is of great significance for achieving ultra-high-speed, high-reliability, and low-latency data transmission. Through the deep integration of IoT technology and 5G, 5G IoT has been widely used in various fields, such as smart transportation, smart home, smart medical care, smart factory, etc. However, with the introduction of new technologies and the continuous expansion of the scale of 5G IoT, 5G IoT is also facing new security threats. On the one hand, the access of large-scale IoT devices may cause network congestion, resulting in abnormal device traffic. On the other hand, IoT nodes sometimes exhibit abnormal behavior, resulting in network interruption, which may be caused by network attacks or device failures. The existing 5G IoT is usually composed of a large number of sensor devices, each of which generates traffic data in the form of time series. Therefore, through intelligent analysis of the time series traffic of 5G IoT devices, timely detection of abnormal traffic in the network is of great security significance for 5G IoT devices.
[0003] Due to the widespread use of 5G networks, the scale and complexity of IoT systems have increased significantly. However, the detection of device failures and abnormal traffic has become increasingly important and challenging. Traditional anomaly detection methods are often unable to cope with large-scale, multi-dimensional, and real-time changing data. To address these problems, advanced machine learning and artificial intelligence technologies have been introduced to improve the accuracy and efficiency of anomaly detection.
[0004] Knowledge distillation is an effective machine learning technique that extracts knowledge from a complex teacher model and transfers it to a smaller student model, which not only simplifies the computational complexity of the model but also retains the high performance of the teacher model. Group learning is a distributed machine learning method that combines the prediction results of multiple models to improve the robustness and accuracy of the overall model. In IoT anomaly detection, the fusion of knowledge distillation and group learning can make full use of the advantages of each model to build an efficient, accurate and stable anomaly detection framework to cope with the ever-changing and complex challenges in the 5G IoT environment. Summary of the invention
[0005] In view of this, the object of the present invention is to provide an anomaly detection method integrating knowledge distillation and group learning.
[0006] In order to achieve the above object, the present invention provides the following technical solutions:
[0007] An anomaly detection method integrating knowledge distillation and group learning includes the following steps:
[0008] S1: In the 5G Internet of Things scenario, an anomaly detection network based on attention and graph convolution is constructed; the anomaly detection network based on attention and graph convolution consists of a time series K-means clustering module based on FastDTW, a variable long-term correlation capture module based on GCN, a neighborhood information extraction module based on multi-scale hole convolution, and a long-term series reconstruction module based on LSH attention;
[0009] S2: Build a group learning network. The participants of the group learning network are composed of various edge servers of the Internet of Things. Anomaly detection teacher model and student model based on attention and graph convolution are deployed on each edge server, and the local teacher model is trained.
[0010] S3: Each edge server shares local teacher model parameters through the group network. The virtual center server aggregates the network parameters of each teacher to form the global teacher model parameters and sends them to each IoT edge server.
[0011] S4: Each edge server receives the global teacher model parameters and trains the local student network using the local dataset and the global teacher model;
[0012] S5: The local student network combines the abnormal conditions of neighboring nodes to learn real-time detection of abnormal traffic in the Internet of Things;
[0013] S6: Develop smart contracts, and the group network regularly updates the global teacher model and local student model to ensure the optimization and update of the global teacher model and local student model.
[0014] Furthermore, the FastDTW-based time series K-means clustering module uses FastDTW instead of the Euclidean distance to measure the similarity between time series, and the K-means layer clusters the time series variables.
[0015] Furthermore, the GCN-based variable long-term correlation capture module has a structure of multiple layers in parallel, and the structure of each layer is graph convolution layer → graph convolution layer; the hidden neurons of the first graph convolution layer are set to 2048, and the hidden neurons of the last convolution layer are set to 512; and each convolution layer is layer normalized and Leaky ReLU activated, and the last convolution layer is passed through a residual structure.
