Rapid root cause positioning method for 5G power distribution network system fault

Abstract

The invention provides a rapid root cause positioning method for a 5G power distribution network system fault, and belongs to the field of power distribution network communication security, and the method comprises the steps: completing the real-time obtaining of a signaling call chain through the collection of a signaling data packet, the feature calculation of the signaling data packet, and the reconstruction of the signaling call chain, constructing a graph neural network for extracting the deep features in the signaling call chain, and carrying out the fault positioning of the 5G power distribution network system. And inputting the deep-level features into a classification model and a positioning model at the same time, and carrying out weighted combination to obtain a fault type and a root cause. According to the design of the invention, the method can rapidly analyze the abnormality of the 5G power distribution network system according to the signaling data of the 5G power distribution network and accurately recognize the root cause of the fault, and achieves the provision of the most possible fault root cause node and root cause problem for the operation and maintenance personnel of the 5G power distribution network system under the condition that the 5G power distribution network is abnormal or attacked. The method is advantaged in that operation and maintenance personnel can be helped to rapidly solve power distribution network faults, fault detection precision, response speed and robustness of a 5G power distribution network system are improved, efficient operation and maintenance of a 5G network are supported, and stable operation of the 5G network is guaranteed.

Description

Technical Field

[0001] The present invention belongs to the field of distribution network communication security, and specifically provides a method for quickly locating the root cause of a 5G distribution network system failure. Background Art

[0002] With the widespread application of 5G networks in distribution network systems, more and more distribution network equipment uses 5G networks as communication carriers for information exchange. 5G networks support multi-node and multi-path transmission, and the network structure is highly dynamic, resulting in complex and changeable signaling interactions and failure modes in the network, which poses huge challenges to network operation and maintenance. Traditional fault diagnosis methods mainly rely on the experience of network operation and maintenance personnel or rule-based monitoring methods. Although these methods can work effectively in relatively simple network environments, they have many limitations in large-scale and complex networks such as 5G distribution network systems. First, it is difficult for diagnostic methods based on artificial rules to fully cover a variety of fault types and complex fault scenarios, especially in the case of multi-node faults and cross-node propagation, which are difficult to identify and accurately locate. In addition, diagnostic methods that rely heavily on artificial experience are often powerless when faced with new or uncommon faults, resulting in low fault diagnosis efficiency and may even cause misjudgment, further affecting the operational stability of the distribution network.

[0003] In addition, in the 5G distribution network system, there are a large number of signaling interactions between various network elements (such as access and mobility management function AMF, user plane function UPF, etc.) to achieve the connection, session management and data transmission of distribution network terminal equipment. In the 5G distribution network system, various faults are prone to occur, affecting the normal operation of the distribution network system. Common fault types include distributed denial of service attacks (DDoS attacks), network element power failure, and sudden increase in business. These faults will not only affect some functional modules of the system, but also may cause fault propagation problems, affecting the service quality of the entire 5G distribution network. Therefore, in the fault location process, it is necessary not only to accurately identify the type of fault, but also to trace the specific location of the fault, that is, root cause location. In addition, solving the fault propagation problem only through the relationship between network element indicators is indirect and inaccurate. In contrast, signaling data, as the key data for controlling communication and collaboration between network elements, can directly describe the interactive relationship between network elements. Analyzing the signaling call chain can accurately depict the failure mode of network elements and effectively solve the problem of fault propagation between network elements. In the root cause location of faults in the 5G distribution network system, the signaling call chain has a complex topological structure and node dependency. Each network element not only has complex interactions with other network elements in multiple hops and multiple paths, but may also be affected by different external factors. Traditional feature extraction methods are difficult to fully characterize such complex interactions and dependencies, resulting in limited effectiveness of fault diagnosis. Therefore, a comprehensive method is needed to accurately identify and locate distribution network faults by analyzing signaling call chain data, thereby achieving rapid self-healing and rapid attack response of 5G distribution network systems. Summary of the invention

[0004] In view of the shortcomings of the existing 5G distribution network system fault root cause location method, the purpose of the present invention is to design a fast fault root cause location method for 5G distribution network access system to improve the accuracy and real-time performance of fault diagnosis.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0006] A method for quickly locating the root cause of a 5G distribution network system fault includes the following steps:

[0007] Step 1: Collect signaling data packets, calculate signaling data packet features, and reconstruct the signaling call chain;

[0008] Step 2: Build a graph neural network and extract deep features in the signaling call chain;

[0009] Step 3: Classification model and positioning model are combined to optimize model construction.

