Abnormal network element detection method and device, equipment and storage medium
By dynamically detecting and iterating clustering of abnormal network elements, accurately locate and combat sample attacks, the security risks during model updates in federated learning scenarios are solved, the accuracy and defense capabilities of data analysis are improved, and the system complexity and attack range are reduced.
Patent Information
- Application Number
- CN202411117899.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-14
- Publication Date
- 2024-12-10
AI Technical Summary
In the horizontal federated learning scenario, local model communication and updates to combat security risks caused by sample attacks. The existing technology uses the introduction of special network functional interfaces to detect, but increases the system complexity and expands the scope of network attacks, and the detection accuracy is achieved. and efficiency is affected by the number of training samples.
By collecting network element feature groups, iteratively clustering and determining network element deviation values, dynamically detecting abnormal network elements, accurately locate network elements that may be trained using adversarial samples, and take measures to deal with it to reduce the negative impact of system performance.
It improves the accuracy and defense capabilities of data analysis, reduces the negative impact of system performance caused by the introduction of NF interfaces, accurately locates and handles abnormal network elements, and improves the defense capabilities for the risk of feature embedded data poisoning.
Smart Images
Figure CN119110292A_ABST
Abstract
Description
Background Art
[0002] To solve the communication privacy issue between multiple 5G network data analysis functions (NWDAFs), the 3rd Generation Partnership Project (3GPP) adopts a distributed architecture of federated learning. NWDAF network elements deployed on multiple clients in the same area collect local sample data, and use their respective sample data to train corresponding local models based on machine learning. The model adjustment parameters of multiple local models are then shared with NWDAF network elements deployed on the server to obtain a global model.
[0003] However, during the process of local model communication and updating, sample data is not protected. Once an attacker inputs adversarial samples into the local model, it will lead to inaccurate model training, and thus cause serious consequences such as performance degradation and misclassification.
[0004] At present, a network function (NF) interface specifically used to detect adversarial samples is often used to perform data detection on sample data collected by NWDAF network elements deployed on each client, filter out adversarial samples that may be intended to attack the network, and retain benign samples that have not attacked the network for model training.
[0005] The newly added NF interface increases the complexity of the system. The NF interface also faces the risk of being targeted by attackers, which further expands the scope of network attacks. In addition, the global detection of the NF interface will be affected by the increase in the number of training samples, and its detection accuracy and efficiency will also be affected. Summary of the invention
[0006] The embodiments of the present application provide an abnormal network element detection, device, equipment and storage medium to solve the security risk problems caused by adversarial sample attacks during model update communication in horizontal federated learning scenarios.
[0007] In a first aspect, an embodiment of the present application provides a method for detecting abnormal network elements, including:
[0008] Collecting network element feature groups of each network element to be detected in the current detection cycle, wherein each network element to be detected includes an original network element detected for the first time and an abnormal network element in a historical detection cycle, and each network element feature group is a plurality of model adjustment parameters obtained when a corresponding local model is trained based on sample data of a network element to be detected;
[0009] Based on the network element feature group of each network element to be detected and the network element feature group of the centroid of each round of network element cluster, iteratively clustering the network elements to be detected to obtain clustering results of each network element cluster;
[0010] Based on the network element characteristic distance between the network elements to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, the network element deviation value of each network element cluster is determined, and the network elements to be detected belonging to the network element cluster whose network element deviation value exceeds the set deviation threshold are regarded as abnormal network elements in the current detection cycle.
[0011] Optionally, the iterative clustering of the network elements to be detected based on the network element feature groups of the network elements to be detected and the network element feature groups of the centroids of each round of network element clusters to obtain clustering results of each network element cluster includes:
[0012] The network elements to be detected are clustered in a cyclic iteration manner until the iteration stop condition is met, and the clustering results of the network element clusters are output:
[0013] Based on the network element feature group of each network element to be detected and the network element feature group of the current round of network element cluster centroids of each network element cluster, determine the network element feature distance between each network element to be detected and the current round of network element cluster centroids, and divide each network element to be detected into a network element cluster whose network element feature distance does not exceed a set distance threshold;
[0014] Based on the network element feature groups of the network elements to be detected in each network element cluster, the centroids of each network element cluster are updated to obtain the next round of network element cluster centroids of each network element cluster.
[0015] Optionally, based on the network element feature group of the network element to be detected in each network element cluster, the centroid of each network element cluster is updated to obtain the next round of network element cluster centroids of each network element cluster, wherein the following operations are performed for each network element cluster respectively:
[0016] Determine a mean value of the network element feature group of the network element cluster based on the network element feature group of the network element to be detected in the network element cluster and the number of network elements in the cluster;
[0017] The network element feature group mean of the network element cluster is determined as the next round of network element cluster centroid of the network element cluster.