[0016] Furthermore, the neighborhood information extraction module based on multi-scale dilated convolution has a structure of the first linear layer → the first multi-scale extended convolution layer → the second linear layer → the second multi-scale extended convolution layer → the third linear layer → the third multi-scale extended convolution layer; the hidden neurons of each linear layer are set to 512, the sparsity of the multi-scale extended convolution layer is set to 1, 2, 3, and the receptive field is set to 3, 5, 7.
[0017] Furthermore, the LSH attention-based long-time series reconstruction module includes multiple random projection hash functions and a self-attention mechanism; the hash function is defined as:
[0018]
[0019] where q i is the eigenvector of the multivariate time series at time t, r i is a random vector drawn from a standard normal distribution, b i It is a random offset drawn from a uniform distribution. A hash function is used to divide the row vectors of the query matrix, key matrix, and value matrix of the sequence into multiple buckets according to their similarity, and self-attention is performed within the bucket to effectively capture long-distance dependencies.
[0020] Furthermore, in S2, an anomaly detection teacher network and a student network based on attention and graph convolution are deployed on each IoT edge server, and the teacher model is trained locally; the training loss of the teacher model is defined as:
[0021]
[0022] Where M represents the sample size, m represents the mth sample, represents the reconstructed output, represents the observed value, Indicates that the matrix size is n rows and m columns.
[0023] Furthermore, in S3, the group network includes multiple IoT edge servers, each of which is deployed with an anomaly detection teacher network and student network based on attention and graph convolution; an edge server is randomly selected as a virtual central server, and the teacher model parameters trained by other edge servers are sent to the virtual central server for aggregation, and the aggregation process adopts the weighted average method:
[0024]
[0025] Where φ gloabl represents the global teacher model parameters, φ local represents the local teacher model parameters of the edge server, s represents the sth edge server, α s ∈[0,1] represents the weight of the aggregation.
[0026] Further, in S4, each edge server receives the global teacher model sent by the virtual central server, uses the local data set and the global teacher model to guide the training of the local student model, and defines the distillation loss of the local student model as:
[0027]
[0028] Where M represents the sample size, k represents the kth sample, n represents the total number of rows in the matrix, i represents the i-th row in the matrix, m represents the total number of columns in the matrix, and j represents the j-th row in the matrix. represents the output of the teacher model, represents the output of the student model; the overall loss function of the student model is defined as the weighted sum of the distillation loss and the training loss:
[0029] L=α·L KD +β·L mse
[0030] Among them, α and β are weight coefficients.
[0031] Furthermore, in S5, the student network is responsible for detecting the traffic anomalies of IoT devices in real time. The degree of anomaly is quantified by an anomaly score. The anomaly score on the local edge server is defined as:
[0032]
[0033] Where n represents the total number of rows in the matrix, i represents the i-th row in the matrix, m represents the total number of columns in the matrix, and j represents the j-th row in the matrix. Indicates the observation window, represents the reconstruction sequence of the student model, Indicates that the matrix size is n rows and m columns. In order to combine the anomaly scores of neighboring devices for anomaly detection, the weighted average method is used to comprehensively consider the anomaly score of the device itself and the anomaly scores of its neighboring devices:
[0034]
[0035] where AS′ i is the comprehensive anomaly score of device i, AS i is the local anomaly score of device i, is the average anomaly score of the neighboring devices of device i, and α∈[0,1] is the weight coefficient.
[0036] Furthermore, in S6, due to the dynamic nature of the IoT environment, the detection accuracy of the global model may decrease over time, and the anomaly detection framework needs to be optimized regularly. By formulating smart contracts, the edge server continuously collects new traffic data, performs preprocessing, expands the local data set, and then uses the latest data to execute S2 to S6 to continuously train and update the local teacher model and student model. The optimization and update of the global teacher model and the local student model are ensured, thereby achieving efficient anomaly detection and rapid response.