[0010] Furthermore, the signaling data packets are mirrored and copied using bypass sniffing technology in the core network switch;

[0011] Extract signaling data packets and header information to verify data length and integrity as well as time validity;

[0012] By shifting and parsing the signaling data packet through the start character of the signaling data packet, the signaling feature file structure is converted from binary to structured, and then the signaling call chain features are calculated and counted;

[0013] Through session backfill, key information between signaling data is aggregated and associated into a signaling call chain.

[0014] Furthermore, the graph neural network is built based on GraphVAE, which includes an encoder and a decoder;

[0015] The signaling call chain generates a hidden variable z through the encoder, and the hidden variable z is reconstructed into a signaling call chain through the decoder to restore the original call chain structure.

[0016] Furthermore, the latent variable z is simultaneously input into the localization model and the classification model. The localization model outputs a list of root cause faults based on the latent variable z, and the classification model outputs a list of fault types based on the latent variable z.

[0017] In order to integrate the calculation results of the positioning model and the classification model, a weighted loss function is introduced:

[0018] Loss = λLlo c +(1―λ)L cl (2)

[0019] Among them, λ∈[0,1] is the weight coefficient, and the output of the positioning model corresponds to the positioning loss L loc , and the output of the classification model corresponds to the classification loss L cl .

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] The present invention significantly improves the fault diagnosis accuracy and speed of the 5G distribution network system. Through the reconstruction of the signaling call chain and the feature extraction model based on GraphVAE, it realizes the capture of key features in complex signaling interactions, so as to quickly and accurately locate the root cause of the fault. At the same time, the combined optimization of fault classification and positioning tasks is used to reduce feature redundancy and ensure the high consistency of diagnostic output and resource utilization efficiency. After detecting a fault, the system can provide real-time feedback on the fault type and location information, helping operation and maintenance personnel to quickly respond to and solve the problem, greatly shortening the fault handling time, and ensuring the stable operation of the 5G distribution network system. In addition, the present invention is highly adaptable to the dynamic and complex 5G distribution network environment, can effectively identify a variety of fault types, reduce operation and maintenance costs, and provide reliable support for the large-scale deployment of 5G distribution network systems. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 This is an architecture diagram of a method for quickly locating the root cause of a 5G distribution network system fault according to an embodiment of the present invention;

[0023] Figure 2 This is a schematic diagram of a signaling call chain feature extraction model based on GraphVAE in an embodiment of the present invention;

[0024] Figure 3 Schematic diagram of the combined optimization model of classification and positioning tasks according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0026] Example

[0027] See also Figure 1-3 ,The 5G distribution network system fault rapid root cause location method includes the following steps:

[0028] 1) Signaling data packet collection, feature calculation and signaling call chain reconstruction.

[0029] Since the 5G distribution network system environment is dynamic and complex, it is impractical to record the various signaling call chains of the distribution network services in real time. In order to effectively locate the root cause of the 5G distribution network system fault by analyzing the signaling call chain, this embodiment uses bypass sniffing technology to collect the original signaling data packets, and designs a signaling call chain reconstruction method to achieve fast and accurate reconstruction of the signaling call process, so as to use the signaling call chain to achieve accurate fault location and classification.

[0030] Specifically, the following steps are included:

[0031] (1) By introducing bypass sniffing technology on the core network switch, all signaling data generated by the core network to complete the distribution network service request is mirrored in real time. Bypass sniffing only analyzes the copied data packets and will not cause any interference or delay to the original data packets, ensuring the normal operation of the 5G distribution network system.

[0032] (2) Cleaning and preprocessing the collected signaling data packets, including extracting the signaling data packets and their header information to verify the collection time range and corresponding version feature information of the signaling data, and ensure the length and integrity of the data as well as the validity of the time.

[0033] (3) Extract the start character of the signaling data packet, perform shift parsing on the signaling data packet according to the 3GPP specification and the signaling sample data packet file, and convert the original binary file into a structured signaling feature file. By analyzing the signaling feature information, calculate and count the relevant signaling call chain features, such as communication interaction delay.