[0018] Optionally, the network element deviation value of each network element cluster is determined based on the network element characteristic distance between the network element to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, wherein the following operations are performed for each network element cluster respectively:
[0019] Based on the network element characteristic distance between the network element to be detected belonging to a network element cluster and the centroid of the last round of network element clusters in the cluster, determine the intra-cluster distance mean, intra-cluster distance difference and intra-cluster standard deviation of the network element cluster;
[0020] A network element deviation value of the network element cluster is determined based on the intra-cluster distance mean, the intra-cluster deviation from the mean, and the intra-cluster standard deviation of the network element cluster.
[0021] Optionally, the method further includes:
[0022] Sending a stop instruction to a management network element for managing multiple network elements to be detected, instructing the management network element to prohibit the abnormal network element from performing model training;
[0023] And, an isolation instruction is issued to the management network element, instructing the management network element to remove the network elements to be detected that are detected as abnormal network elements for multiple times.
[0024] In a second aspect, an embodiment of the present application further provides an abnormal network element detection device, including:
[0025] A feature extraction unit is used to collect network element feature groups of each network element to be detected in the current detection cycle, wherein each network element to be detected includes an original network element detected for the first time and an abnormal network element in a historical detection cycle, and each network element feature group is a plurality of model adjustment parameters obtained when a corresponding local model is trained based on sample data of a network element to be detected;
[0026] A feature detection unit, configured to iteratively cluster the network elements to be detected based on the network element feature groups of the network elements to be detected and the network element feature groups of the centroids of each round of network element clusters, to obtain clustering results of each network element cluster;
[0027] Based on the network element characteristic distance between the network elements to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, the network element deviation value of each network element cluster is determined, and the network elements to be detected belonging to the network element cluster whose network element deviation value exceeds the set deviation threshold are regarded as abnormal network elements in the current detection cycle.
[0028] Optionally, the feature detection unit is used to:
[0029] The network elements to be detected are clustered in a cyclic iteration manner until the iteration stop condition is met, and the clustering results of the network element clusters are output:
[0030] Based on the network element feature group of each network element to be detected and the network element feature group of the current round of network element cluster centroids of each network element cluster, determine the network element feature distance between each network element to be detected and the current round of network element cluster centroids, and divide each network element to be detected into a network element cluster whose network element feature distance does not exceed a set distance threshold;
[0031] Based on the network element feature groups of the network elements to be detected in each network element cluster, the centroids of each network element cluster are updated to obtain the next round of network element cluster centroids of each network element cluster.
[0032] Optionally, the feature detection unit performs the following operations for each network element cluster:
[0033] Determine a mean value of the network element feature group of the network element cluster based on the network element feature group of the network element to be detected in the network element cluster and the number of network elements in the cluster;
[0034] The network element feature group mean of the network element cluster is determined as the next round of network element cluster centroid of the network element cluster.
[0035] Optionally, the feature detection unit performs the following operations for each network element cluster:
[0036] Based on the network element characteristic distance between the network element to be detected belonging to a network element cluster and the centroid of the last round of network element clusters in the cluster, determine the intra-cluster distance mean, intra-cluster distance difference and intra-cluster standard deviation of the network element cluster;
[0037] A network element deviation value of the network element cluster is determined based on the intra-cluster distance mean, the intra-cluster deviation from the mean, and the intra-cluster standard deviation of the network element cluster.
[0038] Optionally, the abnormal network element detection device further includes a poisoning detection unit, and the poisoning detection unit is used to:
[0039] Sending a stop instruction to a management network element for managing multiple network elements to be detected, instructing the management network element to prohibit the abnormal network element from performing model training;
[0040] And, an isolation instruction is issued to the management network element, instructing the management network element to remove the network elements to be detected that are detected as abnormal network elements for multiple times.
[0041] In a third aspect, an embodiment of the present application further provides a computer device, comprising a processor and a memory, wherein the memory stores program code, and when the program code is executed by the processor, the processor executes the steps of any one of the above-mentioned abnormal network element detection methods.
[0042] In a fourth aspect, an embodiment of the present application further provides a computer-readable storage medium, which includes a program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of any one of the above-mentioned abnormal network element detection methods.
[0043] The beneficial effects of this application are as follows:
[0044] The embodiments of the present application provide a method, apparatus, device and storage medium for detecting abnormal network elements, the method comprising: collecting network element feature groups of each network element to be detected in a current detection cycle, each network element to be detected including an original network element detected for the first time and abnormal network elements in a historical detection cycle, each network element feature group being a plurality of model adjustment parameters obtained when a corresponding local model is trained based on sample data of a network element to be detected; then, based on the network element feature groups of each network element to be detected and the network element feature groups of the centroids of each round of network element clusters, iteratively clustering each network element to be detected is performed to obtain clustering results of each network element cluster; finally, based on the network element feature distance between the network element to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, determining the network element deviation value of each network element cluster, and treating the network element to be detected belonging to the network element cluster whose network element deviation value exceeds the set deviation threshold as the abnormal network element in the current detection cycle.