[0037] The beneficial effects of the present invention are as follows: the present invention proposes an anomaly detection model IoT-FKGDL based on attention and graph convolution to address the problem of abnormal device traffic in the IoT environment. The model introduces FastDTW as a distance metric between time series variables, clusters highly correlated variables into the same cluster based on time series K-Means, and uses a graph convolutional network to capture the long-term correlation between variables in each cluster. Secondly, multi-scale hole convolution is used to extract the local correlation of the sequence, and the long-term correlation of multivariate sequences can be effectively captured in combination with local sensitive hashing attention. An IoT-FKGDL-SL anomaly detection framework under a group learning architecture that integrates knowledge distillation is also designed to establish a global anomaly detection framework for the IoT environment. The anomaly detection framework implements an efficient, secure, scalable and privacy-preserving distributed learning method. It is particularly suitable for application scenarios that require real-time response and distributed computing, such as anomaly detection of IoT device traffic.
[0038] Other advantages, objectives and features of the present invention will be described in the following description to some extent, and to some extent, will be obvious to those skilled in the art based on the following examination and study, or can be taught from the practice of the present invention. The objectives and other advantages of the present invention can be realized and obtained through the following description. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention will be described in detail below in conjunction with the accompanying drawings, wherein:
[0040] Figure 1 It is a schematic diagram of the process of the present invention;
[0041] Figure 2 It is the IoT-FKGDL anomaly detection model;
[0042] Figure 3 The IoT-FKGDL anomaly detection framework under the group learning architecture that integrates knowledge distillation. DETAILED DESCRIPTION
[0043] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments only illustrate the basic concept of the present invention in a schematic manner, and the following embodiments and features in the embodiments can be combined with each other without conflict.
[0044] Among them, the drawings are only used for illustrative explanations, and they only represent schematic diagrams rather than actual pictures, and should not be understood as limitations on the present invention. In order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of actual products. For those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.
[0045] The same or similar numbers in the drawings of the embodiments of the present invention correspond to the same or similar parts; in the description of the present invention, it should be understood that if the terms "upper", "lower", "left", "right", "front", "rear", etc. indicate the orientation or position relationship, they are based on the orientation or position relationship shown in the drawings, which is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operate in a specific orientation. Therefore, the terms describing the position relationship in the drawings are only used for illustrative purposes and cannot be understood as limiting the present invention. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to specific circumstances.
[0046] The present invention provides an anomaly detection method integrating knowledge distillation and group learning, comprising:
[0047] S1: In the 5G IoT scenario, we build an anomaly detection network based on attention and graph convolution, including:
[0048] 1) Based on the FastDTW time series K-means clustering module, FastDTW is used as the distance metric between sequences to cluster time series variables with high similarity into the same cluster. The similarity between each time series and the cluster center is:
[0049]
[0050] FastDTW(X i ,C j ) represents the time series X i and cluster center C j The FastDTW distance between them.
[0051] Calculate the mean value of each cluster element as the new cluster center:
[0052]
[0053] Where S j Indicates that it belongs to cluster C j The set of all time series.
[0054] 2) The GCN-based variable long-term correlation capture module is used to capture the correlation between the target variables in each cluster. There are as many parallel GCN networks as there are clusters. The structure of each GCN network layer is graph convolution layer → graph convolution layer; the hidden neurons of the first graph convolution layer are set to 2048, and the hidden neurons of the last convolution layer are set to 512; and each convolution layer is layer normalized and LeakyReLU activated, and the last convolution layer is passed through the residual structure.
[0055] The propagation rule for each convolutional layer is:
[0056]
[0057] in Represents cluster C j The adjacency matrix of the time series variables in I N is the identity matrix, and λ is a learnable parameter. is a trainable weight matrix, m represents the hidden feature dimension of this layer. RELU(·) is a nonlinear activation function.
[0058] 3) The neighborhood information extraction module based on multi-scale dilated convolution is used to capture and analyze the long-term dependencies and dynamic changes of multivariate long time series, so as to have a deeper understanding of data characteristics. Since the data generated by IoT devices is continuous, anomalies usually occur in local areas. The abnormal data points are highly correlated with the adjacent areas, but weakly correlated with the farther data points. This module expresses this neighbor relationship and uses it for anomaly detection. Its structure is first linear layer → first multi-scale dilated convolution layer → second linear layer → second multi-scale dilated convolution layer → third linear layer → third multi-scale dilated convolution layer; the hidden neurons of each linear layer are set to 512, the sparsity of the multi-scale dilated convolution layer is set to 1, 2, 3, and the receptive field is set to 3, 5, 7.