[0034] (4) Signaling call chain data association: Through session backfill, key information aggregation between signaling data is achieved, such as the International Mobile Subscriber Identity (IMSI).

[0035] IMSI is a unique identification code for mobile users, which can uniquely identify mobile users worldwide. Therefore, by grouping signaling data by IMSI number, the signaling message set belonging to a certain user can be uniquely determined. Given that there is a one-to-one correspondence between IMSI and user equipment identification (AMF_UE_NGAP_ID), if the AMF_UE_NGAP_ID field is missing, IMSI can be used for re-association to complete the missing field, thereby facilitating subsequent data backfilling and processing operations. The signaling data is associated into a signaling call chain through the three key fields of IMSI, AMF_UE_NGAP_ID and timestamp. If the IMSI is the same, the timestamp is used to determine the signaling set that belongs to the same user to complete a service request.

[0036] 2) Construct graph neural network and deep feature extraction in signaling call chain.

[0037] Specifically, the signaling call chain is defined as a directed graph of network elements that the signaling data passes through to complete a certain signaling service request of a certain user, and the symbol is G = (A, E, F). A represents the node of the call chain, E represents the edge of the call chain, and F represents the feature of the call chain. Each signaling call chain can be converted into an A, F, E matrix representation.

[0038] In order to preserve the relationship of information transmission between network elements and enhance the feature expression capability, this embodiment constructs a signaling call chain feature extraction neural network based on the GraphVAE model to extract data information and spatial location information rich in signaling call chain data. GraphVAE can effectively fuse the timing information and structural information in the signaling call chain into a high-dimensional feature expression, capture complex patterns across nodes, enable the model to identify fault information across nodes, and provide richer contextual information for subsequent root cause location and fault classification, thereby improving the accuracy and robustness of fault diagnosis.

[0039] GraphVAE consists of two main components: an encoder and a decoder. The encoder converts the input signaling call chain into a vector representation in the latent space, and the decoder restores the vector representation to the original signaling call chain.

[0040] The forward process of the feature extraction model is as follows: First, the signaling call chain is encoded through a graph structure encoder. In the encoder part, the graph structure and attributes of the signaling call chain (such as node feature A, edge feature E, call chain feature F) are input to generate a latent variable z, which represents the deep data and spatial location features of the call chain. Then, the latent variable z is reconstructed into a signaling call chain through a decoder to restore the original call chain structure, thereby retaining its deep features. The encoding and decoding process contains optimization constraints on the KL divergence to ensure that the distribution of the latent variable z is as close to the prior distribution p(z) as possible. Ultimately, this process not only extracts the key features of the signaling call chain, but also enhances the modeling capability of fault modes through latent variable representation.

[0041] There are many signaling call chains on the core network of the 5G distribution network. It is very expensive and unrealistic to directly label each signaling call chain and then train the model. The call chain reconstruction task is a self-supervised learning solution that can train an excellent feature extraction model without labeling the signaling call chain, and then obtain an excellent feature representation z. In the future, only a small amount of labeled data is needed to train the comprehensive application task to achieve network element fault classification and location.

[0042] During the training process, by analyzing and collecting signaling call chain data, GraphVAE minimizes the reconstruction loss and KL divergence to learn the vector representation of signaling call chain features. The encoder and decoder models learn interpretable and reproducible latent space representations to solve the encoder model with the lowest model loss. and decoder model θ. The encoder converts the call chain information G = (A, E, F) into a vector z that approximately conforms to the normal distribution, and the decoder then converts z into G ’ =(A,E,F). The encoder is composed of variational posterior Definition, the decoder is defined by the generative distribution p θ (G'|z), where and θ are learning parameters. In addition, a prior distribution p(z) is imposed on the latent representation as a regularizer; a simple isotropic Gaussian prior p(z)=N(0,1) is used. The overall To train the model. The overall Loss function contains the similarity between the vector z distribution and the normal distribution and the generated G ’ The similarity between the two parts and the original image G is expressed as follows:

[0043]

[0044] 3) Construction of a combined optimization model of classification model and positioning model.