[0045] In view of the security risks such as adversarial sample attacks that may exist during model update communication in horizontal federated learning scenarios, this application makes full use of the characteristics of large feature overlaps and large deviations of abnormal parameters in horizontal federated learning scenarios, dynamically detects the network element feature groups of each network element to be detected in the current detection cycle, and accurately locates abnormal network elements that may use adversarial samples for model training by performing feature analysis on each network element feature group. Corresponding measures are taken for network elements with different levels of security risks to ensure the accuracy of data analysis, improve the NWDAF network element's defense capabilities against data poisoning risks such as feature embedding, and at the same time reduce the negative impact of the system performance caused by the additional introduction of NF interfaces.
[0046] Other features and advantages of the present application will be described in the subsequent description, and partly become apparent from the description, or be understood by practicing the present application. The purpose and other advantages of the present application can be realized and obtained by the structures specifically pointed out in the written description, claims, and drawings. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] The drawings described herein are used to provide a further understanding of the present application and constitute a part of the present application. The illustrative embodiments of the present application and their descriptions are used to explain the present application and do not constitute an improper limitation on the present application. In the drawings:
[0048] Figure 1 A schematic diagram of the structure of the management NWDAF network element provided in the embodiment of the present application;
[0049] Figure 2A A schematic diagram of a process for detecting abnormal network elements provided in an embodiment of the present application;
[0050] Figure 2B A logical diagram of abnormal network element detection provided in an embodiment of the present application;
[0051] Figure 2C A schematic diagram of a process for iteratively clustering each network element to be detected provided in an embodiment of the present application;
[0052] Figure 3 A schematic diagram of a process for managing NWDAF to detect abnormal network elements provided in an embodiment of the present application;
[0053] Figure 4 A schematic diagram of the structure of an abnormal network element detection device provided in an embodiment of the present application;
[0054] Figure 5 A schematic diagram of the structure of a computer device provided in an embodiment of the present application;
[0055] Figure 6 Schematic diagram of the structure of a computing device in an embodiment of the present application. DETAILED DESCRIPTION
[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the technical solution of the present application, rather than all of the embodiments. Based on the embodiments recorded in the application documents, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the technical solution of the present application.
[0057] Some of the terms used in the embodiments of the present application are explained below to facilitate understanding by those skilled in the art.
[0058] 1.1 Artificial Intelligence (AI):
[0059] Artificial intelligence is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.
[0060] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, and mechatronics; artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning. With the research and progress of artificial intelligence technology, artificial intelligence has been studied and applied in many fields, such as common smart homes, smart customer service, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, robots, smart medical care, etc. It is believed that with the development of technology, artificial intelligence will be applied in more fields and play an increasingly important role.
[0061] 2. Machine Learning (ML):
[0062] Machine learning is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.
[0063] 3. Federated Learning
[0064] Federated learning is a distributed machine learning method that aims to protect data privacy and security by using encryption and de-identification technologies. The principle is to disperse the model training process among multiple participants, each participant trains the model locally, and aggregates the model parameters updated by multiple participants to form a global model. This method enables multiple participants to learn together without sharing the original data involving data privacy, and is suitable for cross-organizational or cross-border collaborative learning scenarios.
[0065] 4. NWDAF:
[0066] In the future, communication scenarios will be complex and business needs will be diverse. 5G networks will need to intelligently provide on-demand services and higher network resource utilization efficiency. When designing 5G networks, we should consider introducing artificial intelligence and big data analysis technologies to analyze, reason, judge and deal with mobile network communication data, so as to improve network resource utilization and management efficiency. Therefore, 3GPP introduced the big data analysis network element NWDAF in the 5G core network to provide data analysis services to the network, forming a closed-loop control of data collection, training and reasoning, as well as a network big data analysis system architecture that supports diversified solutions.
[0067] As the AI and big data engine of 5G networks, NWDAF is responsible for collecting network management, business, network, and terminal data, performing data model training, outputting prediction, reasoning, and decision results, assisting network operation and maintenance, and promoting the rational allocation of 5G network resources.
[0068] 5. Adversarial Examples
[0069] Adversarial samples are samples that add some tiny perturbations that are imperceptible to the human eye to the original sample and cause the model to make incorrect inferences. When an attacker generates adversarial samples to cause the machine learning model to make a wrong judgment, it is an adversarial attack.
[0070] Adversarial attacks can be divided into white-box attacks and black-box attacks: a white-box attack means that the attacker can know the machine learning model used, the parameters used by the model, and can interact with the machine learning system; a black-box attack means that the attacker does not know the model and parameters used by machine learning, but can obtain some information through interaction with the machine learning system, such as observing the output and judging the output by passing in arbitrary input.