[0059] 4) The long-term series reconstruction module based on LSH self-attention is to solve the high computational complexity and memory consumption of the self-attention mechanism in long sequence modeling. As the length of the sequence increases, the amount of computation and memory consumption will increase at a quadratic level, and memory limitations will become the main bottleneck for processing long sequences. In practical applications, high computation and memory requirements will lead to high latency, which is difficult to meet the needs of real-time processing of IoT anomaly detection. This module includes multiple random projection hash functions and self-attention mechanisms. The hash function is defined as:
[0060]
[0061] where q i is the eigenvector of the multivariate time series at time t, r i is a random vector drawn from a standard normal distribution, b i It is a random offset drawn from a uniform distribution. A hash function is used to divide the row vectors of the query matrix, key matrix, and value matrix of the sequence into multiple buckets according to their similarity, and self-attention is performed within the bucket to effectively capture long-distance dependencies.
[0062] Finally, the time series K-means clustering module based on FastDTW, the variable long-term correlation capture module based on GCN, the neighborhood information extraction module based on multi-scale hole convolution, and the long-term series reconstruction module based on LSH attention are connected in sequence to form the anomaly detection network based on attention and graph convolution in the present invention.
[0063] S2: Build a group learning network and train a local teacher model, including:
[0064] 1) The participants of the group learning network are composed of various edge servers of the Internet of Things, and the anomaly detection teacher model and student model based on attention and graph convolution are deployed on each edge server.
[0065] 2) Each IoT edge server continuously monitors and collects traffic data as a participant in group learning, and the edge server pre-processes the collected data.
[0066] 3) Then run the anomaly detection algorithm locally to train the teacher model. The training loss of the teacher model is defined as:
[0067]
[0068] Where M represents the sample size, m represents the mth sample, represents the reconstructed output, represents the observed value, Indicates that the matrix size is n rows and m columns.
[0069] S3: Each edge server shares local teacher model parameters through the group network. The virtual center server aggregates the network parameters of each teacher to form the global teacher model parameters and sends them to each IoT edge server, including:
[0070] 1) The parameters of the local teacher model are updated to the group learning network and shared with other participants. An edge server is randomly selected as a virtual central server, responsible for aggregating the teacher model parameters uploaded by each edge server to form a global teacher model. The weighted average method is used to weight the global teacher model parameters according to the node data volume or model performance to ensure that the fused teacher model is more representative and robust. The aggregation process is expressed as:
[0071]
[0072] Where φ gloabl represents the global teacher model parameters, φ local represents the local teacher model parameters of the edge server, s represents the sth edge server, α s ∈[0,1] represents the weight of the aggregation.
[0073] 2) The virtual center server sends the fused global teacher model to each edge node through the group network.
[0074] S4: Each edge server receives the global teacher model parameters and uses the local data set and the global teacher model to train the local student network, including:
[0075] 1) Each edge server receives the global teacher model parameters sent by the virtual center server through the group network.
[0076] 2) Each edge node uses the local dataset and the global teacher model to guide the training of a lightweight student model and defines the distillation loss of the local student model:
[0077]
[0078] Where M represents the sample size, k represents the kth sample, n represents the total number of rows in the matrix, i represents the i-th row in the matrix, m represents the total number of columns in the matrix, and j represents the j-th row in the matrix. represents the output of the teacher model, Represents the output of the student model. The overall loss function of the student model is defined as the weighted sum of the distillation loss and the training loss:
[0079] L=α·L KD +β·L mse
[0080] Among them, α and β are weight coefficients.
[0081] S5: The local student network combines the abnormal conditions of neighboring nodes to learn to detect abnormal traffic in the Internet of Things in real time, including:
[0082] 1) The anomaly score of the local edge server is defined as:
[0083]
[0084] Where n represents the total number of rows in the matrix, i represents the i-th row in the matrix, m represents the total number of columns in the matrix, and j represents the j-th row in the matrix. Indicates the observation window, represents the reconstruction sequence of the student model, Indicates that the matrix size is n rows and m columns.