[0045] Construct a 5G distribution network fault classification and location task model. The model consists of a deep network feature extraction module, a location model, and a classification model. The input signal is used to call the chain G from the restored original signaling through the feature extraction module. ’ Rich feature representations are extracted from the latent variables z to generate deep latent variables z, which are used to characterize the complex characteristics of the fault. The latent variable feature z is then input into the localization model and the classification model at the same time. The localization model outputs a list of root cause faults based on the latent variable z to indicate the potential root causes of the faults; the classification model outputs a list of fault types based on the same latent variable z to identify the fault category.

[0046] The classification model focuses on the discriminative information of the fault type, while the localization model focuses on the specific root cause of the fault. Combined optimization allows the two to share deep graph features, enhance information richness and data complementarity, and improve the overall effect of fault diagnosis. At the same time, the coordinated optimization of classification and localization can ensure the consistency of the output. For example, the type information provided by the classification model can help the localization model focus more accurately on the fault characteristics of a specific type, and vice versa. In addition, combined optimization reduces the feature extraction redundancy caused by separate training, improves the efficiency and resource utilization of the model, and thus achieves more accurate and rapid fault diagnosis.

[0047] During the training process, the model introduces trade-off loss to balance the loss of the positioning task and the classification task, thus achieving the combined optimization of the model's multiple tasks. Specifically, the output of the positioning model corresponds to the positioning loss L loc , and the output of the classification model corresponds to the classification loss L cl In order to comprehensively measure the performance of the model on the two tasks, a weighted loss function is introduced to combine the positioning loss and the classification loss:

[0048] Loss = λL loc +(1―λ)L cl (2)

[0049] Among them, λ∈[0,1] is the weight coefficient, which is used to adjust the relative importance of classification and positioning tasks. By adjusting the value of λ, the attention of the two tasks can be balanced during the model training process. If the classification task is more important, increase the λ value, otherwise decrease the λ value. The combined optimization strategy using trade-off loss can better balance the performance of classification and positioning tasks in model training, so that the model can achieve ideal results in both tasks, thereby improving the accuracy and robustness of overall fault diagnosis. This optimization method makes full use of the complementarity of classification and positioning tasks, laying the foundation for more accurate and rapid fault diagnosis.

[0050] Finally, it should be noted that the above embodiments are only preferred embodiments of the present invention, which are only used to illustrate the technical solution of the present invention rather than to limit it. Any modification, improvement, other modification or equivalent replacement made to the technical solution of the present invention by ordinary technicians in this field, as long as they do not depart from the spirit and scope of the technical solution of the present invention, should be included in the protection scope of the present invention.

Claims

1. A method for quickly locating the root cause of a 5G distribution network system fault, characterized in that: The steps include: Step 1: Collect signaling data packets, calculate signaling data packet features, and reconstruct the signaling call chain; Step 2: Build a graph neural network and extract deep features in the signaling call chain; Step 3: Classification model and positioning model are combined to optimize model construction.

2. The method for quickly locating the root cause of a 5G distribution network system fault according to claim 1 is characterized in that: The signaling data packets are mirrored and copied using bypass sniffing technology in the core network switch; Extract signaling data packets and header information to verify data length and integrity as well as time validity; By shifting and parsing the signaling data packet through the start character of the signaling data packet, the signaling feature file structure is converted from binary to structured, and then the signaling call chain features are calculated and counted; Through session backfill, key information between signaling data is aggregated and associated into a signaling call chain.

3. The method for quickly locating the root cause of a 5G distribution network system fault according to claim 2 is characterized in that: The graph neural network is built on GraphVAE, which includes an encoder and a decoder; The signaling call chain generates a hidden variable z through the encoder, and the hidden variable z is reconstructed into a signaling call chain through the decoder to restore the original call chain structure.

4. The method for quickly locating the root cause of a 5G distribution network system fault according to claim 3 is characterized in that: The latent variable z is input into the localization model and the classification model at the same time. The localization model outputs a list of root cause faults based on the latent variable z, and the classification model outputs a list of fault types based on the latent variable z. In order to integrate the calculation results of the positioning model and the classification model, a weighted loss function is introduced: Loss=λL loc +(1―λ)L cl Among them, λ∈[0,1] is the weight coefficient, and the output of the positioning model corresponds to the positioning loss L loc , and the output of the classification model corresponds to the classification loss L cl .

Images