[0071] The following is a brief introduction to the design concept of the embodiment of the present application:
[0072] To solve the communication privacy issue between NWDAFs in multiple 5G networks, 3GPP adopts a distributed architecture of federated learning. NWDAF network elements deployed on multiple clients in the same area collect local sample data, use their respective sample data to train corresponding local models based on machine learning, and then share the model adjustment parameters of multiple local models with NWDAF network elements deployed on the server to obtain a global model.
[0073] However, during the process of local model communication and updating, sample data is not protected. Once an attacker inputs adversarial samples into the local model, it will lead to inaccurate model training, and thus cause serious consequences such as performance degradation and misclassification.
[0074] At present, NF interfaces specifically used to detect adversarial samples are often used to perform data detection on sample data collected by NWDAF network elements deployed on each client, filter out adversarial samples that may be intended to attack the network, and retain benign samples that have not attacked the network for model training.
[0075] The newly added NF interface increases the complexity of the system. The NF interface also faces the risk of being targeted by attackers, which further expands the scope of network attacks. In addition, the global detection of the NF interface will be affected by the increase in the number of training samples, and its detection accuracy and efficiency will also be affected.
[0076] In view of this, the embodiments of the present application provide a method, device, equipment and storage medium for detecting abnormal network elements. The method specifically includes: collecting network element feature groups of each network element to be detected in the current detection cycle, each network element to be detected includes the original network element detected for the first time and the abnormal network element in the historical detection cycle, and each network element feature group is a plurality of model adjustment parameters obtained when the corresponding local model is trained based on the sample data of a network element to be detected; then, based on the network element feature groups of each network element to be detected and the network element feature groups of the centroid of each round of network element clusters, each network element to be detected is iteratively clustered to obtain the clustering results of each network element cluster; finally, based on the network element feature distance between the network element to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, the network element deviation value of each network element cluster is determined, and the network element to be detected belonging to the network element cluster whose network element deviation value exceeds the set deviation threshold is regarded as the abnormal network element in the current detection cycle.
[0077] In view of the security risks such as adversarial sample attacks that may exist during model update communication in horizontal federated learning scenarios, this application makes full use of the characteristics of large feature overlaps and large deviations of abnormal parameters in horizontal federated learning scenarios, dynamically detects the network element feature groups of each network element to be detected in the current detection cycle, and accurately locates abnormal network elements that may use adversarial samples for model training by performing feature analysis on each network element feature group. Corresponding measures are taken for network elements with different levels of security risks to ensure the accuracy of data analysis, improve the NWDAF network element's defense capabilities against data poisoning risks such as feature embedding, and at the same time reduce the negative impact of the system performance caused by the additional introduction of NF interfaces.
[0078] The preferred embodiments of the present application are described below in conjunction with the drawings in the specification. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present application, and are not used to limit the present application. In addition, the embodiments and features in the embodiments of the present application may be combined with each other if there is no conflict.
[0079] like Figure 1As shown, the present application adds a feature handling module, a sample label library and a poisoning detection module to the management NWDAF network element of the server. Among them, the feature handling module specifically includes a feature extraction unit and a feature detection unit. The feature extraction unit is used to collect the network element feature group of each network element to be detected in the current detection cycle from the data collection module of the management NWDAF network element; the feature detection unit is used to perform feature analysis on each network element feature group, and accurately locate the abnormal network element that may use adversarial samples for model training. The sample label library is used to record the network element detection labels of each unit to be detected in each detection cycle. The network element detection labels include benign network elements and abnormal network elements. The abnormal network elements are further divided into suspicious network elements and attacking network elements. The poisoning detection module is used to continuously detect suspicious network elements, issue a stop instruction to the management network element used to manage multiple network elements to be detected, and instruct the management network element to prohibit the suspicious network element from performing model training; and issue an isolation instruction to the management network element, instructing the management network element to remove the network elements to be detected that have been detected as abnormal network elements for many times.
[0080] Next, combine Figure 2A-2B The schematic diagram shown specifically introduces the process in which the NWDAF network element deployed on the server performs abnormality detection on the NWDAF network elements on each client.
[0081] S201: Collect network element feature groups of each network element to be detected in the current detection cycle, where each network element to be detected includes the original network element detected for the first time and the abnormal network elements in the historical detection cycle. Each network element feature group is a plurality of model adjustment parameters obtained when training a corresponding local model based on sample data of a network element to be detected.
[0082] Through the feature extraction unit in the feature handling module, the network element feature groups of each network element to be detected in the current detection cycle are collected from the original data collection module of the management NWDAF network element. As shown in Table 1, each network element feature group is a plurality of model adjustment parameters obtained when the corresponding local model is trained based on the sample data of a network element to be detected.