[0085] 2) Combine the anomaly scores of neighboring devices for anomaly detection, and use the weighted average method to comprehensively consider the anomaly score of the device itself and the anomaly scores of its neighboring devices:
[0086]
[0087] where AS′ i is the comprehensive anomaly score of device i, AS i is the local anomaly score of device i, is the average anomaly score of the neighboring devices of device i, and α∈[0,1] is the weight coefficient.
[0088] S6: Formulate smart contracts, and the group network regularly updates the global teacher model and local student model to ensure the optimization and update of the global teacher model and local student model, including:
[0089] 1) Due to the dynamic nature of the IoT environment, the detection accuracy of the global model may decrease over time. Therefore, the anomaly detection framework needs to be optimized regularly. S2-S5 is executed regularly, and the edge server continuously collects new traffic data and preprocesses it by formulating smart contracts to expand the local data set. Then, the local teacher model and student model are continuously trained and updated with the latest data to improve the adaptability and accuracy of the model. The virtual center server is responsible for regularly collecting and aggregating the latest teacher model parameters, forming a new global teacher model and issuing it. Ensure the optimization and update of the global teacher model to achieve efficient anomaly detection and rapid response.
[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solution, which should be included in the scope of the claims of the present invention.
Claims
1. An anomaly detection method integrating knowledge distillation and group learning, characterized by: The method comprises the following steps: S1: In the 5G Internet of Things scenario, an anomaly detection network based on attention and graph convolution is constructed; the anomaly detection network based on attention and graph convolution consists of a time series K-means clustering module based on Fast Dynamic Time Warping (FastDTW), a variable long-term correlation capture module based on Graph Convolutional Network (GCN), a neighborhood information extraction module based on multi-scale hole convolution, and a long-term series reconstruction module based on Locality-Sensitive Hashing (LSH) attention; S2: Build a group learning network. The participants of the group learning network are composed of various edge servers of the Internet of Things. Anomaly detection teacher model and student model based on attention and graph convolution are deployed on each edge server, and the local teacher model is trained. S3: Each edge server shares local teacher model parameters through the group network. The virtual center server aggregates the network parameters of each teacher to form the global teacher model parameters and sends them to each IoT edge server. S4: Each edge server receives the global teacher model parameters and trains the local student network using the local dataset and the global teacher model; S5: The local student network combines the abnormal conditions of neighboring nodes to learn real-time detection of abnormal traffic in the Internet of Things; S6: Develop smart contracts, and the group network regularly updates the global teacher model and local student model to ensure the optimization and update of the global teacher model and local student model.
2. The anomaly detection method integrating knowledge distillation and group learning according to claim 1, characterized in that: The FastDTW-based time series K-means clustering module uses FastDTW instead of Euclidean distance to measure the similarity between time series, and clusters time series variables in a K-means layer.
3. The anomaly detection method integrating knowledge distillation and group learning according to claim 1 is characterized in that: The GCN-based variable long-term correlation capture module has a structure of parallel multiple layers, and the structure of each layer is graph convolution layer → graph convolution layer; the hidden neurons of the first graph convolution layer are set to 2048, and the hidden neurons of the last convolution layer are set to 512; and each convolution layer is layer normalized and activated by a leaky linear rectifier unit (Leaky ReLU), and the last convolution layer is passed through a residual structure.
4. The anomaly detection method integrating knowledge distillation and group learning according to claim 1, characterized in that: The neighborhood information extraction module based on multi-scale dilated convolution has a structure of first linear layer→first multi-scale extended convolution layer→second linear layer→second multi-scale extended convolution layer→third linear layer→third multi-scale extended convolution layer; the number of hidden neurons in each linear layer is set to 512, the sparsity of the multi-scale extended convolution layer is set to 1, 2, 3, and the receptive field is set to 3, 5, 7.
5. The anomaly detection method integrating knowledge distillation and group learning according to claim 1, characterized in that: The LSH attention-based long-term series reconstruction module includes multiple random projection hash functions and a self-attention mechanism; the hash function is defined as: where q i is the eigenvector of the multivariate time series at time t, r i is a random vector drawn from a standard normal distribution, b i It is a random offset drawn from a uniform distribution. A hash function is used to divide the row vectors of the query matrix, key matrix, and value matrix of the sequence into multiple buckets according to their similarity, and self-attention is performed within the bucket to effectively capture long-distance dependencies.
6. The anomaly detection method integrating knowledge distillation and group learning according to claim 1, characterized in that: In S2, an anomaly detection teacher network and a student network based on attention and graph convolution are deployed on each IoT edge server, and the teacher model is trained locally; the training loss of the teacher model is defined as: Where M represents the sample size, m represents the mth sample, represents the reconstructed output, represents the observed value, Indicates that the matrix size is n rows and m columns.
7. The anomaly detection method integrating knowledge distillation and group learning according to claim 1 is characterized in that: In S3, the group network includes multiple IoT edge servers, each of which is deployed with an anomaly detection teacher network and student network based on attention and graph convolution; an edge server is randomly selected as the virtual central server, and the teacher model parameters trained by other edge servers are sent to the virtual central server for aggregation. The aggregation process adopts the weighted average method: Where φ gloabl represents the global teacher model parameters, φ local represents the local teacher model parameters of the edge server, s represents the sth edge server, α s ∈[0,1] represents the weight of the aggregation.
8. The anomaly detection method integrating knowledge distillation and group learning according to claim 1, characterized in that: In S4, each edge server receives the global teacher model sent by the virtual central server, uses the local data set and the global teacher model to guide the training of the local student model, and defines the distillation loss of the local student model as: Where M represents the sample size, k represents the kth sample, n represents the total number of rows in the matrix, i represents the i-th row in the matrix, m represents the total number of columns in the matrix, and j represents the j-th row in the matrix. represents the output of the teacher model, represents the output of the student model; the overall loss function of the student model is defined as the weighted sum of the distillation loss and the training loss: L=α·L KD +β·L mse Among them, α and β are weight coefficients.
9. The anomaly detection method integrating knowledge distillation and group learning according to claim 1, characterized in that: In S5, the student network is responsible for real-time detection of traffic anomalies of IoT devices. The degree of anomaly is quantified by an anomaly score. The anomaly score on the local edge server is defined as: Where n represents the total number of rows in the matrix, i represents the i-th row in the matrix, m represents the total number of columns in the matrix, and j represents the j-th row in the matrix. Indicates the observation window, represents the reconstruction sequence of the student model, Indicates that the matrix size is n rows and m columns. In order to combine the anomaly scores of neighboring devices for anomaly detection, the weighted average method is used to comprehensively consider the anomaly score of the device itself and the anomaly scores of its neighboring devices: where AS′ i is the comprehensive anomaly score of device i, AS i is the local anomaly score of device i, is the average anomaly score of the neighboring devices of device i, and α∈[0,1] is the weight coefficient.
10. The anomaly detection method integrating knowledge distillation and group learning according to claim 1, characterized in that: In S6, a smart contract is formulated to enable the edge server to continuously collect new traffic data, perform preprocessing, expand the local data set, and then use the latest data to execute S2 to S6 to continuously train and update the local teacher model and student model; Ensure the optimization and update of the global teacher model and local student model to achieve anomaly detection and response.
Citation Information
Patent Citations
Internet of Things intrusion detection method based on self-supervised learning and self-knowledge distillation
CN114861875A
Service function chain anomaly detection method in network function virtualization scene
CN116896498A
Industrial anomaly detection method and system based on multi-teacher network knowledge distillation
CN117609925A
Intrusion detection method under group learning architecture fused with knowledge distillation
CN117812593A
Cited By
Federal semi-supervised learning method, federal semi-supervised learning device and electronic equipment
CN119337973A
Marine organism community monitoring method and system based on memory hybrid prototype and knowledge distillation
CN120354318A