[0083] Table 1
[0084]
[0085] S202: performing iterative clustering on each network element to be detected based on the network element feature group of each network element to be detected and the network element feature group of the centroid of each round of network element clusters to obtain a clustering result of each network element cluster.
[0086] This application makes full use of the characteristics of large feature overlap and large deviation of abnormal parameters in horizontal federated learning scenarios. The feature detection unit uses clustering analysis method to divide network elements that use similar sample data for model training into a network element cluster. There are five main clustering analysis methods: partition method, hierarchical method, density-based method, grid-based method, and model-based method. Figure 2C Taking the K-means algorithm in the partitioning method as an example, the network elements to be detected are clustered in a cyclic iteration manner until the iteration stop condition is met, and the clustering results of each network element cluster are output:
[0087] S2021: Based on the network element feature group of each network element to be detected and the network element feature group of the current round of network element cluster centroids of each network element cluster, determine the network element feature distance between each network element to be detected and the centroid of each current round of network element clusters, and divide each network element to be detected into a network element cluster whose network element feature distance does not exceed the set distance threshold.
[0088] Assume that the network element feature group of the network element a to be detected is The network element feature group of network element cluster centroid 1 is The network element characteristic distance d between the network element a to be detected and the network element cluster centroid 1 is calculated using formula 1. When there are multiple network element cluster centroids, the network element characteristic distance d between the network element a to be detected and the centroids of each network element cluster is calculated respectively, and the network element a to be detected is divided into the network element cluster with the minimum network element characteristic distance.
[0089]
[0090] S2022: Based on the network element feature groups of the network elements to be detected in each network element cluster, the centroid of each network element cluster is updated to obtain the next round of network element cluster centroids of each network element cluster.
[0091] The centroid of the first-round network element cluster is randomly selected from multiple network elements to be detected, and the centroid of the non-first-round network element cluster is updated based on the network element feature group of the network elements to be detected in each network element cluster. As shown in Formula 2, the following operations are performed for each network element cluster:
[0092] Based on the network element feature group of the network element i to be detected in a network element cluster The mean value of the network element feature group of a network element cluster is determined by the number of network elements n in the cluster. The mean value of the network element feature group of the network element cluster is determined as the next round of network element cluster centroid of the network element cluster.
[0093]
[0094] S2023: Determine whether the iteration stop condition is met, if so, output the clustering result of each network element cluster; otherwise, jump to step 2021.
[0095] Determine whether the centroid of the current round of network element clusters of a network element cluster is the same as the centroid of the next round of network element clusters. If they are the same, it means that the clustering iteration has converged and the clustering results of each network element cluster can be output; otherwise, it means that the clustering iteration has not ended and it is necessary to jump to step 2021 and repeat steps 2021-2023.
[0096] S203: Determine the network element deviation value of each network element cluster based on the network element characteristic distance between the network element to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, and regard the network element to be detected belonging to the network element cluster whose network element deviation value exceeds the set deviation threshold as the abnormal network element in the current detection cycle.
[0097] In step 203, the feature detection unit performs the following operations for each network element cluster:
[0098] Based on the network element characteristic distance between the network element to be detected belonging to a network element cluster and the centroid of the last round of network element clusters in the cluster, the intra-cluster distance mean μ, the intra-cluster deviation from the mean (X-μ) and the intra-cluster standard deviation σ of the network element cluster are determined; as shown in Formula 3, based on the intra-cluster distance mean μ, the intra-cluster deviation from the mean (X-μ) and the intra-cluster standard deviation σ of the network element cluster, the network element deviation value Z of the network element cluster is determined.
[0099] The intra-cluster deviation from the mean (X-μ) is a statistical indicator that measures the relative value between an individual and a group. It is an important indicator used to measure the differences between each network element in a cluster and the group. The intra-cluster standard deviation σ is a statistical indicator used to reflect the degree of dispersion of the network element cluster. The network element deviation value Z is used to indicate the standardized deviation between the network elements in the cluster and the mean distance within the cluster.
[0100] Z=(X-μ) / σ Formula 3;
[0101] The network elements to be detected in the network element cluster whose network element deviation value does not exceed the set deviation threshold are regarded as benign network elements in the current detection cycle, and the benign network elements are allowed to continue to participate in model training. The network elements to be detected in the network element cluster whose network element deviation value exceeds the set deviation threshold are regarded as abnormal network elements in the current detection cycle, and the poisoning detection unit focuses on detecting them until the global model is aggregated or they are detected as benign network elements multiple times. Then, the network element detection labels of each network element to be detected in the current detection cycle are pushed to the sample label library, and registered by the sample label library for easy query and management.
[0102] For the network elements to be detected that are detected as abnormal network elements for the first time, they are marked as suspicious network elements, and the poisoning detection unit will continue to detect them until the global model aggregation is completed. At the same time, the poisoning detection unit sends a stop instruction to the management network element used to manage multiple network elements to be detected, instructing the management network element to prohibit the abnormal network element from model training until the network element is detected as a benign network element multiple times, and then it is allowed to participate in model training.
[0103] For the network elements to be detected that are detected as abnormal network elements multiple times, they are marked as attacking network elements, and the poisoning detection unit sends an isolation instruction to the management network element, instructing the management network element to remove the network elements to be detected that are detected as abnormal network elements multiple times.
[0104] Combination Figure 3 The flowchart shown provides an implementation process of applying the method of the present application to a practical scenario.
[0105] S301: The feature extraction unit in the feature handling module collects the network element feature groups of each network element to be detected in the current detection cycle from the original data collection module managing the NWDAF network element.
[0106] S302: The feature detection unit in the feature handling module iteratively clusters each network element to be detected based on the network element feature group of each network element to be detected and the network element feature group of the centroid of each round of network element clusters to obtain clustering results of each network element cluster.
[0107] S303: Determine the network element deviation value of each network element cluster based on the network element feature distance between the network element to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster through the feature detection unit in the feature handling module.
[0108] S304: Determine whether the network element deviation value exceeds the set deviation threshold 3.0. If so, execute step 305; otherwise, jump to step 306.
[0109] S305: Mark the network elements to be detected in the network element cluster as suspicious network elements in the current detection cycle, record the network element detection labels of the network elements to be detected in the network element cluster in the sample label library, and continuously detect related network elements through the poisoning detection module, and instruct the management NWDAF network element to prohibit related network elements from participating in model training.
[0110] S306: Mark the network elements to be detected in the network element cluster as benign network elements in the current detection cycle, record the network element detection labels of the network elements to be detected in the network element cluster in the sample label library, and continuously detect related network elements through the poisoning detection module, instructing the management NWDAF network element to allow related network elements to participate in model training.
[0111] S307: Determine whether the network element to be detected in the network element cluster is detected as a suspicious network element twice in succession. If so, execute step 308; otherwise, execute step 309.
[0112] S308: The poisoning detection module issues an isolation instruction to instruct the management NWDAF network element to remove the related network elements.
[0113] S309: Continuously detect the detection results of the relevant network elements in the next detection cycle through the poisoning detection module.
[0114] Based on the same inventive concept as the above method embodiment, the present application embodiment also provides a structural schematic diagram of an abnormal network element detection device. Figure 4 As shown, the abnormal network element detection device 400 may include:
[0115] The feature extraction unit 401 is used to collect network element feature groups of each network element to be detected in the current detection cycle, wherein each network element to be detected includes an original network element detected for the first time and an abnormal network element in a historical detection cycle, and each network element feature group is a plurality of model adjustment parameters obtained when a corresponding local model is trained based on sample data of a network element to be detected;
[0116] The feature detection unit 402 is used to iteratively cluster the network elements to be detected based on the network element feature group of each network element to be detected and the network element feature group of the centroid of each round of network element clusters to obtain clustering results of each network element cluster;
[0117] Based on the network element characteristic distance between the network elements to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, the network element deviation value of each network element cluster is determined, and the network elements to be detected belonging to the network element cluster whose network element deviation value exceeds the set deviation threshold are regarded as abnormal network elements in the current detection cycle.
[0118] Optionally, the feature detection unit 402 is used to:
[0119] The network elements to be detected are clustered in a cyclic iteration manner until the iteration stop condition is met, and the clustering results of the network element clusters are output:
[0120] Based on the network element feature group of each network element to be detected and the network element feature group of the current round of network element cluster centroids of each network element cluster, determine the network element feature distance between each network element to be detected and the current round of network element cluster centroids, and divide each network element to be detected into a network element cluster whose network element feature distance does not exceed a set distance threshold;
[0121] Based on the network element feature groups of the network elements to be detected in each network element cluster, the centroids of each network element cluster are updated to obtain the next round of network element cluster centroids of each network element cluster.
[0122] Optionally, the feature detection unit 402 performs the following operations for each network element cluster:
[0123] Determine a mean value of the network element feature group of the network element cluster based on the network element feature group of the network element to be detected in the network element cluster and the number of network elements in the cluster;
[0124] The network element feature group mean of the network element cluster is determined as the next round of network element cluster centroid of the network element cluster.
[0125] Optionally, the feature detection unit 402 performs the following operations for each network element cluster:
[0126] Based on the network element characteristic distance between the network element to be detected belonging to a network element cluster and the centroid of the last round of network element clusters in the cluster, determine the intra-cluster distance mean, intra-cluster distance difference and intra-cluster standard deviation of the network element cluster;
[0127] A network element deviation value of the network element cluster is determined based on the intra-cluster distance mean, the intra-cluster deviation from the mean, and the intra-cluster standard deviation of the network element cluster.
[0128] Optionally, the abnormal network element detection device 400 further includes a poisoning detection unit 403, and the poisoning detection unit 403 is used to:
[0129] Sending a stop instruction to a management network element for managing multiple network elements to be detected, instructing the management network element to prohibit the abnormal network element from performing model training;
[0130] And, an isolation instruction is issued to the management network element, instructing the management network element to remove the network elements to be detected that are detected as abnormal network elements for multiple times.
[0131] For the convenience of description, the above parts are divided into modules (or units) according to their functions and described separately. Of course, when implementing this application, the functions of each module (or unit) can be implemented in the same or multiple software or hardware.
[0132] After introducing the abnormal network element detection method and apparatus according to the exemplary embodiment of the present application, next, a computer device according to another exemplary embodiment of the present application is introduced.
[0133] Those skilled in the art will appreciate that various aspects of the present application may be implemented as a system, method or program product. Therefore, various aspects of the present application may be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to as "circuit", "module" or "system" herein.
[0134] Based on the same inventive concept as the above method embodiment, the present application embodiment also provides a computer device, see Figure 5 As shown, the computer device 500 may include at least a processor 501 and a memory 502. The memory 502 stores program codes, and when the program codes are executed by the processor 501, the processor 501 executes any one of the steps of the above abnormal network element detection method.
[0135] In some possible implementations, the computing device according to the present application may include at least one processor and at least one memory. The memory stores program code, and when the program code is executed by the processor, the processor executes the steps of the abnormal network element detection method according to various exemplary implementations of the present application described above in this specification. For example, the processor may execute the following steps: Figure 2A Follow the steps shown in .
[0136] Refer to the following Figure 6 hereinafter, a computing device 600 according to this embodiment of the present application is described. Figure 6 The computing device 600 is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.
[0137] like Figure 6 As shown, the computing device 600 is in the form of a general computing device. The components of the computing device 600 may include but are not limited to: at least one processing unit 601, at least one storage unit 602, and a bus 603 connecting different system components (including the storage unit 602 and the processing unit 601).
[0138] Bus 603 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, a processor, or a local bus using any of a variety of bus architectures.
[0139] The storage unit 602 may include a readable medium in the form of a volatile memory, such as a random access memory (RAM) 6021 and / or a cache memory unit 6022 , and may further include a read-only memory (ROM) 6023 .
[0140] The storage unit 602 may also include a program / utility 6025 having a set (at least one) of program modules 6024, such program modules 6024 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.
[0141] The computing device 600 may also communicate with one or more external devices 604 (e.g., keyboards, pointing devices, etc.), one or more devices that enable a user to interact with the computing device 600, and / or any device that enables the computing device 600 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 605. In addition, the computing device 600 may also communicate with one or more networks (e.g., a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) via a network adapter 606. As shown, the network adapter 606 communicates with other modules for the computing device 600 via a bus 603. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the computing device 600, including but not limited to: microcode, device drivers, redundant processors, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.
[0142] Based on the same inventive concept as the above method embodiment, various aspects of the abnormal network element detection method provided in the present application can also be implemented in the form of a program product, which includes a program code. When the program product is run on a computer device, the program code is used to enable the computer device to execute the steps of the abnormal network element detection method according to various exemplary embodiments of the present application described above in this specification. For example, the computer device can execute the following steps: Figure 2A Follow the steps shown in .
[0143] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
[0144] Although the preferred embodiments of the present application have been described, those skilled in the art may make additional changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of the present application.
[0145] Obviously, those skilled in the art can make various changes and modifications to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and variations of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.
Claims
1. A method for detecting abnormal network elements, characterized in that: include: Collecting network element feature groups for each network element to be detected in the current detection cycle, wherein each network element to be detected includes the original network element detected for the first time and the abnormal network elements in the historical detection cycle. Each network element feature group is a plurality of model adjustment parameters obtained when training the corresponding local model based on sample data of a network element to be detected; Iteratively clustering the network elements to be detected based on the network element feature groups of the network elements to be detected and the network element feature groups of the centroids of the network element clusters in each round to obtain clustering results of the network element clusters; Based on the network element characteristic distance between the network element to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, the network element deviation value of each network element cluster is determined, and the network element to be detected belonging to the network element cluster whose network element deviation value exceeds the set deviation threshold is regarded as the abnormal network element in the current detection cycle.
2. The method according to claim 1, wherein The iterative clustering of the network elements to be detected based on the network element feature groups of the network elements to be detected and the network element feature groups of the centroids of the network element clusters in each round to obtain clustering results of the network element clusters includes: Cluster the network elements to be detected in a cyclic iterative manner until the iteration stop condition is met, and output the clustering results of the network element clusters: Determine, based on the network element feature group of each network element to be detected and the network element feature group of the current round of network element cluster centroids of each network element cluster, the network element feature distance between each network element to be detected and the current round of network element cluster centroids, and classify each network element to be detected into a network element cluster whose network element feature distance does not exceed a set distance threshold; Based on the network element feature groups of the network elements to be detected in each network element cluster, the centroid of each network element cluster is updated to obtain the next round of network element cluster centroids of each network element cluster.
3. The method according to claim 2, wherein The centroid of each network element cluster is updated based on the network element feature group of the network element to be detected in each network element cluster to obtain the next round of network element cluster centroids of each network element cluster, wherein the following operations are performed for each network element cluster: Determining a mean value of the network element feature group of the network element cluster based on the network element feature group of the network element to be detected in the network element cluster and the number of network elements in the cluster; The network element feature group mean of the network element cluster is determined as the next round of network element cluster centroid of the network element cluster.
4. The method according to claim 1, wherein The network element deviation value of each network element cluster is determined based on the network element characteristic distance between the network element to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, wherein the following operations are performed for each network element cluster: Determine, based on the network element characteristic distance between the network element to be detected belonging to a network element cluster and the centroid of the last round of network element clusters in the cluster, the mean value of the intra-cluster distance, the intra-cluster mean deviation and the intra-cluster standard deviation of the network element cluster; The network element deviation value of the network element cluster is determined based on the intra-cluster distance mean, the intra-cluster deviation from the mean, and the intra-cluster standard deviation of the network element cluster.
5. The method according to any one of claims 1 to 4, characterized in that The method further comprises: issuing a stop instruction to a management network element for managing multiple network elements to be detected, instructing the management network element to prohibit the abnormal network element from performing model training; And, an isolation instruction is issued to the management network element, instructing the management network element to remove the network elements to be detected that are detected as abnormal network elements for multiple times.
6. An abnormal network element detection device, characterized in that: include: A feature extraction unit is configured to collect a network element feature group of each network element to be detected in a current detection cycle, wherein each network element to be detected includes an original network element detected for the first time and abnormal network elements in a historical detection cycle. Each network element feature group is a plurality of model adjustment parameters obtained when a corresponding local model is trained based on sample data of a network element to be detected; a feature detection unit, configured to iteratively cluster the network elements to be detected based on the network element feature groups of the network elements to be detected and the network element feature groups of the centroids of the network element clusters in each round, to obtain clustering results for each network element cluster; Based on the network element characteristic distance between the network element to be detected belonging to each network element cluster and the centroid of the last round of network element clusters in the cluster, the network element deviation value of each network element cluster is determined, and the network element to be detected belonging to the network element cluster whose network element deviation value exceeds the set deviation threshold is regarded as the abnormal network element in the current detection cycle.
7. The device according to claim 6, characterized in that The feature detection unit is used for: Cluster the network elements to be detected in a cyclic iterative manner until the iteration stop condition is met, and output the clustering results of the network element clusters: Determine, based on the network element feature group of each network element to be detected and the network element feature group of the current round of network element cluster centroids of each network element cluster, the network element feature distance between each network element to be detected and the current round of network element cluster centroids, and classify each network element to be detected into a network element cluster whose network element feature distance does not exceed a set distance threshold; Based on the network element feature groups of the network elements to be detected in each network element cluster, the centroid of each network element cluster is updated to obtain the next round of network element cluster centroids of each network element cluster.
8. The device according to claim 7, wherein The feature detection unit performs the following operations on each network element cluster: Determining a mean value of the network element feature group of the network element cluster based on the network element feature group of the network element to be detected in the network element cluster and the number of network elements in the cluster; The network element feature group mean of the network element cluster is determined as the next round of network element cluster centroid of the network element cluster.
9. The device according to claim 6, wherein The feature detection unit performs the following operations on each network element cluster: Determine, based on the network element characteristic distance between the network element to be detected belonging to a network element cluster and the centroid of the last round of network element clusters in the cluster, the mean value of the intra-cluster distance, the intra-cluster mean deviation and the intra-cluster standard deviation of the network element cluster; The network element deviation value of the network element cluster is determined based on the intra-cluster distance mean, the intra-cluster deviation from the mean, and the intra-cluster standard deviation of the network element cluster.
10. The device according to any one of claims 6 to 9, characterized in that The abnormal network element detection device further includes a poisoning detection unit, which is configured to: issuing a stop instruction to a management network element for managing multiple network elements to be detected, instructing the management network element to prohibit the abnormal network element from performing model training; And, an isolation instruction is issued to the management network element, instructing the management network element to remove the network elements to be detected that are detected as abnormal network elements for multiple times.
11. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory stores program codes, and when the program codes are executed by the processor, the processor executes the steps of the method according to any one of claims 1 to 5.
12. A computer-readable storage medium, characterized in that The method comprises a program code, and when the program code is run on a computer device, the program code is used to enable the computer device to execute the steps of the method according to any one of claims 1 to 5.